phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-08-03 17:29:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'main' into feat/litellm_sambanova_usage

Browse source
This commit is contained in:
Jorge Piedrahita Ortiz 2025-03-18 12:06:58 -05:00 • committed by GitHub
commit 5bd1bd30e2
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
76 changed files with 3534 additions and 2843 deletions
Download patch file Download diff file Expand all files Collapse all files

9
.cursor/rules/general.mdc
View file

@ -1,9 +0,0 @@
---
description: General rules always applicable across the project
globs:
alwaysApply: true
---
# Style
- Comments must add value to code. Don't write filler comments explaining what you are doing next; they just add noise.
- Add a comment to clarify surprising behavior which would not be obvious. Good variable naming and clear code organization is more important.

17
.github/dependabot.yml vendored
View file

@ -5,4 +5,19 @@ updates:
- package-ecosystem: "github-actions"
directory: "/" # Will use the default workflow location of `.github/workflows`
schedule:
interval: "daily"
interval: "weekly"
day: "saturday"
commit-message:
prefix: chore(github-deps)
- package-ecosystem: "uv"
directory: "/"
schedule:
interval: "weekly"
day: "saturday"
# ignore all non-security updates: https://docs.github.com/en/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file#open-pull-requests-limit
open-pull-requests-limit: 0
labels:
- type/dependencies
- python
commit-message:
prefix: chore(python-deps)

34
.github/workflows/integration-tests.yml vendored
View file

@ -1,13 +1,28 @@
name: Integration tests
on:
pull_request:
push:
branches: [main]
branches: [ main ]
pull_request:
branches: [ main ]
paths:
- 'distributions/**'
- 'llama_stack/**'
- 'tests/integration/**'
- 'uv.lock'
- 'pyproject.toml'
- 'requirements.txt'
- '.github/workflows/integration-tests.yml' # This workflow
jobs:
ollama:
test-matrix:
runs-on: ubuntu-latest
strategy:
matrix:
# Listing tests manually since some of them currently fail
# TODO: generate matrix list from tests/integration when fixed
test-type: [inference, datasets, inspect, scoring, post_training, providers]
fail-fast: false # we want to run all tests regardless of failure
steps:
- name: Checkout repository
@ -34,6 +49,8 @@ jobs:
run: |
uv sync --extra dev --extra test
uv pip install ollama faiss-cpu
# always test against the latest version of the client
uv pip install git+https://github.com/meta-llama/llama-stack-client-python.git@main
uv pip install -e .
- name: Wait for Ollama to start
@ -56,25 +73,24 @@ jobs:
INFERENCE_MODEL: "meta-llama/Llama-3.2-3B-Instruct"
run: |
source .venv/bin/activate
# TODO: use "llama stack run"
nohup uv run python -m llama_stack.distribution.server.server --yaml-config ./llama_stack/templates/ollama/run.yaml > server.log 2>&1 &
nohup uv run llama stack run ./llama_stack/templates/ollama/run.yaml --image-type venv > server.log 2>&1 &
- name: Wait for Llama Stack server to be ready
run: |
echo "Waiting for Llama Stack server..."
for i in {1..30}; do
if curl -s http://localhost:8321/v1/health | grep -q "OK"; then
echo " Llama Stack server is up!"
echo "Llama Stack server is up!"
exit 0
fi
sleep 1
done
echo " Llama Stack server failed to start"
echo "Llama Stack server failed to start"
cat server.log
exit 1
- name: Run Inference Integration Tests
- name: Run Integration Tests
env:
INFERENCE_MODEL: "meta-llama/Llama-3.2-3B-Instruct"
run: |
uv run pytest -v tests/integration/inference --stack-config=ollama --text-model="meta-llama/Llama-3.2-3B-Instruct" --embedding-model=all-MiniLM-L6-v2
uv run pytest -v tests/integration/${{ matrix.test-type }} --stack-config=ollama --text-model="meta-llama/Llama-3.2-3B-Instruct" --embedding-model=all-MiniLM-L6-v2

5
.github/workflows/providers-build.yml vendored
View file

@ -40,6 +40,7 @@ jobs:
matrix:
template: ${{ fromJson(needs.generate-matrix.outputs.templates) }}
image-type: [venv, container]
fail-fast: false # We want to run all jobs even if some fail
steps:
- name: Checkout repository
@ -67,7 +68,9 @@ jobs:
- name: Run Llama Stack Build
run: |
uv run llama stack build --template ${{ matrix.template }} --image-type ${{ matrix.image-type }} --image-name test
# USE_COPY_NOT_MOUNT is set to true since mounting is not supported by docker buildx, we use COPY instead
# LLAMA_STACK_DIR is set to the current directory so we are building from the source
USE_COPY_NOT_MOUNT=true LLAMA_STACK_DIR=. uv run llama stack build --template ${{ matrix.template }} --image-type ${{ matrix.image-type }} --image-name test
- name: Print dependencies in the image
if: matrix.image-type == 'venv'

8
.github/workflows/unit-tests.yml vendored
View file

@ -5,6 +5,14 @@ on:
branches: [ main ]
pull_request:
branches: [ main ]
paths:
- 'distributions/**'
- 'llama_stack/**'
- 'tests/unit/**'
- 'uv.lock'
- 'pyproject.toml'
- 'requirements.txt'
- '.github/workflows/unit-tests.yml' # This workflow
workflow_dispatch:
jobs:

2
.pre-commit-config.yaml
View file

@ -77,7 +77,7 @@ repos:
name: Distribution Template Codegen
additional_dependencies:
- uv==0.6.0
entry: uv run --extra codegen python -m llama_stack.scripts.distro_codegen
entry: uv run --extra codegen ./scripts/distro_codegen.py
language: python
pass_filenames: false
require_serial: true

2
CONTRIBUTING.md
View file

@ -159,7 +159,7 @@ LLAMA_STACK_DIR=$(pwd) LLAMA_STACK_CLIENT_DIR=../llama-stack-client-python llama
### Updating Provider Configurations
If you have made changes to a provider's configuration in any form (introducing a new config key, or changing models, etc.), you should run `python llama_stack/scripts/distro_codegen.py` to re-generate various YAML files as well as the documentation. You should not change `docs/source/.../distributions/` files manually as they are auto-generated.
If you have made changes to a provider's configuration in any form (introducing a new config key, or changing models, etc.), you should run `./scripts/distro_codegen.py` to re-generate various YAML files as well as the documentation. You should not change `docs/source/.../distributions/` files manually as they are auto-generated.
### Building the Documentation

3
distributions/dependencies.json
View file

@ -401,16 +401,13 @@
],
"nvidia": [
"aiosqlite",
"autoevals",
"blobfile",
"chardet",
"datasets",
"faiss-cpu",
"fastapi",
"fire",
"httpx",
"matplotlib",
"mcp",
"nltk",
"numpy",
"openai",

715
docs/_static/llama-stack-spec.html vendored
View file

@ -40,75 +40,7 @@
}
],
"paths": {
"/v1/datasetio/rows": {
"get": {
"responses": {
"200": {
"description": "OK",
"content": {
"application/json": {
"schema": {
"$ref": "#/components/schemas/PaginatedRowsResult"
}
}
}
},
"400": {
"$ref": "#/components/responses/BadRequest400"
},
"429": {
"$ref": "#/components/responses/TooManyRequests429"
},
"500": {
"$ref": "#/components/responses/InternalServerError500"
},
"default": {
"$ref": "#/components/responses/DefaultError"
}
},
"tags": [
"DatasetIO"
],
"description": "Get a paginated list of rows from a dataset.",
"parameters": [
{
"name": "dataset_id",
"in": "query",
"description": "The ID of the dataset to get the rows from.",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "rows_in_page",
"in": "query",
"description": "The number of rows to get per page.",
"required": true,
"schema": {
"type": "integer"
}
},
{
"name": "page_token",
"in": "query",
"description": "The token to get the next page of rows.",
"required": false,
"schema": {
"type": "string"
}
},
{
"name": "filter_condition",
"in": "query",
"description": "(Optional) A condition to filter the rows by.",
"required": false,
"schema": {
"type": "string"
}
}
]
},
"/v1/datasetio/append-rows/{dataset_id}": {
"post": {
"responses": {
"200": {
@ -131,7 +63,16 @@
"DatasetIO"
],
"description": "",
"parameters": [],
"parameters": [
{
"name": "dataset_id",
"in": "path",
"required": true,
"schema": {
"type": "string"
}
}
],
"requestBody": {
"content": {
"application/json": {
@ -583,7 +524,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "List all buckets.",
"parameters": [
@ -623,7 +564,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "Create a new upload session for a file identified by a bucket and key.",
"parameters": [],
@ -850,7 +791,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "Get a file info identified by a bucket and key.",
"parameters": [
@ -900,7 +841,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "Delete a file identified by a bucket and key.",
"parameters": [
@ -1889,7 +1830,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "Returns information about an existsing upload session",
"parameters": [
@ -1937,7 +1878,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "Upload file content to an existing upload session. On the server, request body will have the raw bytes that are uploaded.",
"parameters": [
@ -2236,6 +2177,67 @@
}
}
},
"/v1/datasetio/iterrows/{dataset_id}": {
"get": {
"responses": {
"200": {
"description": "OK",
"content": {
"application/json": {
"schema": {
"$ref": "#/components/schemas/IterrowsResponse"
}
}
}
},
"400": {
"$ref": "#/components/responses/BadRequest400"
},
"429": {
"$ref": "#/components/responses/TooManyRequests429"
},
"500": {
"$ref": "#/components/responses/InternalServerError500"
},
"default": {
"$ref": "#/components/responses/DefaultError"
}
},
"tags": [
"DatasetIO"
],
"description": "Get a paginated list of rows from a dataset. Uses cursor-based pagination.",
"parameters": [
{
"name": "dataset_id",
"in": "path",
"description": "The ID of the dataset to get the rows from.",
"required": true,
"schema": {
"type": "string"
}
},
{
"name": "start_index",
"in": "query",
"description": "Index into dataset for the first row to get. Get all rows if None.",
"required": false,
"schema": {
"type": "integer"
}
},
{
"name": "limit",
"in": "query",
"description": "The number of rows to get.",
"required": false,
"schema": {
"type": "integer"
}
}
]
}
},
"/v1/eval/benchmarks/{benchmark_id}/jobs/{job_id}": {
"get": {
"responses": {
@ -2535,7 +2537,14 @@
"post": {
"responses": {
"200": {
"description": "OK"
"description": "OK",
"content": {
"application/json": {
"schema": {
"$ref": "#/components/schemas/Dataset"
}
}
}
},
"400": {
"$ref": "#/components/responses/BadRequest400"
@ -2553,7 +2562,7 @@
"tags": [
"Datasets"
],
"description": "",
"description": "Register a new dataset.",
"parameters": [],
"requestBody": {
"content": {
@ -2594,7 +2603,7 @@
}
},
"tags": [
"Files (Coming Soon)"
"Files"
],
"description": "List all files in a bucket.",
"parameters": [
@ -3824,9 +3833,6 @@
"AppendRowsRequest": {
"type": "object",
"properties": {
"dataset_id": {
"type": "string"
},
"rows": {
"type": "array",
"items": {
@ -3858,7 +3864,6 @@
},
"additionalProperties": false,
"required": [
"dataset_id",
"rows"
],
"title": "AppendRowsRequest"
@ -6824,6 +6829,224 @@
],
"title": "Benchmark"
},
"DataSource": {
"oneOf": [
{
"$ref": "#/components/schemas/URIDataSource"
},
{
"$ref": "#/components/schemas/RowsDataSource"
}
],
"discriminator": {
"propertyName": "type",
"mapping": {
"uri": "#/components/schemas/URIDataSource",
"rows": "#/components/schemas/RowsDataSource"
}
}
},
"Dataset": {
"type": "object",
"properties": {
"identifier": {
"type": "string"
},
"provider_resource_id": {
"type": "string"
},
"provider_id": {
"type": "string"
},
"type": {
"type": "string",
"const": "dataset",
"default": "dataset"
},
"purpose": {
"type": "string",
"enum": [
"post-training/messages",
"eval/question-answer",
"eval/messages-answer"
],
"title": "DatasetPurpose",
"description": "Purpose of the dataset. Each purpose has a required input data schema."
},
"source": {
"$ref": "#/components/schemas/DataSource"
},
"metadata": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
}
},
"additionalProperties": false,
"required": [
"identifier",
"provider_resource_id",
"provider_id",
"type",
"purpose",
"source",
"metadata"
],
"title": "Dataset"
},
"RowsDataSource": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "rows",
"default": "rows"
},
"rows": {
"type": "array",
"items": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
},
"description": "The dataset is stored in rows. E.g. - [ {\"messages\": [{\"role\": \"user\", \"content\": \"Hello, world!\"}, {\"role\": \"assistant\", \"content\": \"Hello, world!\"}]} ]"
}
},
"additionalProperties": false,
"required": [
"type",
"rows"
],
"title": "RowsDataSource",
"description": "A dataset stored in rows."
},
"URIDataSource": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "uri",
"default": "uri"
},
"uri": {
"type": "string",
"description": "The dataset can be obtained from a URI. E.g. - \"https://mywebsite.com/mydata.jsonl\" - \"lsfs://mydata.jsonl\" - \"data:csv;base64,{base64_content}\""
}
},
"additionalProperties": false,
"required": [
"type",
"uri"
],
"title": "URIDataSource",
"description": "A dataset that can be obtained from a URI."
},
"Model": {
"type": "object",
"properties": {
"identifier": {
"type": "string"
},
"provider_resource_id": {
"type": "string"
},
"provider_id": {
"type": "string"
},
"type": {
"type": "string",
"const": "model",
"default": "model"
},
"metadata": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
},
"model_type": {
"$ref": "#/components/schemas/ModelType",
"default": "llm"
}
},
"additionalProperties": false,
"required": [
"identifier",
"provider_resource_id",
"provider_id",
"type",
"metadata",
"model_type"
],
"title": "Model"
},
"ModelType": {
"type": "string",
"enum": [
"llm",
"embedding"
],
"title": "ModelType"
},
"AgentTurnInputType": {
"type": "object",
"properties": {
@ -6899,70 +7122,6 @@
],
"title": "CompletionInputType"
},
"Dataset": {
"type": "object",
"properties": {
"identifier": {
"type": "string"
},
"provider_resource_id": {
"type": "string"
},
"provider_id": {
"type": "string"
},
"type": {
"type": "string",
"const": "dataset",
"default": "dataset"
},
"dataset_schema": {
"type": "object",
"additionalProperties": {
"$ref": "#/components/schemas/ParamType"
}
},
"url": {
"$ref": "#/components/schemas/URL"
},
"metadata": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
}
},
"additionalProperties": false,
"required": [
"identifier",
"provider_resource_id",
"provider_id",
"type",
"dataset_schema",
"url",
"metadata"
],
"title": "Dataset"
},
"JsonType": {
"type": "object",
"properties": {
@ -7057,151 +7216,6 @@
}
}
},
"StringType": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "string",
"default": "string"
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "StringType"
},
"UnionType": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "union",
"default": "union"
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "UnionType"
},
"Model": {
"type": "object",
"properties": {
"identifier": {
"type": "string"
},
"provider_resource_id": {
"type": "string"
},
"provider_id": {
"type": "string"
},
"type": {
"type": "string",
"const": "model",
"default": "model"
},
"metadata": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
},
"model_type": {
"$ref": "#/components/schemas/ModelType",
"default": "llm"
}
},
"additionalProperties": false,
"required": [
"identifier",
"provider_resource_id",
"provider_id",
"type",
"metadata",
"model_type"
],
"title": "Model"
},
"ModelType": {
"type": "string",
"enum": [
"llm",
"embedding"
],
"title": "ModelType"
},
"PaginatedRowsResult": {
"type": "object",
"properties": {
"rows": {
"type": "array",
"items": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
},
"description": "The rows in the current page."
},
"total_count": {
"type": "integer",
"description": "The total number of rows in the dataset."
},
"next_page_token": {
"type": "string",
"description": "The token to get the next page of rows."
}
},
"additionalProperties": false,
"required": [
"rows",
"total_count"
],
"title": "PaginatedRowsResult",
"description": "A paginated list of rows from a dataset."
},
"ScoringFn": {
"type": "object",
"properties": {
@ -7265,6 +7279,36 @@
],
"title": "ScoringFn"
},
"StringType": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "string",
"default": "string"
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "StringType"
},
"UnionType": {
"type": "object",
"properties": {
"type": {
"type": "string",
"const": "union",
"default": "union"
}
},
"additionalProperties": false,
"required": [
"type"
],
"title": "UnionType"
},
"Shield": {
"type": "object",
"properties": {
@ -8084,6 +8128,50 @@
],
"title": "ToolInvocationResult"
},
"IterrowsResponse": {
"type": "object",
"properties": {
"data": {
"type": "array",
"items": {
"type": "object",
"additionalProperties": {
"oneOf": [
{
"type": "null"
},
{
"type": "boolean"
},
{
"type": "number"
},
{
"type": "string"
},
{
"type": "array"
},
{
"type": "object"
}
]
}
},
"description": "The rows in the current page."
},
"next_start_index": {
"type": "integer",
"description": "Index into dataset for the first row in the next page. None if there are no more rows."
}
},
"additionalProperties": false,
"required": [
"data"
],
"title": "IterrowsResponse",
"description": "A paginated list of rows from a dataset."
},
"ListAgentSessionsResponse": {
"type": "object",
"properties": {
@ -9330,23 +9418,18 @@
"RegisterDatasetRequest": {
"type": "object",
"properties": {
"dataset_id": {
"type": "string"
"purpose": {
"type": "string",
"enum": [
"post-training/messages",
"eval/question-answer",
"eval/messages-answer"
],
"description": "The purpose of the dataset. One of - \"post-training/messages\": The dataset contains a messages column with list of messages for post-training. { \"messages\": [ {\"role\": \"user\", \"content\": \"Hello, world!\"}, {\"role\": \"assistant\", \"content\": \"Hello, world!\"}, ] } - \"eval/question-answer\": The dataset contains a question column and an answer column for evaluation. { \"question\": \"What is the capital of France?\", \"answer\": \"Paris\" } - \"eval/messages-answer\": The dataset contains a messages column with list of messages and an answer column for evaluation. { \"messages\": [ {\"role\": \"user\", \"content\": \"Hello, my name is John Doe.\"}, {\"role\": \"assistant\", \"content\": \"Hello, John Doe. How can I help you today?\"}, {\"role\": \"user\", \"content\": \"What's my name?\"}, ], \"answer\": \"John Doe\" }"
},
"dataset_schema": {
"type": "object",
"additionalProperties": {
"$ref": "#/components/schemas/ParamType"
}
},
"url": {
"$ref": "#/components/schemas/URL"
},
"provider_dataset_id": {
"type": "string"
},
"provider_id": {
"type": "string"
"source": {
"$ref": "#/components/schemas/DataSource",
"description": "The data source of the dataset. Ensure that the data source schema is compatible with the purpose of the dataset. Examples: - { \"type\": \"uri\", \"uri\": \"https://mywebsite.com/mydata.jsonl\" } - { \"type\": \"uri\", \"uri\": \"lsfs://mydata.jsonl\" } - { \"type\": \"uri\", \"uri\": \"data:csv;base64,{base64_content}\" } - { \"type\": \"uri\", \"uri\": \"huggingface://llamastack/simpleqa?split=train\" } - { \"type\": \"rows\", \"rows\": [ { \"messages\": [ {\"role\": \"user\", \"content\": \"Hello, world!\"}, {\"role\": \"assistant\", \"content\": \"Hello, world!\"}, ] } ] }"
},
"metadata": {
"type": "object",
@ -9371,14 +9454,18 @@
"type": "object"
}
]
}
},
"description": "The metadata for the dataset. - E.g. {\"description\": \"My dataset\"}"
},
"dataset_id": {
"type": "string",
"description": "The ID of the dataset. If not provided, an ID will be generated."
}
},
"additionalProperties": false,
"required": [
"dataset_id",
"dataset_schema",
"url"
"purpose",
"source"
],
"title": "RegisterDatasetRequest"
},
@ -10197,7 +10284,7 @@
"x-displayName": "Llama Stack Evaluation API for running evaluations on model and agent candidates."
},
{
"name": "Files (Coming Soon)"
"name": "Files"
},
{
"name": "Inference",
@ -10258,7 +10345,7 @@
"DatasetIO",
"Datasets",
"Eval",
"Files (Coming Soon)",
"Files",
"Inference",
"Inspect",
"Models",

499
docs/_static/llama-stack-spec.yaml vendored
View file

@ -10,56 +10,7 @@ info:
servers:
- url: http://any-hosted-llama-stack.com
paths:
/v1/datasetio/rows:
get:
responses:
'200':
description: OK
content:
application/json:
schema:
$ref: '#/components/schemas/PaginatedRowsResult'
'400':
$ref: '#/components/responses/BadRequest400'
'429':
$ref: >-
#/components/responses/TooManyRequests429
'500':
$ref: >-
#/components/responses/InternalServerError500
default:
$ref: '#/components/responses/DefaultError'
tags:
- DatasetIO
description: >-
Get a paginated list of rows from a dataset.
parameters:
- name: dataset_id
in: query
description: >-
The ID of the dataset to get the rows from.
required: true
schema:
type: string
- name: rows_in_page
in: query
description: The number of rows to get per page.
required: true
schema:
type: integer
- name: page_token
in: query
description: The token to get the next page of rows.
required: false
schema:
type: string
- name: filter_condition
in: query
description: >-
(Optional) A condition to filter the rows by.
required: false
schema:
type: string
/v1/datasetio/append-rows/{dataset_id}:
post:
responses:
'200':
@ -77,7 +28,12 @@ paths:
tags:
- DatasetIO
description: ''
parameters: []
parameters:
- name: dataset_id
in: path
required: true
schema:
type: string
requestBody:
content:
application/json:
@ -394,7 +350,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: List all buckets.
parameters:
- name: bucket
@ -421,7 +377,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: >-
Create a new upload session for a file identified by a bucket and key.
parameters: []
@ -580,7 +536,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: >-
Get a file info identified by a bucket and key.
parameters:
@ -616,7 +572,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: >-
Delete a file identified by a bucket and key.
parameters:
@ -1268,7 +1224,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: >-
Returns information about an existsing upload session
parameters:
@ -1299,7 +1255,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: >-
Upload file content to an existing upload session. On the server, request
body will have the raw bytes that are uploaded.
@ -1501,6 +1457,50 @@ paths:
schema:
$ref: '#/components/schemas/InvokeToolRequest'
required: true
/v1/datasetio/iterrows/{dataset_id}:
get:
responses:
'200':
description: OK
content:
application/json:
schema:
$ref: '#/components/schemas/IterrowsResponse'
'400':
$ref: '#/components/responses/BadRequest400'
'429':
$ref: >-
#/components/responses/TooManyRequests429
'500':
$ref: >-
#/components/responses/InternalServerError500
default:
$ref: '#/components/responses/DefaultError'
tags:
- DatasetIO
description: >-
Get a paginated list of rows from a dataset. Uses cursor-based pagination.
parameters:
- name: dataset_id
in: path
description: >-
The ID of the dataset to get the rows from.
required: true
schema:
type: string
- name: start_index
in: query
description: >-
Index into dataset for the first row to get. Get all rows if None.
required: false
schema:
type: integer
- name: limit
in: query
description: The number of rows to get.
required: false
schema:
type: integer
/v1/eval/benchmarks/{benchmark_id}/jobs/{job_id}:
get:
responses:
@ -1710,6 +1710,10 @@ paths:
responses:
'200':
description: OK
content:
application/json:
schema:
$ref: '#/components/schemas/Dataset'
'400':
$ref: '#/components/responses/BadRequest400'
'429':
@ -1722,7 +1726,7 @@ paths:
$ref: '#/components/responses/DefaultError'
tags:
- Datasets
description: ''
description: Register a new dataset.
parameters: []
requestBody:
content:
@ -1750,7 +1754,7 @@ paths:
default:
$ref: '#/components/responses/DefaultError'
tags:
- Files (Coming Soon)
- Files
description: List all files in a bucket.
parameters:
- name: bucket
@ -2607,8 +2611,6 @@ components:
AppendRowsRequest:
type: object
properties:
dataset_id:
type: string
rows:
type: array
items:
@ -2623,7 +2625,6 @@ components:
- type: object
additionalProperties: false
required:
- dataset_id
- rows
title: AppendRowsRequest
CompletionMessage:
@ -4726,6 +4727,148 @@ components:
- scoring_functions
- metadata
title: Benchmark
DataSource:
oneOf:
- $ref: '#/components/schemas/URIDataSource'
- $ref: '#/components/schemas/RowsDataSource'
discriminator:
propertyName: type
mapping:
uri: '#/components/schemas/URIDataSource'
rows: '#/components/schemas/RowsDataSource'
Dataset:
type: object
properties:
identifier:
type: string
provider_resource_id:
type: string
provider_id:
type: string
type:
type: string
const: dataset
default: dataset
purpose:
type: string
enum:
- post-training/messages
- eval/question-answer
- eval/messages-answer
title: DatasetPurpose
description: >-
Purpose of the dataset. Each purpose has a required input data schema.
source:
$ref: '#/components/schemas/DataSource'
metadata:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
additionalProperties: false
required:
- identifier
- provider_resource_id
- provider_id
- type
- purpose
- source
- metadata
title: Dataset
RowsDataSource:
type: object
properties:
type:
type: string
const: rows
default: rows
rows:
type: array
items:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: >-
The dataset is stored in rows. E.g. - [ {"messages": [{"role": "user",
"content": "Hello, world!"}, {"role": "assistant", "content": "Hello,
world!"}]} ]
additionalProperties: false
required:
- type
- rows
title: RowsDataSource
description: A dataset stored in rows.
URIDataSource:
type: object
properties:
type:
type: string
const: uri
default: uri
uri:
type: string
description: >-
The dataset can be obtained from a URI. E.g. - "https://mywebsite.com/mydata.jsonl"
- "lsfs://mydata.jsonl" - "data:csv;base64,{base64_content}"
additionalProperties: false
required:
- type
- uri
title: URIDataSource
description: >-
A dataset that can be obtained from a URI.
Model:
type: object
properties:
identifier:
type: string
provider_resource_id:
type: string
provider_id:
type: string
type:
type: string
const: model
default: model
metadata:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
model_type:
$ref: '#/components/schemas/ModelType'
default: llm
additionalProperties: false
required:
- identifier
- provider_resource_id
- provider_id
- type
- metadata
- model_type
title: Model
ModelType:
type: string
enum:
- llm
- embedding
title: ModelType
AgentTurnInputType:
type: object
properties:
@ -4781,45 +4924,6 @@ components:
required:
- type
title: CompletionInputType
Dataset:
type: object
properties:
identifier:
type: string
provider_resource_id:
type: string
provider_id:
type: string
type:
type: string
const: dataset
default: dataset
dataset_schema:
type: object
additionalProperties:
$ref: '#/components/schemas/ParamType'
url:
$ref: '#/components/schemas/URL'
metadata:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
additionalProperties: false
required:
- identifier
- provider_resource_id
- provider_id
- type
- dataset_schema
- url
- metadata
title: Dataset
JsonType:
type: object
properties:
@ -4878,97 +4982,6 @@ components:
chat_completion_input: '#/components/schemas/ChatCompletionInputType'
completion_input: '#/components/schemas/CompletionInputType'
agent_turn_input: '#/components/schemas/AgentTurnInputType'
StringType:
type: object
properties:
type:
type: string
const: string
default: string
additionalProperties: false
required:
- type
title: StringType
UnionType:
type: object
properties:
type:
type: string
const: union
default: union
additionalProperties: false
required:
- type
title: UnionType
Model:
type: object
properties:
identifier:
type: string
provider_resource_id:
type: string
provider_id:
type: string
type:
type: string
const: model
default: model
metadata:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
model_type:
$ref: '#/components/schemas/ModelType'
default: llm
additionalProperties: false
required:
- identifier
- provider_resource_id
- provider_id
- type
- metadata
- model_type
title: Model
ModelType:
type: string
enum:
- llm
- embedding
title: ModelType
PaginatedRowsResult:
type: object
properties:
rows:
type: array
items:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: The rows in the current page.
total_count:
type: integer
description: The total number of rows in the dataset.
next_page_token:
type: string
description: The token to get the next page of rows.
additionalProperties: false
required:
- rows
- total_count
title: PaginatedRowsResult
description: A paginated list of rows from a dataset.
ScoringFn:
type: object
properties:
@ -5007,6 +5020,28 @@ components:
- metadata
- return_type
title: ScoringFn
StringType:
type: object
properties:
type:
type: string
const: string
default: string
additionalProperties: false
required:
- type
title: StringType
UnionType:
type: object
properties:
type:
type: string
const: union
default: union
additionalProperties: false
required:
- type
title: UnionType
Shield:
type: object
properties:
@ -5506,6 +5541,32 @@ components:
required:
- content
title: ToolInvocationResult
IterrowsResponse:
type: object
properties:
data:
type: array
items:
type: object
additionalProperties:
oneOf:
- type: 'null'
- type: boolean
- type: number
- type: string
- type: array
- type: object
description: The rows in the current page.
next_start_index:
type: integer
description: >-
Index into dataset for the first row in the next page. None if there are
no more rows.
additionalProperties: false
required:
- data
title: IterrowsResponse
description: A paginated list of rows from a dataset.
ListAgentSessionsResponse:
type: object
properties:
@ -6313,18 +6374,35 @@ components:
RegisterDatasetRequest:
type: object
properties:
dataset_id:
type: string
dataset_schema:
type: object
additionalProperties:
$ref: '#/components/schemas/ParamType'
url:
$ref: '#/components/schemas/URL'
provider_dataset_id:
type: string
provider_id:
purpose:
type: string
enum:
- post-training/messages
- eval/question-answer
- eval/messages-answer
description: >-
The purpose of the dataset. One of - "post-training/messages": The dataset
contains a messages column with list of messages for post-training. {
"messages": [ {"role": "user", "content": "Hello, world!"}, {"role": "assistant",
"content": "Hello, world!"}, ] } - "eval/question-answer": The dataset
contains a question column and an answer column for evaluation. { "question":
"What is the capital of France?", "answer": "Paris" } - "eval/messages-answer":
The dataset contains a messages column with list of messages and an answer
column for evaluation. { "messages": [ {"role": "user", "content": "Hello,
my name is John Doe."}, {"role": "assistant", "content": "Hello, John
Doe. How can I help you today?"}, {"role": "user", "content": "What's
my name?"}, ], "answer": "John Doe" }
source:
$ref: '#/components/schemas/DataSource'
description: >-
The data source of the dataset. Ensure that the data source schema is
compatible with the purpose of the dataset. Examples: - { "type": "uri",
"uri": "https://mywebsite.com/mydata.jsonl" } - { "type": "uri", "uri":
"lsfs://mydata.jsonl" } - { "type": "uri", "uri": "data:csv;base64,{base64_content}"
} - { "type": "uri", "uri": "huggingface://llamastack/simpleqa?split=train"
} - { "type": "rows", "rows": [ { "messages": [ {"role": "user", "content":
"Hello, world!"}, {"role": "assistant", "content": "Hello, world!"}, ]
} ] }
metadata:
type: object
additionalProperties:
@ -6335,11 +6413,16 @@ components:
- type: string
- type: array
- type: object
description: >-
The metadata for the dataset. - E.g. {"description": "My dataset"}
dataset_id:
type: string
description: >-
The ID of the dataset. If not provided, an ID will be generated.
additionalProperties: false
required:
- dataset_id
- dataset_schema
- url
- purpose
- source
title: RegisterDatasetRequest
RegisterModelRequest:
type: object
@ -6855,7 +6938,7 @@ tags:
- name: Eval
x-displayName: >-
Llama Stack Evaluation API for running evaluations on model and agent candidates.
- name: Files (Coming Soon)
- name: Files
- name: Inference
description: >-
This API provides the raw interface to the underlying models. Two kinds of models
@ -6893,7 +6976,7 @@ x-tagGroups:
- DatasetIO
- Datasets
- Eval
- Files (Coming Soon)
- Files
- Inference
- Inspect
- Models

2178
docs/notebooks/Alpha_Llama_Stack_Post_Training.ipynb
View file

File diff suppressed because one or more lines are too long

352
docs/notebooks/Llama_Stack_Benchmark_Evals.ipynb
View file

@ -84,16 +84,14 @@
"name": "stdout",
"output_type": "stream",
"text": [
"Not in Google Colab environment\n",
"\u001b[33mWarning: `bwrap` is not available. Code interpreter tool will not work correctly.\u001b[0m\n"
"Not in Google Colab environment\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"/opt/anaconda3/envs/master/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
" from .autonotebook import tqdm as notebook_tqdm\n"
"Warning: `bwrap` is not available. Code interpreter tool will not work correctly.\n"
]
},
{
@ -117,76 +115,146 @@
"- datasetio\n",
"- eval\n",
"- inference\n",
"- memory\n",
"- safety\n",
"- scoring\n",
"- telemetry\n",
"- tool_runtime\n",
"datasets: <span style=\"font-weight: bold\">[]</span>\n",
"container_image: null\n",
"- vector_io\n",
"benchmarks: <span style=\"font-weight: bold\">[]</span>\n",
"container_image: null\n",
"datasets: <span style=\"font-weight: bold\">[]</span>\n",
"image_name: together\n",
"memory_banks: <span style=\"font-weight: bold\">[]</span>\n",
"logging: null\n",
"metadata_store:\n",
" db_path: <span style=\"color: #800080; text-decoration-color: #800080\">/Users/xiyan/.llama/distributions/together/</span><span style=\"color: #ff00ff; text-decoration-color: #ff00ff\">registry.db</span>\n",
" namespace: null\n",
" type: sqlite\n",
"models:\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-8B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-8B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-8B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-8B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-70B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-70B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-70B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-70B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-405B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-405B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-405B-Instruct-FP8\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.1</span>-405B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-3B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-3B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-3B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-3B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-11B-Vision-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-11B-Vision-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-11B-Vision-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-11B-Vision-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-90B-Vision-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-90B-Vision-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-90B-Vision-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.2</span>-90B-Vision-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.3</span>-70B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.3</span>-70B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.3</span>-70B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3.3</span>-70B-Instruct-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Meta-Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-8B\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-8B\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-8B\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-8B\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-11B-Vision-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-11B-Vision-Turbo\n",
"- metadata: <span style=\"font-weight: bold\">{}</span>\n",
" model_id: meta-llama/Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-11B-Vision\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-Guard-<span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">3</span>-11B-Vision-Turbo\n",
"- metadata:\n",
" context_length: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">8192</span>\n",
" embedding_dimension: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">768</span>\n",
" model_id: togethercomputer/m2-bert-80M-8k-retrieval\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - embedding\n",
" provider_id: together\n",
" provider_model_id: togethercomputer/m2-bert-80M-8k-retrieval\n",
"- metadata:\n",
" context_length: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">32768</span>\n",
" embedding_dimension: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">768</span>\n",
" model_id: togethercomputer/m2-bert-80M-32k-retrieval\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - embedding\n",
" provider_id: together\n",
" provider_model_id: togethercomputer/m2-bert-80M-32k-retrieval\n",
"- metadata:\n",
" embedding_dimension: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">384</span>\n",
" model_id: all-MiniLM-L6-v2\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
@ -203,14 +271,26 @@
" provider_id: meta-reference\n",
" provider_type: inline::meta-reference\n",
" datasetio:\n",
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" - config:\n",
" kvstore:\n",
" db_path: <span style=\"color: #800080; text-decoration-color: #800080\">/Users/xiyan/.llama/distributions/together/</span><span style=\"color: #ff00ff; text-decoration-color: #ff00ff\">huggingface_datasetio.db</span>\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: huggingface\n",
" provider_type: remote::huggingface\n",
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" - config:\n",
" kvstore:\n",
" db_path: <span style=\"color: #800080; text-decoration-color: #800080\">/Users/xiyan/.llama/distributions/together/</span><span style=\"color: #ff00ff; text-decoration-color: #ff00ff\">localfs_datasetio.db</span>\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: localfs\n",
" provider_type: inline::localfs\n",
" eval:\n",
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" - config:\n",
" kvstore:\n",
" db_path: <span style=\"color: #800080; text-decoration-color: #800080\">/Users/xiyan/.llama/distributions/together/</span><span style=\"color: #ff00ff; text-decoration-color: #ff00ff\">meta_reference_eval.db</span>\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: meta-reference\n",
" provider_type: inline::meta-reference\n",
" inference:\n",
@ -222,16 +302,9 @@
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" provider_id: sentence-transformers\n",
" provider_type: inline::sentence-transformers\n",
" memory:\n",
" - config:\n",
" kvstore:\n",
" db_path: <span style=\"color: #800080; text-decoration-color: #800080\">/Users/xiyan/.llama/distributions/together/</span><span style=\"color: #ff00ff; text-decoration-color: #ff00ff\">faiss_store.db</span>\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: faiss\n",
" provider_type: inlin<span style=\"color: #00ff00; text-decoration-color: #00ff00; font-weight: bold\">e::fa</span>iss\n",
" safety:\n",
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" - config:\n",
" excluded_categories: <span style=\"font-weight: bold\">[]</span>\n",
" provider_id: llama-guard\n",
" provider_type: inline::llama-guard\n",
" scoring:\n",
@ -269,7 +342,26 @@
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" provider_id: rag-runtime\n",
" provider_type: inline::rag-runtime\n",
" - config: <span style=\"font-weight: bold\">{}</span>\n",
" provider_id: model-context-protocol\n",
" provider_type: remote::model-context-protocol\n",
" - config:\n",
" api_key: <span style=\"color: #008000; text-decoration-color: #008000\">'********'</span>\n",
" provider_id: wolfram-alpha\n",
" provider_type: remote::wolfram-alpha\n",
" vector_io:\n",
" - config:\n",
" kvstore:\n",
" db_path: <span style=\"color: #800080; text-decoration-color: #800080\">/Users/xiyan/.llama/distributions/together/</span><span style=\"color: #ff00ff; text-decoration-color: #ff00ff\">faiss_store.db</span>\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: faiss\n",
" provider_type: inlin<span style=\"color: #00ff00; text-decoration-color: #00ff00; font-weight: bold\">e::fa</span>iss\n",
"scoring_fns: <span style=\"font-weight: bold\">[]</span>\n",
"server:\n",
" port: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">8321</span>\n",
" tls_certfile: null\n",
" tls_keyfile: null\n",
"shields:\n",
"- params: null\n",
" provider_id: null\n",
@ -288,6 +380,11 @@
" mcp_endpoint: null\n",
" provider_id: code-interpreter\n",
" toolgroup_id: builtin::code_interpreter\n",
"- args: null\n",
" mcp_endpoint: null\n",
" provider_id: wolfram-alpha\n",
" toolgroup_id: builtin::wolfram_alpha\n",
"vector_dbs: <span style=\"font-weight: bold\">[]</span>\n",
"version: <span style=\"color: #008000; text-decoration-color: #008000\">'2'</span>\n",
"\n",
"</pre>\n"
@ -298,76 +395,146 @@
"- datasetio\n",
"- eval\n",
"- inference\n",
"- memory\n",
"- safety\n",
"- scoring\n",
"- telemetry\n",
"- tool_runtime\n",
"datasets: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
"container_image: null\n",
"- vector_io\n",
"benchmarks: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
"container_image: null\n",
"datasets: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
"image_name: together\n",
"memory_banks: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
"logging: null\n",
"metadata_store:\n",
" db_path: \u001b[35m/Users/xiyan/.llama/distributions/together/\u001b[0m\u001b[95mregistry.db\u001b[0m\n",
" namespace: null\n",
" type: sqlite\n",
"models:\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-8B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-8B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.1\u001b[0m-8B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-8B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-70B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-70B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.1\u001b[0m-70B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-70B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-405B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-405B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.1\u001b[0m-405B-Instruct-FP8\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-\u001b[1;36m3.1\u001b[0m-405B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-3B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-3B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-3B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-3B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-11B-Vision-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-11B-Vision-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-11B-Vision-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-11B-Vision-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-90B-Vision-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-90B-Vision-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-90B-Vision-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.2\u001b[0m-90B-Vision-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.3\u001b[0m-70B-Instruct-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.3\u001b[0m-70B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-\u001b[1;36m3.3\u001b[0m-70B-Instruct\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-\u001b[1;36m3.3\u001b[0m-70B-Instruct-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Meta-Llama-Guard-\u001b[1;36m3\u001b[0m-8B\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-Guard-\u001b[1;36m3\u001b[0m-8B\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-Guard-\u001b[1;36m3\u001b[0m-8B\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Meta-Llama-Guard-\u001b[1;36m3\u001b[0m-8B\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-Guard-\u001b[1;36m3\u001b[0m-11B-Vision-Turbo\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-Guard-\u001b[1;36m3\u001b[0m-11B-Vision-Turbo\n",
"- metadata: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" model_id: meta-llama/Llama-Guard-\u001b[1;36m3\u001b[0m-11B-Vision\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - llm\n",
" provider_id: together\n",
" provider_model_id: meta-llama/Llama-Guard-\u001b[1;36m3\u001b[0m-11B-Vision-Turbo\n",
"- metadata:\n",
" context_length: \u001b[1;36m8192\u001b[0m\n",
" embedding_dimension: \u001b[1;36m768\u001b[0m\n",
" model_id: togethercomputer/m2-bert-80M-8k-retrieval\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - embedding\n",
" provider_id: together\n",
" provider_model_id: togethercomputer/m2-bert-80M-8k-retrieval\n",
"- metadata:\n",
" context_length: \u001b[1;36m32768\u001b[0m\n",
" embedding_dimension: \u001b[1;36m768\u001b[0m\n",
" model_id: togethercomputer/m2-bert-80M-32k-retrieval\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
" - embedding\n",
" provider_id: together\n",
" provider_model_id: togethercomputer/m2-bert-80M-32k-retrieval\n",
"- metadata:\n",
" embedding_dimension: \u001b[1;36m384\u001b[0m\n",
" model_id: all-MiniLM-L6-v2\n",
" model_type: !!python/object/apply:llama_stack.apis.models.models.ModelType\n",
@ -384,14 +551,26 @@
" provider_id: meta-reference\n",
" provider_type: inline::meta-reference\n",
" datasetio:\n",
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" - config:\n",
" kvstore:\n",
" db_path: \u001b[35m/Users/xiyan/.llama/distributions/together/\u001b[0m\u001b[95mhuggingface_datasetio.db\u001b[0m\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: huggingface\n",
" provider_type: remote::huggingface\n",
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" - config:\n",
" kvstore:\n",
" db_path: \u001b[35m/Users/xiyan/.llama/distributions/together/\u001b[0m\u001b[95mlocalfs_datasetio.db\u001b[0m\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: localfs\n",
" provider_type: inline::localfs\n",
" eval:\n",
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" - config:\n",
" kvstore:\n",
" db_path: \u001b[35m/Users/xiyan/.llama/distributions/together/\u001b[0m\u001b[95mmeta_reference_eval.db\u001b[0m\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: meta-reference\n",
" provider_type: inline::meta-reference\n",
" inference:\n",
@ -403,16 +582,9 @@
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" provider_id: sentence-transformers\n",
" provider_type: inline::sentence-transformers\n",
" memory:\n",
" - config:\n",
" kvstore:\n",
" db_path: \u001b[35m/Users/xiyan/.llama/distributions/together/\u001b[0m\u001b[95mfaiss_store.db\u001b[0m\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: faiss\n",
" provider_type: inlin\u001b[1;92me::fa\u001b[0miss\n",
" safety:\n",
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" - config:\n",
" excluded_categories: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
" provider_id: llama-guard\n",
" provider_type: inline::llama-guard\n",
" scoring:\n",
@ -450,7 +622,26 @@
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" provider_id: rag-runtime\n",
" provider_type: inline::rag-runtime\n",
" - config: \u001b[1m{\u001b[0m\u001b[1m}\u001b[0m\n",
" provider_id: model-context-protocol\n",
" provider_type: remote::model-context-protocol\n",
" - config:\n",
" api_key: \u001b[32m'********'\u001b[0m\n",
" provider_id: wolfram-alpha\n",
" provider_type: remote::wolfram-alpha\n",
" vector_io:\n",
" - config:\n",
" kvstore:\n",
" db_path: \u001b[35m/Users/xiyan/.llama/distributions/together/\u001b[0m\u001b[95mfaiss_store.db\u001b[0m\n",
" namespace: null\n",
" type: sqlite\n",
" provider_id: faiss\n",
" provider_type: inlin\u001b[1;92me::fa\u001b[0miss\n",
"scoring_fns: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
"server:\n",
" port: \u001b[1;36m8321\u001b[0m\n",
" tls_certfile: null\n",
" tls_keyfile: null\n",
"shields:\n",
"- params: null\n",
" provider_id: null\n",
@ -469,6 +660,11 @@
" mcp_endpoint: null\n",
" provider_id: code-interpreter\n",
" toolgroup_id: builtin::code_interpreter\n",
"- args: null\n",
" mcp_endpoint: null\n",
" provider_id: wolfram-alpha\n",
" toolgroup_id: builtin::wolfram_alpha\n",
"vector_dbs: \u001b[1m[\u001b[0m\u001b[1m]\u001b[0m\n",
"version: \u001b[32m'2'\u001b[0m\n",
"\n"
]
@ -532,7 +728,7 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 3,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
@ -643,17 +839,7 @@
"id": "DJkmoG2kq1_P",
"outputId": "8493ee59-c6ff-4bb6-d787-f295944db1cf"
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Generating dev split: 100%|██████████| 5/5 [00:00<00:00, 139.81 examples/s]\n",
"Generating validation split: 100%|██████████| 30/30 [00:00<00:00, 258.29 examples/s]\n",
"Generating test split: 100%|██████████| 287/287 [00:01<00:00, 197.69 examples/s]\n"
]
}
],
"outputs": [],
"source": [
"import datasets\n",
"\n",
@ -676,7 +862,7 @@
},
{
"cell_type": "code",
"execution_count": 7,
"execution_count": 4,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
@ -691,7 +877,7 @@
"name": "stderr",
"output_type": "stream",
"text": [
"100%|██████████| 5/5 [00:42<00:00, 8.60s/it]\n"
"100%|██████████| 5/5 [00:33<00:00, 6.71s/it]\n"
]
},
{
@ -699,16 +885,18 @@
"text/html": [
"<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"color: #800080; text-decoration-color: #800080; font-weight: bold\">EvaluateResponse</span><span style=\"font-weight: bold\">(</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ </span><span style=\"color: #808000; text-decoration-color: #808000\">generations</span>=<span style=\"font-weight: bold\">[</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'Answer: D'</span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'The image shows a sunflower leaf with small, dark spots and white powdery patches. The dark spots are likely caused by a fungal pathogen, such as rust or septoria leaf spot, while the white powdery patches are likely caused by a fungal pathogen, such as powdery mildew.\\n\\nSince there are two distinct types of lesions on the leaf, it is likely that there are two different pathogens infecting the leaf.\\n\\n**Answer:** B) Two pathogens'</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'**Potato Pests**\\n\\nThe two insects depicted are:\\n\\n* **Colorado Potato Beetle (Leptinotarsa decemlineata)**: Characterized by black and yellow stripes, this beetle is a significant pest of potatoes. It feeds on the leaves and can cause substantial damage to the crop.\\n* **False Potato Beetle (Leptinotarsa juncta)**: Also known as the false Colorado beetle, this species has similar coloring but is not as harmful to potatoes as the Colorado potato beetle.'</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">\"The question requires the identification of the reason behind the massive gum production on the trunks of grapefruit trees in Cyprus, despite appearing healthy from a distance. The correct answer can be deduced by analyzing the symptoms and considering the possible causes.\\n\\nTo determine the correct answer, let's evaluate each option:\\n\\nA) Don't know or not sure: This option is incorrect because it does not provide a specific reason for the gum production.\\n\\nB) Physiological stress: This option is also incorrect because it is too broad and does not specifically explain the gum production.\\n\\nC) Bacterial disease: This option is incorrect because bacterial diseases typically cause different symptoms such as leaf spots, blights, or wilting.\\n\\nD) Harvesting damage when cutting with knives: This option is incorrect because harvesting damage would likely cause wounds or scars on the tree, but it would not lead to massive gum production.\\n\\nE) Fungal gummosis: This option is the most likely cause of the gum production. Fungal gummosis is a common disease in citrus trees, including grapefruit, that causes the production of gum or sap on the trunks and branches. The disease is typically caused by fungi such as Phytophthora or Diplodia, which infect the tree through wounds or natural openings. The gum production is a defense mechanism by the tree to try to seal off the infection and prevent further damage.\\n\\nTherefore, the correct answer is:\\n\\nAnswer: E\"</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">\"The image shows a sunflower leaf with a powdery mildew, which is a fungal disease caused by various species of fungi. The white powdery coating on the leaves is a characteristic symptom of this disease. The leaf also has some black spots, which could be indicative of a secondary infection or another type of disease. However, without more information or a closer examination, it's difficult to determine the exact cause of the black spots.\\n\\nBased on the image alone, we can see at least two types of symptoms: the powdery mildew and the black spots. This suggests that there may be more than one pathogen involved, but it's also possible that the black spots are a result of the same fungal infection causing the powdery mildew.\\n\\nAnswer: B) Two pathogens\"</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'The symptoms observed, characterized by the massive gum production on the trunks of the grapefruit trees in Cyprus, suggest a physiological or pathological response. Given the absence of visible signs of damage or pests from a higher point on a hillside, and considering the specific nature of the symptom (gum production), we can infer that the cause is more likely related to an internal process within the tree rather than external damage from harvesting. While physiological stress (B) could lead to such symptoms, the primary reason for gum production in trees, especially in citrus species, is typically linked to disease. Among the options provided, fungal gummosis (E) is a condition known to cause gumming in citrus trees, which aligns with the observed symptoms. Therefore, without direct evidence of external damage (harvesting) or confirmation of physiological stress being the primary cause, the most appropriate answer based on the information given is:\\n\\nAnswer: E'</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'Answer: D'</span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'**Causes of Splitting Petioles in Rhubarb**\\n\\nThe following factors can cause the petioles of rhubarb to split:\\n\\n* **Physiological Problems**: Issues such as water stress, nutrient deficiencies, or extreme temperatures can lead to splitting.\\n* **Phytoplasma Infection**: A bacterial infection caused by phytoplasma can lead to splitting of the petioles.\\n* **Animal Damage**: Pests like slugs, snails, or rodents can damage the plant and cause splitting.\\n* **Bacterial Infection**: Bacterial infections can also cause splitting.\\n\\nAs a result, the correct answer is:\\n\\n*Answer*: A) Physiological problems'</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">\"**Analysis of the Image**\\n\\nThe image provided shows a rhubarb plant with split petioles. To determine the cause of this issue, we need to consider various factors that could lead to such damage.\\n\\n**Possible Causes of Petiole Splitting**\\n\\n* **Physiological Problems**: Rhubarb plants can experience physiological stress due to environmental factors like extreme temperatures, waterlogging, or nutrient deficiencies. This stress can cause the petioles to split.\\n* **Phytoplasma Infection**: Phytoplasma is a type of bacteria that can infect plants, including rhubarb. It can cause symptoms such as yellowing leaves, stunted growth, and splitting of petioles.\\n* **Animal Damage**: Animals like rabbits, deer, or insects can damage rhubarb plants by eating the leaves or stems, which can lead to splitting of the petioles.\\n* **Bacteria**: Bacterial infections can also cause damage to rhubarb plants, including splitting of the petioles.\\n\\n**Conclusion**\\n\\nBased on the analysis, it is clear that all the options listed (A) Physiological problems, B) Phytoplasma infection, D) Animal damage, and E) Bacteria) could potentially cause the petioles of the rhubarb plant to split. Therefore, there is no single option that would not be a cause for the petioles splitting.\\n\\n**Answer**: C) I don't know and don't want to guess.\"</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">}</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ </span><span style=\"font-weight: bold\">]</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ </span><span style=\"color: #808000; text-decoration-color: #808000\">scores</span>=<span style=\"font-weight: bold\">{</span>\n",
@ -723,16 +911,18 @@
"text/plain": [
"\u001b[1;35mEvaluateResponse\u001b[0m\u001b[1m(\u001b[0m\n",
"\u001b[2;32m│ \u001b[0m\u001b[33mgenerations\u001b[0m=\u001b[1m[\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'Answer: D'\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'The image shows a sunflower leaf with small, dark spots and white powdery patches. The dark spots are likely caused by a fungal pathogen, such as rust or septoria leaf spot, while the white powdery patches are likely caused by a fungal pathogen, such as powdery mildew.\\n\\nSince there are two distinct types of lesions on the leaf, it is likely that there are two different pathogens infecting the leaf.\\n\\n**Answer:** B\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Two pathogens'\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'**Potato Pests**\\n\\nThe two insects depicted are:\\n\\n* **Colorado Potato Beetle \u001b[0m\u001b[32m(\u001b[0m\u001b[32mLeptinotarsa decemlineata\u001b[0m\u001b[32m)\u001b[0m\u001b[32m**: Characterized by black and yellow stripes, this beetle is a significant pest of potatoes. It feeds on the leaves and can cause substantial damage to the crop.\\n* **False Potato Beetle \u001b[0m\u001b[32m(\u001b[0m\u001b[32mLeptinotarsa juncta\u001b[0m\u001b[32m)\u001b[0m\u001b[32m**: Also known as the false Colorado beetle, this species has similar coloring but is not as harmful to potatoes as the Colorado potato beetle.'\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m\"The question requires the identification of the reason behind the massive gum production on the trunks of grapefruit trees in Cyprus, despite appearing healthy from a distance. The correct answer can be deduced by analyzing the symptoms and considering the possible causes.\\n\\nTo determine the correct answer, let's evaluate each option:\\n\\nA\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Don't know or not sure: This option is incorrect because it does not provide a specific reason for the gum production.\\n\\nB\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Physiological stress: This option is also incorrect because it is too broad and does not specifically explain the gum production.\\n\\nC\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Bacterial disease: This option is incorrect because bacterial diseases typically cause different symptoms such as leaf spots, blights, or wilting.\\n\\nD\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Harvesting damage when cutting with knives: This option is incorrect because harvesting damage would likely cause wounds or scars on the tree, but it would not lead to massive gum production.\\n\\nE\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Fungal gummosis: This option is the most likely cause of the gum production. Fungal gummosis is a common disease in citrus trees, including grapefruit, that causes the production of gum or sap on the trunks and branches. The disease is typically caused by fungi such as Phytophthora or Diplodia, which infect the tree through wounds or natural openings. The gum production is a defense mechanism by the tree to try to seal off the infection and prevent further damage.\\n\\nTherefore, the correct answer is:\\n\\nAnswer: E\"\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m\"The image shows a sunflower leaf with a powdery mildew, which is a fungal disease caused by various species of fungi. The white powdery coating on the leaves is a characteristic symptom of this disease. The leaf also has some black spots, which could be indicative of a secondary infection or another type of disease. However, without more information or a closer examination, it's difficult to determine the exact cause of the black spots.\\n\\nBased on the image alone, we can see at least two types of symptoms: the powdery mildew and the black spots. This suggests that there may be more than one pathogen involved, but it's also possible that the black spots are a result of the same fungal infection causing the powdery mildew.\\n\\nAnswer: B\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Two pathogens\"\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'The symptoms observed, characterized by the massive gum production on the trunks of the grapefruit trees in Cyprus, suggest a physiological or pathological response. Given the absence of visible signs of damage or pests from a higher point on a hillside, and considering the specific nature of the symptom \u001b[0m\u001b[32m(\u001b[0m\u001b[32mgum production\u001b[0m\u001b[32m)\u001b[0m\u001b[32m, we can infer that the cause is more likely related to an internal process within the tree rather than external damage from harvesting. While physiological stress \u001b[0m\u001b[32m(\u001b[0m\u001b[32mB\u001b[0m\u001b[32m)\u001b[0m\u001b[32m could lead to such symptoms, the primary reason for gum production in trees, especially in citrus species, is typically linked to disease. Among the options provided, fungal gummosis \u001b[0m\u001b[32m(\u001b[0m\u001b[32mE\u001b[0m\u001b[32m)\u001b[0m\u001b[32m is a condition known to cause gumming in citrus trees, which aligns with the observed symptoms. Therefore, without direct evidence of external damage \u001b[0m\u001b[32m(\u001b[0m\u001b[32mharvesting\u001b[0m\u001b[32m)\u001b[0m\u001b[32m or confirmation of physiological stress being the primary cause, the most appropriate answer based on the information given is:\\n\\nAnswer: E'\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'Answer: D'\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'**Causes of Splitting Petioles in Rhubarb**\\n\\nThe following factors can cause the petioles of rhubarb to split:\\n\\n* **Physiological Problems**: Issues such as water stress, nutrient deficiencies, or extreme temperatures can lead to splitting.\\n* **Phytoplasma Infection**: A bacterial infection caused by phytoplasma can lead to splitting of the petioles.\\n* **Animal Damage**: Pests like slugs, snails, or rodents can damage the plant and cause splitting.\\n* **Bacterial Infection**: Bacterial infections can also cause splitting.\\n\\nAs a result, the correct answer is:\\n\\n*Answer*: A\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Physiological problems'\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m\"**Analysis of the Image**\\n\\nThe image provided shows a rhubarb plant with split petioles. To determine the cause of this issue, we need to consider various factors that could lead to such damage.\\n\\n**Possible Causes of Petiole Splitting**\\n\\n* **Physiological Problems**: Rhubarb plants can experience physiological stress due to environmental factors like extreme temperatures, waterlogging, or nutrient deficiencies. This stress can cause the petioles to split.\\n* **Phytoplasma Infection**: Phytoplasma is a type of bacteria that can infect plants, including rhubarb. It can cause symptoms such as yellowing leaves, stunted growth, and splitting of petioles.\\n* **Animal Damage**: Animals like rabbits, deer, or insects can damage rhubarb plants by eating the leaves or stems, which can lead to splitting of the petioles.\\n* **Bacteria**: Bacterial infections can also cause damage to rhubarb plants, including splitting of the petioles.\\n\\n**Conclusion**\\n\\nBased on the analysis, it is clear that all the options listed \u001b[0m\u001b[32m(\u001b[0m\u001b[32mA\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Physiological problems, B\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Phytoplasma infection, D\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Animal damage, and E\u001b[0m\u001b[32m)\u001b[0m\u001b[32m Bacteria\u001b[0m\u001b[32m)\u001b[0m\u001b[32m could potentially cause the petioles of the rhubarb plant to split. Therefore, there is no single option that would not be a cause for the petioles splitting.\\n\\n**Answer**: C\u001b[0m\u001b[32m)\u001b[0m\u001b[32m I don't know and don't want to guess.\"\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m}\u001b[0m\n",
"\u001b[2;32m│ \u001b[0m\u001b[1m]\u001b[0m,\n",
"\u001b[2;32m│ \u001b[0m\u001b[33mscores\u001b[0m=\u001b[1m{\u001b[0m\n",
@ -815,7 +1005,7 @@
},
{
"cell_type": "code",
"execution_count": 8,
"execution_count": 5,
"metadata": {
"id": "HXmZf3Ymw-aX"
},
@ -823,39 +1013,33 @@
"source": [
"simpleqa_dataset_id = \"huggingface::simpleqa\"\n",
"\n",
"_ = client.datasets.register(\n",
"register_dataset_response = client.datasets.register(\n",
" purpose=\"eval/messages-answer\",\n",
" source={\n",
" \"type\": \"uri\",\n",
" \"uri\": \"huggingface://datasets/llamastack/simpleqa?split=train\",\n",
" },\n",
" dataset_id=simpleqa_dataset_id,\n",
" provider_id=\"huggingface\",\n",
" url={\"uri\": \"https://huggingface.co/datasets/llamastack/simpleqa\"},\n",
" metadata={\n",
" \"path\": \"llamastack/simpleqa\",\n",
" \"split\": \"train\",\n",
" },\n",
" dataset_schema={\n",
" \"input_query\": {\"type\": \"string\"},\n",
" \"expected_answer\": {\"type\": \"string\"},\n",
" \"chat_completion_input\": {\"type\": \"chat_completion_input\"},\n",
" },\n",
")\n"
")"
]
},
{
"cell_type": "code",
"execution_count": 9,
"execution_count": 6,
"metadata": {
"id": "Gc8azb4Rxr5J"
},
"outputs": [],
"source": [
"eval_rows = client.datasetio.get_rows_paginated(\n",
"eval_rows = client.datasets.iterrows(\n",
" dataset_id=simpleqa_dataset_id,\n",
" rows_in_page=5,\n",
")\n"
" limit=5,\n",
")"
]
},
{
"cell_type": "code",
"execution_count": 12,
"execution_count": 7,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
@ -876,7 +1060,7 @@
"name": "stderr",
"output_type": "stream",
"text": [
"100%|██████████| 5/5 [00:31<00:00, 6.38s/it]\n"
"100%|██████████| 5/5 [00:13<00:00, 2.71s/it]\n"
]
},
{
@ -889,14 +1073,14 @@
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">\"Radcliffe College was a women's liberal arts college in Cambridge, Massachusetts. However, it merged with Harvard University in 1977 and is now known as the Radcliffe Institute for Advanced Study at Harvard University.\"</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'I do not have information on the Leipzig 1877 tournament.'</span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'I am unable to verify in whose honor the Leipzig 1877 tournament was organized.'</span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'generated_answer'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">\"I am unable to verify what Empress Elizabeth of Austria's favorite sculpture depicted at her villa Achilleion at Corfu, according to Karl Küchler.\"</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"font-weight: bold\">}</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ </span><span style=\"font-weight: bold\">]</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ </span><span style=\"color: #808000; text-decoration-color: #808000\">scores</span>=<span style=\"font-weight: bold\">{</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ </span><span style=\"color: #008000; text-decoration-color: #008000\">'llm-as-judge::405b-simpleqa'</span>: <span style=\"color: #800080; text-decoration-color: #800080; font-weight: bold\">ScoringResult</span><span style=\"font-weight: bold\">(</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #808000; text-decoration-color: #808000\">aggregated_results</span>=<span style=\"font-weight: bold\">{}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #808000; text-decoration-color: #808000\">aggregated_results</span>=<span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'categorical_count'</span>: <span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'categorical_count'</span>: <span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'A'</span>: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">1</span>, <span style=\"color: #008000; text-decoration-color: #008000\">'C'</span>: <span style=\"color: #008080; text-decoration-color: #008080; font-weight: bold\">4</span><span style=\"font-weight: bold\">}}}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ </span><span style=\"color: #808000; text-decoration-color: #808000\">score_rows</span>=<span style=\"font-weight: bold\">[</span>\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ │ </span><span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'score'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'C'</span>, <span style=\"color: #008000; text-decoration-color: #008000\">'judge_feedback'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'C'</span><span style=\"font-weight: bold\">}</span>,\n",
"<span style=\"color: #7fbf7f; text-decoration-color: #7fbf7f\">│ │ │ │ </span><span style=\"font-weight: bold\">{</span><span style=\"color: #008000; text-decoration-color: #008000\">'score'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'C'</span>, <span style=\"color: #008000; text-decoration-color: #008000\">'judge_feedback'</span>: <span style=\"color: #008000; text-decoration-color: #008000\">'C'</span><span style=\"font-weight: bold\">}</span>,\n",
@ -917,14 +1101,14 @@
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m\"Radcliffe College was a women's liberal arts college in Cambridge, Massachusetts. However, it merged with Harvard University in 1977 and is now known as the Radcliffe Institute for Advanced Study at Harvard University.\"\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'I do not have information on the Leipzig 1877 tournament.'\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m'I am unable to verify in whose honor the Leipzig 1877 tournament was organized.'\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[32m'generated_answer'\u001b[0m: \u001b[32m\"I am unable to verify what Empress Elizabeth of Austria's favorite sculpture depicted at her villa Achilleion at Corfu, according to Karl Küchler.\"\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[1m}\u001b[0m\n",
"\u001b[2;32m│ \u001b[0m\u001b[1m]\u001b[0m,\n",
"\u001b[2;32m│ \u001b[0m\u001b[33mscores\u001b[0m=\u001b[1m{\u001b[0m\n",
"\u001b[2;32m│ │ \u001b[0m\u001b[32m'llm-as-judge::405b-simpleqa'\u001b[0m: \u001b[1;35mScoringResult\u001b[0m\u001b[1m(\u001b[0m\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[33maggregated_results\u001b[0m=\u001b[1m{\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[33maggregated_results\u001b[0m=\u001b[1m{\u001b[0m\u001b[32m'categorical_count'\u001b[0m: \u001b[1m{\u001b[0m\u001b[32m'categorical_count'\u001b[0m: \u001b[1m{\u001b[0m\u001b[32m'A'\u001b[0m: \u001b[1;36m1\u001b[0m, \u001b[32m'C'\u001b[0m: \u001b[1;36m4\u001b[0m\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ │ \u001b[0m\u001b[33mscore_rows\u001b[0m=\u001b[1m[\u001b[0m\n",
"\u001b[2;32m│ │ │ │ \u001b[0m\u001b[1m{\u001b[0m\u001b[32m'score'\u001b[0m: \u001b[32m'C'\u001b[0m, \u001b[32m'judge_feedback'\u001b[0m: \u001b[32m'C'\u001b[0m\u001b[1m}\u001b[0m,\n",
"\u001b[2;32m│ │ │ │ \u001b[0m\u001b[1m{\u001b[0m\u001b[32m'score'\u001b[0m: \u001b[32m'C'\u001b[0m, \u001b[32m'judge_feedback'\u001b[0m: \u001b[32m'C'\u001b[0m\u001b[1m}\u001b[0m,\n",
@ -957,7 +1141,7 @@
"\n",
"response = client.eval.evaluate_rows_alpha(\n",
" benchmark_id=\"meta-reference::simpleqa\",\n",
" input_rows=eval_rows.rows,\n",
" input_rows=eval_rows.data,\n",
" scoring_functions=[\"llm-as-judge::405b-simpleqa\"],\n",
" benchmark_config={\n",
" \"type\": \"benchmark\",\n",
@ -1106,7 +1290,7 @@
"\n",
"response = client.eval.evaluate_rows_alpha(\n",
" benchmark_id=\"meta-reference::simpleqa\",\n",
" input_rows=eval_rows.rows,\n",
" input_rows=eval_rows.data,\n",
" scoring_functions=[\"llm-as-judge::405b-simpleqa\"],\n",
" benchmark_config={\n",
" \"type\": \"benchmark\",\n",

39
docs/openapi_generator/pyopenapi/generator.py
View file

@ -435,7 +435,7 @@ class Generator:
)
self.schema_builder = SchemaBuilder(schema_generator)
self.responses = {}
# Create standard error responses
self._create_standard_error_responses()
@ -446,7 +446,7 @@ class Generator:
"""
# Get the Error schema
error_schema = self.schema_builder.classdef_to_ref(Error)
# Create standard error responses
self.responses["BadRequest400"] = Response(
description="The request was invalid or malformed",
@ -457,11 +457,11 @@ class Generator:
"status": 400,
"title": "Bad Request",
"detail": "The request was invalid or malformed",
}
},
)
}
},
)
self.responses["TooManyRequests429"] = Response(
description="The client has sent too many requests in a given amount of time",
content={
@ -471,11 +471,11 @@ class Generator:
"status": 429,
"title": "Too Many Requests",
"detail": "You have exceeded the rate limit. Please try again later.",
}
},
)
}
},
)
self.responses["InternalServerError500"] = Response(
description="The server encountered an unexpected error",
content={
@ -485,11 +485,11 @@ class Generator:
"status": 500,
"title": "Internal Server Error",
"detail": "An unexpected error occurred. Our team has been notified.",
}
},
)
}
},
)
# Add a default error response for any unhandled error cases
self.responses["DefaultError"] = Response(
description="An unexpected error occurred",
@ -500,9 +500,9 @@ class Generator:
"status": 0,
"title": "Error",
"detail": "An unexpected error occurred",
}
},
)
}
},
)
def _build_type_tag(self, ref: str, schema: Schema) -> Tag:
@ -547,11 +547,14 @@ class Generator:
"SyntheticDataGeneration",
"PostTraining",
"BatchInference",
"Files",
]:
op.defining_class.__name__ = f"{op.defining_class.__name__} (Coming Soon)"
print(op.defining_class.__name__)
# TODO (xiyan): temporary fix for datasetio inner impl + datasets api
# if op.defining_class.__name__ in ["DatasetIO"]:
# op.defining_class.__name__ = "Datasets"
doc_string = parse_type(op.func_ref)
doc_params = dict(
(param.name, param.description) for param in doc_string.params.values()
@ -598,7 +601,9 @@ class Generator:
# data passed in request body as raw bytes cannot have request parameters
if raw_bytes_request_body and op.request_params:
raise ValueError("Cannot have both raw bytes request body and request parameters")
raise ValueError(
"Cannot have both raw bytes request body and request parameters"
)
# data passed in request body as raw bytes
if raw_bytes_request_body:
@ -719,7 +724,7 @@ class Generator:
responses.update(response_builder.build_response(response_options))
assert len(responses.keys()) > 0, f"No responses found for {op.name}"
# Add standard error response references
if self.options.include_standard_error_responses:
if "400" not in responses:
@ -730,7 +735,7 @@ class Generator:
responses["500"] = ResponseRef("InternalServerError500")
if "default" not in responses:
responses["default"] = ResponseRef("DefaultError")
if op.event_type is not None:
builder = ContentBuilder(self.schema_builder)
callbacks = {

2
docs/source/contributing/new_api_provider.md
View file

@ -6,7 +6,7 @@ This guide will walk you through the process of adding a new API provider to Lla
- Begin by reviewing the [core concepts](../concepts/index.md) of Llama Stack and choose the API your provider belongs to (Inference, Safety, VectorIO, etc.)
- Determine the provider type ({repopath}`Remote::llama_stack/providers/remote` or {repopath}`Inline::llama_stack/providers/inline`). Remote providers make requests to external services, while inline providers execute implementation locally.
- Add your provider to the appropriate {repopath}`Registry::llama_stack/providers/registry/`. Specify pip dependencies necessary.
- Update any distribution {repopath}`Templates::llama_stack/templates/` build.yaml and run.yaml files if they should include your provider by default. Run {repopath}`llama_stack/scripts/distro_codegen.py` if necessary. Note that `distro_codegen.py` will fail if the new provider causes any distribution template to attempt to import provider-specific dependencies. This usually means the distribution's `get_distribution_template()` code path should only import any necessary Config or model alias definitions from each provider and not the provider's actual implementation.
- Update any distribution {repopath}`Templates::llama_stack/templates/` build.yaml and run.yaml files if they should include your provider by default. Run {repopath}`./scripts/distro_codegen.py` if necessary. Note that `distro_codegen.py` will fail if the new provider causes any distribution template to attempt to import provider-specific dependencies. This usually means the distribution's `get_distribution_template()` code path should only import any necessary Config or model alias definitions from each provider and not the provider's actual implementation.
Here are some example PRs to help you get started:

8
docs/source/distributions/building_distro.md
View file

@ -185,8 +185,12 @@ llama stack build --config llama_stack/templates/ollama/build.yaml
:::
:::{tab-item} Building Container
> [!TIP]
> Podman is supported as an alternative to Docker. Set `CONTAINER_BINARY` to `podman` in your environment to use Podman.
```{admonition} Podman Alternative
:class: tip
Podman is supported as an alternative to Docker. Set `CONTAINER_BINARY` to `podman` in your environment to use Podman.
```
To build a container image, you may start off from a template and use the `--image-type container` flag to specify `container` as the build image type.

12
docs/source/distributions/remote_hosted_distro/nvidia.md
View file

@ -6,13 +6,13 @@ The `llamastack/distribution-nvidia` distribution consists of the following prov
| API | Provider(s) |
|-----|-------------|
| agents | `inline::meta-reference` |
| datasetio | `remote::huggingface`, `inline::localfs` |
| datasetio | `inline::localfs` |
| eval | `inline::meta-reference` |
| inference | `remote::nvidia` |
| safety | `inline::llama-guard` |
| scoring | `inline::basic`, `inline::llm-as-judge`, `inline::braintrust` |
| safety | `remote::nvidia` |
| scoring | `inline::basic` |
| telemetry | `inline::meta-reference` |
| tool_runtime | `remote::brave-search`, `remote::tavily-search`, `inline::code-interpreter`, `inline::rag-runtime`, `remote::model-context-protocol` |
| tool_runtime | `inline::rag-runtime` |
| vector_io | `inline::faiss` |
@ -20,8 +20,10 @@ The `llamastack/distribution-nvidia` distribution consists of the following prov
The following environment variables can be configured:
- `LLAMASTACK_PORT`: Port for the Llama Stack distribution server (default: `5001`)
- `NVIDIA_API_KEY`: NVIDIA API Key (default: ``)
- `GUARDRAILS_SERVICE_URL`: URL for the NeMo Guardrails Service (default: `http://0.0.0.0:7331`)
- `INFERENCE_MODEL`: Inference model (default: `Llama3.1-8B-Instruct`)
- `SAFETY_MODEL`: Name of the model to use for safety (default: `meta/llama-3.1-8b-instruct`)
### Models

27
docs/source/playground/index.md
View file

@ -92,6 +92,8 @@ Interactive pages for users to play with and explore Llama Stack API capabilitie
## Starting the Llama Stack Playground
### Llama CLI
To start the Llama Stack Playground, run the following commands:
1. Start up the Llama Stack API server
@ -107,3 +109,28 @@ cd llama_stack/distribution/ui
pip install -r requirements.txt
streamlit run app.py
```
### Docker
Playground can also be started in a docker image:
```sh
export LLAMA_STACK_URL=http://localhost:11434
docker run \
-p 8501:8501 \
-e LLAMA_STACK_ENDPOINT=$LLAMA_STACK_URL \
quay.io/jland/llama-stack-playground
```
## Configurable Environment Variables
## Environment Variables
| Environment Variable | Description | Default Value |
|----------------------------|------------------------------------|---------------------------|
| LLAMA_STACK_ENDPOINT | The endpoint for the Llama Stack | http://localhost:8321 |
| FIREWORKS_API_KEY | API key for Fireworks provider | (empty string) |
| TOGETHER_API_KEY | API key for Together provider | (empty string) |
| SAMBANOVA_API_KEY | API key for SambaNova provider | (empty string) |
| OPENAI_API_KEY | API key for OpenAI provider | (empty string) |

24
docs/source/references/evals_reference/index.md
View file

@ -114,23 +114,17 @@ pprint(response)
simpleqa_dataset_id = "huggingface::simpleqa"
_ = client.datasets.register(
purpose="eval/messages-answer",
source={
"type": "uri",
"uri": "huggingface://datasets/llamastack/simpleqa?split=train",
},
dataset_id=simpleqa_dataset_id,
provider_id="huggingface",
url={"uri": "https://huggingface.co/datasets/llamastack/simpleqa"},
metadata={
"path": "llamastack/simpleqa",
"split": "train",
},
dataset_schema={
"input_query": {"type": "string"},
"expected_answer": {"type": "string"},
"chat_completion_input": {"type": "chat_completion_input"},
},
)
eval_rows = client.datasetio.get_rows_paginated(
eval_rows = client.datasets.iterrows(
dataset_id=simpleqa_dataset_id,
rows_in_page=5,
limit=5,
)
```
@ -143,7 +137,7 @@ client.benchmarks.register(
response = client.eval.evaluate_rows(
benchmark_id="meta-reference::simpleqa",
input_rows=eval_rows.rows,
input_rows=eval_rows.data,
scoring_functions=["llm-as-judge::405b-simpleqa"],
benchmark_config={
"eval_candidate": {
@ -191,7 +185,7 @@ agent_config = {
response = client.eval.evaluate_rows(
benchmark_id="meta-reference::simpleqa",
input_rows=eval_rows.rows,
input_rows=eval_rows.data,
scoring_functions=["llm-as-judge::405b-simpleqa"],
benchmark_config={
"eval_candidate": {

109
docs/source/references/llama_stack_client_cli_reference.md
View file

@ -6,17 +6,32 @@ The `llama-stack-client` CLI allows you to query information about the distribut
### `llama-stack-client`
```bash
llama-stack-client -h
llama-stack-client
Usage: llama-stack-client [OPTIONS] COMMAND [ARGS]...
usage: llama-stack-client [-h] {models,memory_banks,shields} ...
Welcome to the LlamaStackClient CLI
Welcome to the LlamaStackClient CLI
Options:
--version Show the version and exit.
--endpoint TEXT Llama Stack distribution endpoint
--api-key TEXT Llama Stack distribution API key
--config TEXT Path to config file
--help Show this message and exit.
options:
-h, --help show this help message and exit
subcommands:
{models,memory_banks,shields}
Commands:
configure Configure Llama Stack Client CLI.
datasets Manage datasets.
eval Run evaluation tasks.
eval_tasks Manage evaluation tasks.
inference Inference (chat).
inspect Inspect server configuration.
models Manage GenAI models.
post_training Post-training.
providers Manage API providers.
scoring_functions Manage scoring functions.
shields Manage safety shield services.
toolgroups Manage available tool groups.
vector_dbs Manage vector databases.
```
### `llama-stack-client configure`
@ -127,11 +142,11 @@ llama-stack-client vector_dbs list
llama-stack-client vector_dbs register <vector-db-id> [--provider-id <provider-id>] [--provider-vector-db-id <provider-vector-db-id>] [--embedding-model <embedding-model>] [--embedding-dimension <embedding-dimension>]
```
Options:
- `--provider-id`: Optional. Provider ID for the vector db
- `--provider-vector-db-id`: Optional. Provider's vector db ID
- `--embedding-model`: Optional. Embedding model to use. Default: "all-MiniLM-L6-v2"
- `--embedding-dimension`: Optional. Dimension of embeddings. Default: 384
Optional arguments:
- `--provider-id`: Provider ID for the vector db
- `--provider-vector-db-id`: Provider's vector db ID
- `--embedding-model`: Embedding model to use. Default: "all-MiniLM-L6-v2"
- `--embedding-dimension`: Dimension of embeddings. Default: 384
### `llama-stack-client vector_dbs unregister`
```bash
@ -157,11 +172,13 @@ llama-stack-client shields list
llama-stack-client shields register --shield-id <shield-id> [--provider-id <provider-id>] [--provider-shield-id <provider-shield-id>] [--params <params>]
```
Options:
- `--shield-id`: Required. ID of the shield
- `--provider-id`: Optional. Provider ID for the shield
- `--provider-shield-id`: Optional. Provider's shield ID
- `--params`: Optional. JSON configuration parameters for the shield
Required arguments:
- `--shield-id`: ID of the shield
Optional arguments:
- `--provider-id`: Provider ID for the shield
- `--provider-shield-id`: Provider's shield ID
- `--params`: JSON configuration parameters for the shield
## Eval Task Management
@ -175,13 +192,15 @@ llama-stack-client benchmarks list
llama-stack-client benchmarks register --eval-task-id <eval-task-id> --dataset-id <dataset-id> --scoring-functions <function1> [<function2> ...] [--provider-id <provider-id>] [--provider-eval-task-id <provider-eval-task-id>] [--metadata <metadata>]
```
Options:
- `--eval-task-id`: Required. ID of the eval task
- `--dataset-id`: Required. ID of the dataset to evaluate
- `--scoring-functions`: Required. One or more scoring functions to use for evaluation
- `--provider-id`: Optional. Provider ID for the eval task
- `--provider-eval-task-id`: Optional. Provider's eval task ID
- `--metadata`: Optional. Metadata for the eval task in JSON format
Required arguments:
- `--eval-task-id`: ID of the eval task
- `--dataset-id`: ID of the dataset to evaluate
- `--scoring-functions`: One or more scoring functions to use for evaluation
Optional arguments:
- `--provider-id`: Provider ID for the eval task
- `--provider-eval-task-id`: Provider's eval task ID
- `--metadata`: Metadata for the eval task in JSON format
## Eval execution
### `llama-stack-client eval run-benchmark`
@ -189,11 +208,13 @@ Options:
llama-stack-client eval run-benchmark <eval-task-id1> [<eval-task-id2> ...] --eval-task-config <config-file> --output-dir <output-dir> [--num-examples <num>] [--visualize]
```
Options:
- `--eval-task-config`: Required. Path to the eval task config file in JSON format
- `--output-dir`: Required. Path to the directory where evaluation results will be saved
- `--num-examples`: Optional. Number of examples to evaluate (useful for debugging)
- `--visualize`: Optional flag. If set, visualizes evaluation results after completion
Required arguments:
- `--eval-task-config`: Path to the eval task config file in JSON format
- `--output-dir`: Path to the directory where evaluation results will be saved
Optional arguments:
- `--num-examples`: Number of examples to evaluate (useful for debugging)
- `--visualize`: If set, visualizes evaluation results after completion
Example benchmark_config.json:
```json
@ -214,11 +235,13 @@ Example benchmark_config.json:
llama-stack-client eval run-scoring <eval-task-id> --eval-task-config <config-file> --output-dir <output-dir> [--num-examples <num>] [--visualize]
```
Options:
- `--eval-task-config`: Required. Path to the eval task config file in JSON format
- `--output-dir`: Required. Path to the directory where scoring results will be saved
- `--num-examples`: Optional. Number of examples to evaluate (useful for debugging)
- `--visualize`: Optional flag. If set, visualizes scoring results after completion
Required arguments:
- `--eval-task-config`: Path to the eval task config file in JSON format
- `--output-dir`: Path to the directory where scoring results will be saved
Optional arguments:
- `--num-examples`: Number of examples to evaluate (useful for debugging)
- `--visualize`: If set, visualizes scoring results after completion
## Tool Group Management
@ -230,11 +253,11 @@ llama-stack-client toolgroups list
+---------------------------+------------------+------+---------------+
| identifier | provider_id | args | mcp_endpoint |
+===========================+==================+======+===============+
| builtin::code_interpreter | code-interpreter | None | None |
| builtin::code_interpreter | code-interpreter | None | None |
+---------------------------+------------------+------+---------------+
| builtin::rag | rag-runtime | None | None |
| builtin::rag | rag-runtime | None | None |
+---------------------------+------------------+------+---------------+
| builtin::websearch | tavily-search | None | None |
| builtin::websearch | tavily-search | None | None |
+---------------------------+------------------+------+---------------+
```
@ -250,11 +273,11 @@ Shows detailed information about a specific toolgroup. If the toolgroup is not f
llama-stack-client toolgroups register <toolgroup_id> [--provider-id <provider-id>] [--provider-toolgroup-id <provider-toolgroup-id>] [--mcp-config <mcp-config>] [--args <args>]
```
Options:
- `--provider-id`: Optional. Provider ID for the toolgroup
- `--provider-toolgroup-id`: Optional. Provider's toolgroup ID
- `--mcp-config`: Optional. JSON configuration for the MCP endpoint
- `--args`: Optional. JSON arguments for the toolgroup
Optional arguments:
- `--provider-id`: Provider ID for the toolgroup
- `--provider-toolgroup-id`: Provider's toolgroup ID
- `--mcp-config`: JSON configuration for the MCP endpoint
- `--args`: JSON arguments for the toolgroup
### `llama-stack-client toolgroups unregister`
```bash

34
llama_stack/apis/datasetio/datasetio.py
View file

@ -13,19 +13,16 @@ from llama_stack.schema_utils import json_schema_type, webmethod
@json_schema_type
class PaginatedRowsResult(BaseModel):
class IterrowsResponse(BaseModel):
"""
A paginated list of rows from a dataset.
:param rows: The rows in the current page.
:param total_count: The total number of rows in the dataset.
:param next_page_token: The token to get the next page of rows.
:param data: The rows in the current page.
:param next_start_index: Index into dataset for the first row in the next page. None if there are no more rows.
"""
# the rows obey the DatasetSchema for the given dataset
rows: List[Dict[str, Any]]
total_count: int
next_page_token: Optional[str] = None
data: List[Dict[str, Any]]
next_start_index: Optional[int] = None
class DatasetStore(Protocol):
@ -37,22 +34,21 @@ class DatasetIO(Protocol):
# keeping for aligning with inference/safety, but this is not used
dataset_store: DatasetStore
@webmethod(route="/datasetio/rows", method="GET")
async def get_rows_paginated(
# TODO(xiyan): there's a flakiness here where setting route to "/datasets/" here will not result in proper routing
@webmethod(route="/datasetio/iterrows/{dataset_id:path}", method="GET")
async def iterrows(
self,
dataset_id: str,
rows_in_page: int,
page_token: Optional[str] = None,
filter_condition: Optional[str] = None,
) -> PaginatedRowsResult:
"""Get a paginated list of rows from a dataset.
start_index: Optional[int] = None,
limit: Optional[int] = None,
) -> IterrowsResponse:
"""Get a paginated list of rows from a dataset. Uses cursor-based pagination.
:param dataset_id: The ID of the dataset to get the rows from.
:param rows_in_page: The number of rows to get per page.
:param page_token: The token to get the next page of rows.
:param filter_condition: (Optional) A condition to filter the rows by.
:param start_index: Index into dataset for the first row to get. Get all rows if None.
:param limit: The number of rows to get.
"""
...
@webmethod(route="/datasetio/rows", method="POST")
@webmethod(route="/datasetio/append-rows/{dataset_id:path}", method="POST")
async def append_rows(self, dataset_id: str, rows: List[Dict[str, Any]]) -> None: ...

163
llama_stack/apis/datasets/datasets.py
View file

@ -4,19 +4,102 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from typing import Any, Dict, List, Literal, Optional, Protocol
from enum import Enum
from typing import Annotated, Any, Dict, List, Literal, Optional, Protocol, Union
from pydantic import BaseModel, Field
from llama_stack.apis.common.content_types import URL
from llama_stack.apis.common.type_system import ParamType
from llama_stack.apis.resource import Resource, ResourceType
from llama_stack.schema_utils import json_schema_type, webmethod
from llama_stack.schema_utils import json_schema_type, register_schema, webmethod
class DatasetPurpose(str, Enum):
"""
Purpose of the dataset. Each purpose has a required input data schema.
:cvar post-training/messages: The dataset contains messages used for post-training.
{
"messages": [
{"role": "user", "content": "Hello, world!"},
{"role": "assistant", "content": "Hello, world!"},
]
}
:cvar eval/question-answer: The dataset contains a question column and an answer column.
{
"question": "What is the capital of France?",
"answer": "Paris"
}
:cvar eval/messages-answer: The dataset contains a messages column with list of messages and an answer column.
{
"messages": [
{"role": "user", "content": "Hello, my name is John Doe."},
{"role": "assistant", "content": "Hello, John Doe. How can I help you today?"},
{"role": "user", "content": "What's my name?"},
],
"answer": "John Doe"
}
"""
post_training_messages = "post-training/messages"
eval_question_answer = "eval/question-answer"
eval_messages_answer = "eval/messages-answer"
# TODO: add more schemas here
class DatasetType(Enum):
"""
Type of the dataset source.
:cvar uri: The dataset can be obtained from a URI.
:cvar rows: The dataset is stored in rows.
"""
uri = "uri"
rows = "rows"
@json_schema_type
class URIDataSource(BaseModel):
"""A dataset that can be obtained from a URI.
:param uri: The dataset can be obtained from a URI. E.g.
- "https://mywebsite.com/mydata.jsonl"
- "lsfs://mydata.jsonl"
- "data:csv;base64,{base64_content}"
"""
type: Literal["uri"] = "uri"
uri: str
@json_schema_type
class RowsDataSource(BaseModel):
"""A dataset stored in rows.
:param rows: The dataset is stored in rows. E.g.
- [
{"messages": [{"role": "user", "content": "Hello, world!"}, {"role": "assistant", "content": "Hello, world!"}]}
]
"""
type: Literal["rows"] = "rows"
rows: List[Dict[str, Any]]
DataSource = register_schema(
Annotated[
Union[URIDataSource, RowsDataSource],
Field(discriminator="type"),
],
name="DataSource",
)
class CommonDatasetFields(BaseModel):
dataset_schema: Dict[str, ParamType]
url: URL
"""
Common fields for a dataset.
"""
purpose: DatasetPurpose
source: DataSource
metadata: Dict[str, Any] = Field(
default_factory=dict,
description="Any additional metadata for this dataset",
@ -50,13 +133,69 @@ class Datasets(Protocol):
@webmethod(route="/datasets", method="POST")
async def register_dataset(
self,
dataset_id: str,
dataset_schema: Dict[str, ParamType],
url: URL,
provider_dataset_id: Optional[str] = None,
provider_id: Optional[str] = None,
purpose: DatasetPurpose,
source: DataSource,
metadata: Optional[Dict[str, Any]] = None,
) -> None: ...
dataset_id: Optional[str] = None,
) -> Dataset:
"""
Register a new dataset.
:param purpose: The purpose of the dataset. One of
- "post-training/messages": The dataset contains a messages column with list of messages for post-training.
{
"messages": [
{"role": "user", "content": "Hello, world!"},
{"role": "assistant", "content": "Hello, world!"},
]
}
- "eval/question-answer": The dataset contains a question column and an answer column for evaluation.
{
"question": "What is the capital of France?",
"answer": "Paris"
}
- "eval/messages-answer": The dataset contains a messages column with list of messages and an answer column for evaluation.
{
"messages": [
{"role": "user", "content": "Hello, my name is John Doe."},
{"role": "assistant", "content": "Hello, John Doe. How can I help you today?"},
{"role": "user", "content": "What's my name?"},
],
"answer": "John Doe"
}
:param source: The data source of the dataset. Ensure that the data source schema is compatible with the purpose of the dataset. Examples:
- {
"type": "uri",
"uri": "https://mywebsite.com/mydata.jsonl"
}
- {
"type": "uri",
"uri": "lsfs://mydata.jsonl"
}
- {
"type": "uri",
"uri": "data:csv;base64,{base64_content}"
}
- {
"type": "uri",
"uri": "huggingface://llamastack/simpleqa?split=train"
}
- {
"type": "rows",
"rows": [
{
"messages": [
{"role": "user", "content": "Hello, world!"},
{"role": "assistant", "content": "Hello, world!"},
]
}
]
}
:param metadata: The metadata for the dataset.
- E.g. {"description": "My dataset"}
:param dataset_id: The ID of the dataset. If not provided, an ID will be generated.
"""
...
@webmethod(route="/datasets/{dataset_id:path}", method="GET")
async def get_dataset(

4
llama_stack/cli/stack/_build.py
View file

@ -38,7 +38,7 @@ from llama_stack.distribution.distribution import get_provider_registry
from llama_stack.distribution.resolver import InvalidProviderError
from llama_stack.distribution.utils.config_dirs import DISTRIBS_BASE_DIR
from llama_stack.distribution.utils.dynamic import instantiate_class_type
from llama_stack.distribution.utils.exec import formulate_run_args, run_with_pty
from llama_stack.distribution.utils.exec import formulate_run_args, run_command
from llama_stack.distribution.utils.image_types import LlamaStackImageType
from llama_stack.providers.datatypes import Api
@ -213,7 +213,7 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
config = parse_and_maybe_upgrade_config(config_dict)
run_args = formulate_run_args(args.image_type, args.image_name, config, args.template)
run_args.extend([run_config, str(os.getenv("LLAMA_STACK_PORT", 8321))])
run_with_pty(run_args)
run_command(run_args)
def _generate_run_config(

4
llama_stack/cli/stack/run.py
View file

@ -82,7 +82,7 @@ class StackRun(Subcommand):
from llama_stack.distribution.configure import parse_and_maybe_upgrade_config
from llama_stack.distribution.utils.config_dirs import DISTRIBS_BASE_DIR
from llama_stack.distribution.utils.exec import formulate_run_args, run_with_pty
from llama_stack.distribution.utils.exec import formulate_run_args, run_command
config_file = Path(args.config)
has_yaml_suffix = args.config.endswith(".yaml")
@ -136,4 +136,4 @@ class StackRun(Subcommand):
if args.tls_keyfile and args.tls_certfile:
run_args.extend(["--tls-keyfile", args.tls_keyfile, "--tls-certfile", args.tls_certfile])
run_with_pty(run_args)
run_command(run_args)

9
llama_stack/distribution/build.py
View file

@ -6,7 +6,6 @@
import importlib.resources
import logging
import sys
from pathlib import Path
from typing import Dict, List
@ -15,7 +14,7 @@ from termcolor import cprint
from llama_stack.distribution.datatypes import BuildConfig, Provider
from llama_stack.distribution.distribution import get_provider_registry
from llama_stack.distribution.utils.exec import run_command, run_with_pty
from llama_stack.distribution.utils.exec import run_command
from llama_stack.distribution.utils.image_types import LlamaStackImageType
from llama_stack.providers.datatypes import Api
@ -123,11 +122,7 @@ def build_image(
if special_deps:
args.append("#".join(special_deps))
is_terminal = sys.stdin.isatty()
if is_terminal:
return_code = run_with_pty(args)
else:
return_code = run_command(args)
return_code = run_command(args)
if return_code != 0:
log.error(

5
llama_stack/distribution/build_container.sh
View file

@ -43,7 +43,7 @@ RED='\033[0;31m'
NC='\033[0m' # No Color
CONTAINER_BINARY=${CONTAINER_BINARY:-docker}
CONTAINER_OPTS=${CONTAINER_OPTS:-}
CONTAINER_OPTS=${CONTAINER_OPTS:---progress=plain}
TEMP_DIR=$(mktemp -d)
@ -253,8 +253,7 @@ $CONTAINER_BINARY build \
"${CLI_ARGS[@]}" \
-t "$image_tag" \
-f "$TEMP_DIR/Containerfile" \
"." \
--progress=plain
"."
# clean up tmp/configs
set +x

7
llama_stack/distribution/providers.py
View file

@ -8,10 +8,13 @@
from pydantic import BaseModel
from llama_stack.apis.providers import ListProvidersResponse, ProviderInfo, Providers
from llama_stack.log import get_logger
from .datatypes import StackRunConfig
from .stack import redact_sensitive_fields
logger = get_logger(name=__name__, category="core")
class ProviderImplConfig(BaseModel):
run_config: StackRunConfig
@ -31,6 +34,10 @@ class ProviderImpl(Providers):
async def initialize(self) -> None:
pass
async def shutdown(self) -> None:
logger.debug("ProviderImpl.shutdown")
pass
async def list_providers(self) -> ListProvidersResponse:
run_config = self.config.run_config
safe_config = StackRunConfig(**redact_sensitive_fields(run_config.model_dump()))

51
llama_stack/distribution/routers/routers.py
View file

@ -12,7 +12,8 @@ from llama_stack.apis.common.content_types import (
InterleavedContent,
InterleavedContentItem,
)
from llama_stack.apis.datasetio import DatasetIO, PaginatedRowsResult
from llama_stack.apis.datasetio import DatasetIO, IterrowsResponse
from llama_stack.apis.datasets import DatasetPurpose, DataSource
from llama_stack.apis.eval import (
BenchmarkConfig,
Eval,
@ -160,7 +161,11 @@ class InferenceRouter(Inference):
await self.routing_table.register_model(model_id, provider_model_id, provider_id, metadata, model_type)
def _construct_metrics(
self, prompt_tokens: int, completion_tokens: int, total_tokens: int, model: Model
self,
prompt_tokens: int,
completion_tokens: int,
total_tokens: int,
model: Model,
) -> List[MetricEvent]:
"""Constructs a list of MetricEvent objects containing token usage metrics.
@ -298,7 +303,12 @@ class InferenceRouter(Inference):
completion_text += chunk.event.delta.text
if chunk.event.event_type == ChatCompletionResponseEventType.complete:
completion_tokens = await self._count_tokens(
[CompletionMessage(content=completion_text, stop_reason=StopReason.end_of_turn)],
[
CompletionMessage(
content=completion_text,
stop_reason=StopReason.end_of_turn,
)
],
tool_config.tool_prompt_format,
)
total_tokens = (prompt_tokens or 0) + (completion_tokens or 0)
@ -471,21 +481,36 @@ class DatasetIORouter(DatasetIO):
logger.debug("DatasetIORouter.shutdown")
pass
async def get_rows_paginated(
async def register_dataset(
self,
purpose: DatasetPurpose,
source: DataSource,
metadata: Optional[Dict[str, Any]] = None,
dataset_id: Optional[str] = None,
) -> None:
logger.debug(
f"DatasetIORouter.register_dataset: {purpose=} {source=} {metadata=} {dataset_id=}",
)
await self.routing_table.register_dataset(
purpose=purpose,
source=source,
metadata=metadata,
dataset_id=dataset_id,
)
async def iterrows(
self,
dataset_id: str,
rows_in_page: int,
page_token: Optional[str] = None,
filter_condition: Optional[str] = None,
) -> PaginatedRowsResult:
start_index: Optional[int] = None,
limit: Optional[int] = None,
) -> IterrowsResponse:
logger.debug(
f"DatasetIORouter.get_rows_paginated: {dataset_id}, rows_in_page={rows_in_page}",
f"DatasetIORouter.iterrows: {dataset_id}, {start_index=} {limit=}",
)
return await self.routing_table.get_provider_impl(dataset_id).get_rows_paginated(
return await self.routing_table.get_provider_impl(dataset_id).iterrows(
dataset_id=dataset_id,
rows_in_page=rows_in_page,
page_token=page_token,
filter_condition=filter_condition,
start_index=start_index,
limit=limit,
)
async def append_rows(self, dataset_id: str, rows: List[Dict[str, Any]]) -> None:

52
llama_stack/distribution/routers/routing_tables.py
View file

@ -5,6 +5,7 @@
# the root directory of this source tree.
import logging
import uuid
from typing import Any, Dict, List, Optional
from pydantic import TypeAdapter
@ -12,7 +13,14 @@ from pydantic import TypeAdapter
from llama_stack.apis.benchmarks import Benchmark, Benchmarks, ListBenchmarksResponse
from llama_stack.apis.common.content_types import URL
from llama_stack.apis.common.type_system import ParamType
from llama_stack.apis.datasets import Dataset, Datasets, ListDatasetsResponse
from llama_stack.apis.datasets import (
Dataset,
DatasetPurpose,
Datasets,
DatasetType,
DataSource,
ListDatasetsResponse,
)
from llama_stack.apis.models import ListModelsResponse, Model, Models, ModelType
from llama_stack.apis.resource import ResourceType
from llama_stack.apis.scoring_functions import (
@ -352,34 +360,42 @@ class DatasetsRoutingTable(CommonRoutingTableImpl, Datasets):
async def register_dataset(
self,
dataset_id: str,
dataset_schema: Dict[str, ParamType],
url: URL,
provider_dataset_id: Optional[str] = None,
provider_id: Optional[str] = None,
purpose: DatasetPurpose,
source: DataSource,
metadata: Optional[Dict[str, Any]] = None,
) -> None:
if provider_dataset_id is None:
provider_dataset_id = dataset_id
if provider_id is None:
# If provider_id not specified, use the only provider if it supports this dataset
if len(self.impls_by_provider_id) == 1:
provider_id = list(self.impls_by_provider_id.keys())[0]
dataset_id: Optional[str] = None,
) -> Dataset:
if not dataset_id:
dataset_id = f"dataset-{str(uuid.uuid4())}"
provider_dataset_id = dataset_id
# infer provider from source
if source.type == DatasetType.rows.value:
provider_id = "localfs"
elif source.type == DatasetType.uri.value:
# infer provider from uri
if source.uri.startswith("huggingface"):
provider_id = "huggingface"
else:
raise ValueError(
f"No provider specified and multiple providers available. Please specify a provider_id. Available providers: {self.impls_by_provider_id.keys()}"
)
provider_id = "localfs"
else:
raise ValueError(f"Unknown data source type: {source.type}")
if metadata is None:
metadata = {}
dataset = Dataset(
identifier=dataset_id,
provider_resource_id=provider_dataset_id,
provider_id=provider_id,
dataset_schema=dataset_schema,
url=url,
purpose=purpose,
source=source,
metadata=metadata,
)
await self.register_object(dataset)
return dataset
async def unregister_dataset(self, dataset_id: str) -> None:
dataset = await self.get_dataset(dataset_id)

11
llama_stack/distribution/ui/Containerfile Normal file
View file

@ -0,0 +1,11 @@
# More info on playground configuration can be found here:
# https://llama-stack.readthedocs.io/en/latest/playground
FROM python:3.9-slim
WORKDIR /app
COPY . /app/
RUN /usr/local/bin/python -m pip install --upgrade pip && \
/usr/local/bin/pip3 install -r requirements.txt
EXPOSE 8501
ENTRYPOINT ["streamlit", "run", "app.py", "--server.port=8501", "--server.address=0.0.0.0"]

10
llama_stack/distribution/ui/README.md
View file

@ -40,3 +40,13 @@ cd llama_stack/distribution/ui
pip install -r requirements.txt
streamlit run app.py
```
## Environment Variables
| Environment Variable | Description | Default Value |
|----------------------------|------------------------------------|---------------------------|
| LLAMA_STACK_ENDPOINT | The endpoint for the Llama Stack | http://localhost:8321 |
| FIREWORKS_API_KEY | API key for Fireworks provider | (empty string) |
| TOGETHER_API_KEY | API key for Together provider | (empty string) |
| SAMBANOVA_API_KEY | API key for SambaNova provider | (empty string) |
| OPENAI_API_KEY | API key for OpenAI provider | (empty string) |

7
llama_stack/distribution/ui/page/evaluations/native_eval.py
View file

@ -166,11 +166,10 @@ def run_evaluation_3():
eval_candidate = st.session_state["eval_candidate"]
dataset_id = benchmarks[selected_benchmark].dataset_id
rows = llama_stack_api.client.datasetio.get_rows_paginated(
rows = llama_stack_api.client.datasets.iterrows(
dataset_id=dataset_id,
rows_in_page=-1,
)
total_rows = len(rows.rows)
total_rows = len(rows.data)
# Add number of examples control
num_rows = st.number_input(
"Number of Examples to Evaluate",
@ -195,7 +194,7 @@ def run_evaluation_3():
if st.button("Run Evaluation"):
progress_text = "Running evaluation..."
progress_bar = st.progress(0, text=progress_text)
rows = rows.rows
rows = rows.data
if num_rows < total_rows:
rows = rows[:num_rows]

137
llama_stack/distribution/utils/exec.py
View file

@ -4,13 +4,10 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import errno
import logging
import os
import select
import signal
import subprocess
import sys
from termcolor import cprint
@ -88,13 +85,6 @@ def formulate_run_args(image_type, image_name, config, template_name) -> list:
return run_args
def run_with_pty(command):
if sys.platform.startswith("win"):
return _run_with_pty_win(command)
else:
return _run_with_pty_unix(command)
def in_notebook():
try:
from IPython import get_ipython
@ -108,19 +98,19 @@ def in_notebook():
return True
# run a command in a pseudo-terminal, with interrupt handling,
# useful when you want to run interactive things
def _run_with_pty_unix(command):
import pty
import termios
def run_command(command: list[str]) -> int:
"""
Run a command with interrupt handling and output capture.
Uses subprocess.run with direct stream piping for better performance.
master, slave = pty.openpty()
Args:
command (list): The command to run.
old_settings = termios.tcgetattr(sys.stdin)
Returns:
int: The return code of the command.
"""
original_sigint = signal.getsignal(signal.SIGINT)
ctrl_c_pressed = False
process = None
def sigint_handler(signum, frame):
nonlocal ctrl_c_pressed
@ -131,106 +121,19 @@ def _run_with_pty_unix(command):
# Set up the signal handler
signal.signal(signal.SIGINT, sigint_handler)
new_settings = termios.tcgetattr(sys.stdin)
new_settings[3] = new_settings[3] & ~termios.ECHO # Disable echo
new_settings[3] = new_settings[3] & ~termios.ICANON # Disable canonical mode
termios.tcsetattr(sys.stdin, termios.TCSADRAIN, new_settings)
process = subprocess.Popen(
# Run the command with stdout/stderr piped directly to system streams
result = subprocess.run(
command,
stdin=slave,
stdout=slave,
stderr=slave,
universal_newlines=True,
preexec_fn=os.setsid,
text=True,
check=False,
)
# Close the slave file descriptor as it's now owned by the subprocess
os.close(slave)
def handle_io():
while not ctrl_c_pressed:
try:
rlist, _, _ = select.select([sys.stdin, master], [], [], 0.1)
if sys.stdin in rlist:
data = os.read(sys.stdin.fileno(), 1024)
if not data:
break
os.write(master, data)
if master in rlist:
data = os.read(master, 1024)
if not data:
break
sys.stdout.buffer.write(data)
sys.stdout.flush()
except KeyboardInterrupt:
# This will be raised when Ctrl+C is pressed
break
if process.poll() is not None:
break
handle_io()
except (EOFError, KeyboardInterrupt):
pass
except OSError as e:
if e.errno != errno.EIO:
raise
finally:
# Clean up
termios.tcsetattr(sys.stdin, termios.TCSADRAIN, old_settings)
signal.signal(signal.SIGINT, original_sigint)
os.close(master)
if process and process.poll() is None:
process.terminate()
process.wait()
return process.returncode
# run a command in a pseudo-terminal in windows, with interrupt handling,
def _run_with_pty_win(command):
"""
Runs a command with interactive support using subprocess directly.
"""
try:
# For shell scripts on Windows, use appropriate shell
if isinstance(command, (list, tuple)):
if command[0].endswith(".sh"):
if os.path.exists("/usr/bin/bash"): # WSL
command = ["bash"] + command
else:
# Use cmd.exe with bash while preserving all arguments
command = ["cmd.exe", "/c", "bash"] + command
process = subprocess.Popen(
command,
shell=True,
universal_newlines=True,
)
process.wait()
return result.returncode
except subprocess.SubprocessError as e:
log.error(f"Subprocess error: {e}")
return 1
except Exception as e:
print(f"Error: {str(e)}")
log.exception(f"Unexpected error: {e}")
return 1
finally:
if process and process.poll() is None:
process.terminate()
process.wait()
return process.returncode
def run_command(command):
try:
result = subprocess.run(command, capture_output=True, text=True, check=True)
print("Script Output\n", result.stdout)
return result.returncode
except subprocess.CalledProcessError as e:
print("Error running script:", e)
print("Error output:", e.stderr)
return e.returncode
# Restore the original signal handler
signal.signal(signal.SIGINT, original_sigint)

236
llama_stack/providers/inline/agents/meta_reference/agent_instance.py
View file

@ -614,118 +614,133 @@ class ChatAgent(ShieldRunnerMixin):
logger.debug(f"completion message with EOM (iter: {n_iter}): {str(message)}")
input_messages = input_messages + [message]
else:
logger.debug(f"completion message (iter: {n_iter}) from the model: {str(message)}")
# 1. Start the tool execution step and progress
step_id = str(uuid.uuid4())
yield AgentTurnResponseStreamChunk(
event=AgentTurnResponseEvent(
payload=AgentTurnResponseStepStartPayload(
step_type=StepType.tool_execution.value,
step_id=step_id,
)
)
)
tool_call = message.tool_calls[0]
yield AgentTurnResponseStreamChunk(
event=AgentTurnResponseEvent(
payload=AgentTurnResponseStepProgressPayload(
step_type=StepType.tool_execution.value,
step_id=step_id,
tool_call=tool_call,
delta=ToolCallDelta(
parse_status=ToolCallParseStatus.in_progress,
tool_call=tool_call,
),
)
)
)
input_messages = input_messages + [message]
# If tool is a client tool, yield CompletionMessage and return
if tool_call.tool_name in client_tools:
# NOTE: mark end_of_message to indicate to client that it may
# call the tool and continue the conversation with the tool's response.
message.stop_reason = StopReason.end_of_message
# Process tool calls in the message
client_tool_calls = []
non_client_tool_calls = []
# Separate client and non-client tool calls
for tool_call in message.tool_calls:
if tool_call.tool_name in client_tools:
client_tool_calls.append(tool_call)
else:
non_client_tool_calls.append(tool_call)
# Process non-client tool calls first
for tool_call in non_client_tool_calls:
step_id = str(uuid.uuid4())
yield AgentTurnResponseStreamChunk(
event=AgentTurnResponseEvent(
payload=AgentTurnResponseStepStartPayload(
step_type=StepType.tool_execution.value,
step_id=step_id,
)
)
)
yield AgentTurnResponseStreamChunk(
event=AgentTurnResponseEvent(
payload=AgentTurnResponseStepProgressPayload(
step_type=StepType.tool_execution.value,
step_id=step_id,
delta=ToolCallDelta(
parse_status=ToolCallParseStatus.in_progress,
tool_call=tool_call,
),
)
)
)
# Execute the tool call
async with tracing.span(
"tool_execution",
{
"tool_name": tool_call.tool_name,
"input": message.model_dump_json(),
},
) as span:
tool_execution_start_time = datetime.now(timezone.utc).isoformat()
tool_result = await self.execute_tool_call_maybe(
session_id,
tool_call,
)
if tool_result.content is None:
raise ValueError(
f"Tool call result (id: {tool_call.call_id}, name: {tool_call.tool_name}) does not have any content"
)
result_message = ToolResponseMessage(
call_id=tool_call.call_id,
content=tool_result.content,
)
span.set_attribute("output", result_message.model_dump_json())
# Store tool execution step
tool_execution_step = ToolExecutionStep(
step_id=step_id,
turn_id=turn_id,
tool_calls=[tool_call],
tool_responses=[
ToolResponse(
call_id=tool_call.call_id,
tool_name=tool_call.tool_name,
content=tool_result.content,
metadata=tool_result.metadata,
)
],
started_at=tool_execution_start_time,
completed_at=datetime.now(timezone.utc).isoformat(),
)
# Yield the step completion event
yield AgentTurnResponseStreamChunk(
event=AgentTurnResponseEvent(
payload=AgentTurnResponseStepCompletePayload(
step_type=StepType.tool_execution.value,
step_id=step_id,
step_details=tool_execution_step,
)
)
)
# Add the result message to input_messages for the next iteration
input_messages.append(result_message)
# TODO: add tool-input touchpoint and a "start" event for this step also
# but that needs a lot more refactoring of Tool code potentially
if (type(result_message.content) is str) and (
out_attachment := _interpret_content_as_attachment(result_message.content)
):
# NOTE: when we push this message back to the model, the model may ignore the
# attached file path etc. since the model is trained to only provide a user message
# with the summary. We keep all generated attachments and then attach them to final message
output_attachments.append(out_attachment)
# If there are client tool calls, yield a message with only those tool calls
if client_tool_calls:
await self.storage.set_in_progress_tool_call_step(
session_id,
turn_id,
ToolExecutionStep(
step_id=step_id,
turn_id=turn_id,
tool_calls=[tool_call],
tool_calls=client_tool_calls,
tool_responses=[],
started_at=datetime.now(timezone.utc).isoformat(),
),
)
yield message
# Create a copy of the message with only client tool calls
client_message = message.model_copy(deep=True)
client_message.tool_calls = client_tool_calls
# NOTE: mark end_of_message to indicate to client that it may
# call the tool and continue the conversation with the tool's response.
client_message.stop_reason = StopReason.end_of_message
# Yield the message with client tool calls
yield client_message
return
# If tool is a builtin server tool, execute it
tool_name = tool_call.tool_name
if isinstance(tool_name, BuiltinTool):
tool_name = tool_name.value
async with tracing.span(
"tool_execution",
{
"tool_name": tool_name,
"input": message.model_dump_json(),
},
) as span:
tool_execution_start_time = datetime.now(timezone.utc).isoformat()
tool_call = message.tool_calls[0]
tool_result = await self.execute_tool_call_maybe(
session_id,
tool_call,
)
if tool_result.content is None:
raise ValueError(
f"Tool call result (id: {tool_call.call_id}, name: {tool_call.tool_name}) does not have any content"
)
result_messages = [
ToolResponseMessage(
call_id=tool_call.call_id,
content=tool_result.content,
)
]
assert len(result_messages) == 1, "Currently not supporting multiple messages"
result_message = result_messages[0]
span.set_attribute("output", result_message.model_dump_json())
yield AgentTurnResponseStreamChunk(
event=AgentTurnResponseEvent(
payload=AgentTurnResponseStepCompletePayload(
step_type=StepType.tool_execution.value,
step_id=step_id,
step_details=ToolExecutionStep(
step_id=step_id,
turn_id=turn_id,
tool_calls=[tool_call],
tool_responses=[
ToolResponse(
call_id=result_message.call_id,
tool_name=tool_call.tool_name,
content=result_message.content,
metadata=tool_result.metadata,
)
],
started_at=tool_execution_start_time,
completed_at=datetime.now(timezone.utc).isoformat(),
),
)
)
)
# TODO: add tool-input touchpoint and a "start" event for this step also
# but that needs a lot more refactoring of Tool code potentially
if (type(result_message.content) is str) and (
out_attachment := _interpret_content_as_attachment(result_message.content)
):
# NOTE: when we push this message back to the model, the model may ignore the
# attached file path etc. since the model is trained to only provide a user message
# with the summary. We keep all generated attachments and then attach them to final message
output_attachments.append(out_attachment)
input_messages = input_messages + [message, result_message]
async def _initialize_tools(
self,
toolgroups_for_turn: Optional[List[AgentToolGroup]] = None,
@ -891,16 +906,14 @@ class ChatAgent(ShieldRunnerMixin):
if memory_tool and code_interpreter_tool:
# if both memory and code_interpreter are available, we download the URLs
# and attach the data to the last message.
msg = await attachment_message(self.tempdir, url_items)
input_messages.append(msg)
await attachment_message(self.tempdir, url_items, input_messages[-1])
# Since memory is present, add all the data to the memory bank
await self.add_to_session_vector_db(session_id, documents)
elif code_interpreter_tool:
# if only code_interpreter is available, we download the URLs to a tempdir
# and attach the path to them as a message to inference with the
# assumption that the model invokes the code_interpreter tool with the path
msg = await attachment_message(self.tempdir, url_items)
input_messages.append(msg)
await attachment_message(self.tempdir, url_items, input_messages[-1])
elif memory_tool:
# if only memory is available, we load the data from the URLs and content items to the memory bank
await self.add_to_session_vector_db(session_id, documents)
@ -967,8 +980,8 @@ async def load_data_from_urls(urls: List[URL]) -> List[str]:
return data
async def attachment_message(tempdir: str, urls: List[URL]) -> ToolResponseMessage:
content = []
async def attachment_message(tempdir: str, urls: List[URL], message: UserMessage) -> None:
contents = []
for url in urls:
uri = url.uri
@ -988,16 +1001,19 @@ async def attachment_message(tempdir: str, urls: List[URL]) -> ToolResponseMessa
else:
raise ValueError(f"Unsupported URL {url}")
content.append(
contents.append(
TextContentItem(
text=f'# User provided a file accessible to you at "{filepath}"\nYou can use code_interpreter to load and inspect it.'
)
)
return ToolResponseMessage(
call_id="",
content=content,
)
if isinstance(message.content, list):
message.content.extend(contents)
else:
if isinstance(message.content, str):
message.content = [TextContentItem(text=message.content)] + contents
else:
message.content = [message.content] + contents
def _interpret_content_as_attachment(

139
llama_stack/providers/inline/datasetio/localfs/datasetio.py
View file

@ -3,20 +3,14 @@
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import base64
import os
from abc import ABC, abstractmethod
from dataclasses import dataclass
from typing import Any, Dict, List, Optional
from urllib.parse import urlparse
import pandas
from llama_stack.apis.common.content_types import URL
from llama_stack.apis.datasetio import DatasetIO, PaginatedRowsResult
from llama_stack.apis.datasetio import DatasetIO, IterrowsResponse
from llama_stack.apis.datasets import Dataset
from llama_stack.providers.datatypes import DatasetsProtocolPrivate
from llama_stack.providers.utils.datasetio.url_utils import get_dataframe_from_url
from llama_stack.providers.utils.datasetio.url_utils import get_dataframe_from_uri
from llama_stack.providers.utils.kvstore import kvstore_impl
from .config import LocalFSDatasetIOConfig
@ -24,30 +18,7 @@ from .config import LocalFSDatasetIOConfig
DATASETS_PREFIX = "localfs_datasets:"
class BaseDataset(ABC):
def __init__(self, *args, **kwargs) -> None:
super().__init__(*args, **kwargs)
@abstractmethod
def __len__(self) -> int:
raise NotImplementedError()
@abstractmethod
def __getitem__(self, idx):
raise NotImplementedError()
@abstractmethod
def load(self):
raise NotImplementedError()
@dataclass
class DatasetInfo:
dataset_def: Dataset
dataset_impl: BaseDataset
class PandasDataframeDataset(BaseDataset):
class PandasDataframeDataset:
def __init__(self, dataset_def: Dataset, *args, **kwargs) -> None:
super().__init__(*args, **kwargs)
self.dataset_def = dataset_def
@ -64,23 +35,19 @@ class PandasDataframeDataset(BaseDataset):
else:
return self.df.iloc[idx].to_dict()
def _validate_dataset_schema(self, df) -> pandas.DataFrame:
# note that we will drop any columns in dataset that are not in the schema
df = df[self.dataset_def.dataset_schema.keys()]
# check all columns in dataset schema are present
assert len(df.columns) == len(self.dataset_def.dataset_schema)
# TODO: type checking against column types in dataset schema
return df
def load(self) -> None:
if self.df is not None:
return
df = get_dataframe_from_url(self.dataset_def.url)
if df is None:
raise ValueError(f"Failed to load dataset from {self.dataset_def.url}")
if self.dataset_def.source.type == "uri":
self.df = get_dataframe_from_uri(self.dataset_def.source.uri)
elif self.dataset_def.source.type == "rows":
self.df = pandas.DataFrame(self.dataset_def.source.rows)
else:
raise ValueError(f"Unsupported dataset source type: {self.dataset_def.source.type}")
self.df = self._validate_dataset_schema(df)
if self.df is None:
raise ValueError(f"Failed to load dataset from {self.dataset_def.url}")
class LocalFSDatasetIOImpl(DatasetIO, DatasetsProtocolPrivate):
@ -99,95 +66,55 @@ class LocalFSDatasetIOImpl(DatasetIO, DatasetsProtocolPrivate):
for dataset in stored_datasets:
dataset = Dataset.model_validate_json(dataset)
dataset_impl = PandasDataframeDataset(dataset)
self.dataset_infos[dataset.identifier] = DatasetInfo(
dataset_def=dataset,
dataset_impl=dataset_impl,
)
self.dataset_infos[dataset.identifier] = dataset
async def shutdown(self) -> None: ...
async def register_dataset(
self,
dataset: Dataset,
dataset_def: Dataset,
) -> None:
# Store in kvstore
key = f"{DATASETS_PREFIX}{dataset.identifier}"
key = f"{DATASETS_PREFIX}{dataset_def.identifier}"
await self.kvstore.set(
key=key,
value=dataset.json(),
)
dataset_impl = PandasDataframeDataset(dataset)
self.dataset_infos[dataset.identifier] = DatasetInfo(
dataset_def=dataset,
dataset_impl=dataset_impl,
value=dataset_def.model_dump_json(),
)
self.dataset_infos[dataset_def.identifier] = dataset_def
async def unregister_dataset(self, dataset_id: str) -> None:
key = f"{DATASETS_PREFIX}{dataset_id}"
await self.kvstore.delete(key=key)
del self.dataset_infos[dataset_id]
async def get_rows_paginated(
async def iterrows(
self,
dataset_id: str,
rows_in_page: int,
page_token: Optional[str] = None,
filter_condition: Optional[str] = None,
) -> PaginatedRowsResult:
dataset_info = self.dataset_infos.get(dataset_id)
dataset_info.dataset_impl.load()
start_index: Optional[int] = None,
limit: Optional[int] = None,
) -> IterrowsResponse:
dataset_def = self.dataset_infos[dataset_id]
dataset_impl = PandasDataframeDataset(dataset_def)
dataset_impl.load()
if page_token and not page_token.isnumeric():
raise ValueError("Invalid page_token")
start_index = start_index or 0
if page_token is None or len(page_token) == 0:
next_page_token = 0
if limit is None or limit == -1:
end = len(dataset_impl)
else:
next_page_token = int(page_token)
end = min(start_index + limit, len(dataset_impl))
start = next_page_token
if rows_in_page == -1:
end = len(dataset_info.dataset_impl)
else:
end = min(start + rows_in_page, len(dataset_info.dataset_impl))
rows = dataset_impl[start_index:end]
rows = dataset_info.dataset_impl[start:end]
return PaginatedRowsResult(
rows=rows,
total_count=len(rows),
next_page_token=str(end),
return IterrowsResponse(
data=rows,
next_start_index=end if end < len(dataset_impl) else None,
)
async def append_rows(self, dataset_id: str, rows: List[Dict[str, Any]]) -> None:
dataset_info = self.dataset_infos.get(dataset_id)
if dataset_info is None:
raise ValueError(f"Dataset with id {dataset_id} not found")
dataset_impl = dataset_info.dataset_impl
dataset_def = self.dataset_infos[dataset_id]
dataset_impl = PandasDataframeDataset(dataset_def)
dataset_impl.load()
new_rows_df = pandas.DataFrame(rows)
new_rows_df = dataset_impl._validate_dataset_schema(new_rows_df)
dataset_impl.df = pandas.concat([dataset_impl.df, new_rows_df], ignore_index=True)
url = str(dataset_info.dataset_def.url.uri)
parsed_url = urlparse(url)
if parsed_url.scheme == "file" or not parsed_url.scheme:
file_path = parsed_url.path
os.makedirs(os.path.dirname(file_path), exist_ok=True)
dataset_impl.df.to_csv(file_path, index=False)
elif parsed_url.scheme == "data":
# For data URLs, we need to update the base64-encoded content
if not parsed_url.path.startswith("text/csv;base64,"):
raise ValueError("Data URL must be a base64-encoded CSV")
csv_buffer = dataset_impl.df.to_csv(index=False)
base64_content = base64.b64encode(csv_buffer.encode("utf-8")).decode("utf-8")
dataset_info.dataset_def.url = URL(uri=f"data:text/csv;base64,{base64_content}")
else:
raise ValueError(
f"Unsupported URL scheme: {parsed_url.scheme}. Only file:// and data: URLs are supported for writing."
)

19
llama_stack/providers/inline/eval/meta_reference/eval.py
View file

@ -14,16 +14,11 @@ from llama_stack.apis.datasetio import DatasetIO
from llama_stack.apis.datasets import Datasets
from llama_stack.apis.inference import Inference, SystemMessage, UserMessage
from llama_stack.apis.scoring import Scoring
from llama_stack.distribution.datatypes import Api
from llama_stack.providers.datatypes import BenchmarksProtocolPrivate
from llama_stack.providers.inline.agents.meta_reference.agent_instance import (
MEMORY_QUERY_TOOL,
)
from llama_stack.providers.utils.common.data_schema_validator import (
ColumnName,
get_valid_schemas,
validate_dataset_schema,
)
from llama_stack.providers.utils.common.data_schema_validator import ColumnName
from llama_stack.providers.utils.kvstore import kvstore_impl
from .....apis.common.job_types import Job
@ -88,15 +83,17 @@ class MetaReferenceEvalImpl(
task_def = self.benchmarks[benchmark_id]
dataset_id = task_def.dataset_id
scoring_functions = task_def.scoring_functions
dataset_def = await self.datasets_api.get_dataset(dataset_id=dataset_id)
validate_dataset_schema(dataset_def.dataset_schema, get_valid_schemas(Api.eval.value))
all_rows = await self.datasetio_api.get_rows_paginated(
# TODO (xiyan): validate dataset schema
# dataset_def = await self.datasets_api.get_dataset(dataset_id=dataset_id)
all_rows = await self.datasetio_api.iterrows(
dataset_id=dataset_id,
rows_in_page=(-1 if benchmark_config.num_examples is None else benchmark_config.num_examples),
limit=(-1 if benchmark_config.num_examples is None else benchmark_config.num_examples),
)
res = await self.evaluate_rows(
benchmark_id=benchmark_id,
input_rows=all_rows.rows,
input_rows=all_rows.data,
scoring_functions=scoring_functions,
benchmark_config=benchmark_config,
)

6
llama_stack/providers/inline/post_training/torchtune/recipes/lora_finetuning_single_device.py
View file

@ -328,13 +328,13 @@ class LoraFinetuningSingleDevice:
batch_size: int,
) -> Tuple[DistributedSampler, DataLoader]:
async def fetch_rows(dataset_id: str):
return await self.datasetio_api.get_rows_paginated(
return await self.datasetio_api.iterrows(
dataset_id=dataset_id,
rows_in_page=-1,
limit=-1,
)
all_rows = await fetch_rows(dataset_id)
rows = all_rows.rows
rows = all_rows.data
await validate_input_dataset_schema(
datasets_api=self.datasets_api,

7
llama_stack/providers/inline/safety/llama_guard/llama_guard.py
View file

@ -227,13 +227,6 @@ class LlamaGuardShield:
if len(messages) >= 2 and (messages[0].role == Role.user.value and messages[1].role == Role.user.value):
messages = messages[1:]
for i in range(1, len(messages)):
if messages[i].role == messages[i - 1].role:
for i, m in enumerate(messages):
print(f"{i}: {m.role}: {m.content}")
raise ValueError(
f"Messages must alternate between user and assistant. Message {i} has the same role as message {i - 1}"
)
return messages
async def run(self, messages: List[Message]) -> RunShieldResponse:

10
llama_stack/providers/inline/scoring/basic/scoring.py
View file

@ -24,7 +24,9 @@ from llama_stack.providers.utils.common.data_schema_validator import (
from .config import BasicScoringConfig
from .scoring_fn.bfcl_scoring_fn import BFCLScoringFn
from .scoring_fn.equality_scoring_fn import EqualityScoringFn
from .scoring_fn.regex_parser_math_response_scoring_fn import RegexParserMathResponseScoringFn
from .scoring_fn.regex_parser_math_response_scoring_fn import (
RegexParserMathResponseScoringFn,
)
from .scoring_fn.regex_parser_scoring_fn import RegexParserScoringFn
from .scoring_fn.subset_of_scoring_fn import SubsetOfScoringFn
@ -82,12 +84,12 @@ class BasicScoringImpl(
dataset_def = await self.datasets_api.get_dataset(dataset_id=dataset_id)
validate_dataset_schema(dataset_def.dataset_schema, get_valid_schemas(Api.scoring.value))
all_rows = await self.datasetio_api.get_rows_paginated(
all_rows = await self.datasetio_api.iterrows(
dataset_id=dataset_id,
rows_in_page=-1,
limit=-1,
)
res = await self.score(
input_rows=all_rows.rows,
input_rows=all_rows.data,
scoring_functions=scoring_functions,
)
if save_results_dataset:

6
llama_stack/providers/inline/scoring/braintrust/braintrust.py
View file

@ -167,11 +167,11 @@ class BraintrustScoringImpl(
dataset_def = await self.datasets_api.get_dataset(dataset_id=dataset_id)
validate_dataset_schema(dataset_def.dataset_schema, get_valid_schemas(Api.scoring.value))
all_rows = await self.datasetio_api.get_rows_paginated(
all_rows = await self.datasetio_api.iterrows(
dataset_id=dataset_id,
rows_in_page=-1,
limit=-1,
)
res = await self.score(input_rows=all_rows.rows, scoring_functions=scoring_functions)
res = await self.score(input_rows=all_rows.data, scoring_functions=scoring_functions)
if save_results_dataset:
# TODO: persist and register dataset on to server for reading
# self.datasets_api.register_dataset()

6
llama_stack/providers/inline/scoring/llm_as_judge/scoring.py
View file

@ -72,12 +72,12 @@ class LlmAsJudgeScoringImpl(
dataset_def = await self.datasets_api.get_dataset(dataset_id=dataset_id)
validate_dataset_schema(dataset_def.dataset_schema, get_valid_schemas(Api.scoring.value))
all_rows = await self.datasetio_api.get_rows_paginated(
all_rows = await self.datasetio_api.iterrows(
dataset_id=dataset_id,
rows_in_page=-1,
limit=-1,
)
res = await self.score(
input_rows=all_rows.rows,
input_rows=all_rows.data,
scoring_functions=scoring_functions,
)
if save_results_dataset:

9
llama_stack/providers/registry/safety.py
View file

@ -55,4 +55,13 @@ def available_providers() -> List[ProviderSpec]:
config_class="llama_stack.providers.remote.safety.bedrock.BedrockSafetyConfig",
),
),
remote_provider_spec(
api=Api.safety,
adapter=AdapterSpec(
adapter_type="nvidia",
pip_packages=["requests"],
module="llama_stack.providers.remote.safety.nvidia",
config_class="llama_stack.providers.remote.safety.nvidia.NVIDIASafetyConfig",
),
),
]

65
llama_stack/providers/remote/datasetio/huggingface/huggingface.py
View file

@ -4,13 +4,13 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from typing import Any, Dict, List, Optional
from urllib.parse import parse_qs, urlparse
import datasets as hf_datasets
from llama_stack.apis.datasetio import DatasetIO, PaginatedRowsResult
from llama_stack.apis.datasetio import DatasetIO, IterrowsResponse
from llama_stack.apis.datasets import Dataset
from llama_stack.providers.datatypes import DatasetsProtocolPrivate
from llama_stack.providers.utils.datasetio.url_utils import get_dataframe_from_url
from llama_stack.providers.utils.kvstore import kvstore_impl
from .config import HuggingfaceDatasetIOConfig
@ -18,22 +18,14 @@ from .config import HuggingfaceDatasetIOConfig
DATASETS_PREFIX = "datasets:"
def load_hf_dataset(dataset_def: Dataset):
if dataset_def.metadata.get("path", None):
dataset = hf_datasets.load_dataset(**dataset_def.metadata)
else:
df = get_dataframe_from_url(dataset_def.url)
def parse_hf_params(dataset_def: Dataset):
uri = dataset_def.source.uri
parsed_uri = urlparse(uri)
params = parse_qs(parsed_uri.query)
params = {k: v[0] for k, v in params.items()}
path = parsed_uri.path.lstrip("/")
if df is None:
raise ValueError(f"Failed to load dataset from {dataset_def.url}")
dataset = hf_datasets.Dataset.from_pandas(df)
# drop columns not specified by schema
if dataset_def.dataset_schema:
dataset = dataset.select_columns(list(dataset_def.dataset_schema.keys()))
return dataset
return path, params
class HuggingfaceDatasetIOImpl(DatasetIO, DatasetsProtocolPrivate):
@ -64,7 +56,7 @@ class HuggingfaceDatasetIOImpl(DatasetIO, DatasetsProtocolPrivate):
key = f"{DATASETS_PREFIX}{dataset_def.identifier}"
await self.kvstore.set(
key=key,
value=dataset_def.json(),
value=dataset_def.model_dump_json(),
)
self.dataset_infos[dataset_def.identifier] = dataset_def
@ -73,41 +65,34 @@ class HuggingfaceDatasetIOImpl(DatasetIO, DatasetsProtocolPrivate):
await self.kvstore.delete(key=key)
del self.dataset_infos[dataset_id]
async def get_rows_paginated(
async def iterrows(
self,
dataset_id: str,
rows_in_page: int,
page_token: Optional[str] = None,
filter_condition: Optional[str] = None,
) -> PaginatedRowsResult:
start_index: Optional[int] = None,
limit: Optional[int] = None,
) -> IterrowsResponse:
dataset_def = self.dataset_infos[dataset_id]
loaded_dataset = load_hf_dataset(dataset_def)
path, params = parse_hf_params(dataset_def)
loaded_dataset = hf_datasets.load_dataset(path, **params)
if page_token and not page_token.isnumeric():
raise ValueError("Invalid page_token")
start_index = start_index or 0
if page_token is None or len(page_token) == 0:
next_page_token = 0
else:
next_page_token = int(page_token)
start = next_page_token
if rows_in_page == -1:
if limit is None or limit == -1:
end = len(loaded_dataset)
else:
end = min(start + rows_in_page, len(loaded_dataset))
end = min(start_index + limit, len(loaded_dataset))
rows = [loaded_dataset[i] for i in range(start, end)]
rows = [loaded_dataset[i] for i in range(start_index, end)]
return PaginatedRowsResult(
rows=rows,
total_count=len(rows),
next_page_token=str(end),
return IterrowsResponse(
data=rows,
next_start_index=end if end < len(loaded_dataset) else None,
)
async def append_rows(self, dataset_id: str, rows: List[Dict[str, Any]]) -> None:
dataset_def = self.dataset_infos[dataset_id]
loaded_dataset = load_hf_dataset(dataset_def)
path, params = parse_hf_params(dataset_def)
loaded_dataset = hf_datasets.load_dataset(path, **params)
# Convert rows to HF Dataset format
new_dataset = hf_datasets.Dataset.from_list(rows)

15
llama_stack/providers/remote/inference/passthrough/passthrough.py
View file

@ -12,6 +12,7 @@ from llama_stack.apis.common.content_types import InterleavedContent
from llama_stack.apis.inference import (
ChatCompletionResponse,
ChatCompletionResponseStreamChunk,
CompletionMessage,
EmbeddingsResponse,
EmbeddingTaskType,
Inference,
@ -160,12 +161,14 @@ class PassthroughInferenceAdapter(Inference):
client = self._get_client()
response = await client.inference.chat_completion(**json_params)
response = response.to_dict()
# temporary hack to remove the metrics from the response
response["metrics"] = []
return convert_to_pydantic(ChatCompletionResponse, response)
return ChatCompletionResponse(
completion_message=CompletionMessage(
content=response.completion_message.content.text,
stop_reason=response.completion_message.stop_reason,
tool_calls=response.completion_message.tool_calls,
),
logprobs=response.logprobs,
)
async def _stream_chat_completion(self, json_params: Dict[str, Any]) -> AsyncGenerator:
client = self._get_client()

5
llama_stack/providers/remote/inference/vllm/config.py
View file

@ -25,6 +25,10 @@ class VLLMInferenceAdapterConfig(BaseModel):
default="fake",
description="The API token",
)
tls_verify: bool = Field(
default=True,
description="Whether to verify TLS certificates",
)
@classmethod
def sample_run_config(
@ -36,4 +40,5 @@ class VLLMInferenceAdapterConfig(BaseModel):
"url": url,
"max_tokens": "${env.VLLM_MAX_TOKENS:4096}",
"api_token": "${env.VLLM_API_TOKEN:fake}",
"tls_verify": "${env.VLLM_TLS_VERIFY:true}",
}

7
llama_stack/providers/remote/inference/vllm/vllm.py
View file

@ -7,6 +7,7 @@ import json
import logging
from typing import AsyncGenerator, List, Optional, Union
import httpx
from openai import AsyncOpenAI
from openai.types.chat.chat_completion_chunk import (
ChatCompletionChunk as OpenAIChatCompletionChunk,
@ -229,7 +230,11 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
async def initialize(self) -> None:
log.info(f"Initializing VLLM client with base_url={self.config.url}")
self.client = AsyncOpenAI(base_url=self.config.url, api_key=self.config.api_token)
self.client = AsyncOpenAI(
base_url=self.config.url,
api_key=self.config.api_token,
http_client=None if self.config.tls_verify else httpx.AsyncClient(verify=False),
)
async def shutdown(self) -> None:
pass

18
llama_stack/providers/remote/safety/nvidia/__init__.py Normal file
View file

@ -0,0 +1,18 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from typing import Any
from .config import NVIDIASafetyConfig
async def get_adapter_impl(config: NVIDIASafetyConfig, _deps) -> Any:
from .nvidia import NVIDIASafetyAdapter
impl = NVIDIASafetyAdapter(config)
await impl.initialize()
return impl

37
llama_stack/providers/remote/safety/nvidia/config.py Normal file
View file

@ -0,0 +1,37 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import os
from typing import Any, Dict, Optional
from pydantic import BaseModel, Field
from llama_stack.schema_utils import json_schema_type
@json_schema_type
class NVIDIASafetyConfig(BaseModel):
"""
Configuration for the NVIDIA Guardrail microservice endpoint.
Attributes:
guardrails_service_url (str): A base url for accessing the NVIDIA guardrail endpoint, e.g. http://0.0.0.0:7331
config_id (str): The ID of the guardrails configuration to use from the configuration store
(https://developer.nvidia.com/docs/nemo-microservices/guardrails/source/guides/configuration-store-guide.html)
"""
guardrails_service_url: str = Field(
default_factory=lambda: os.getenv("GUARDRAILS_SERVICE_URL", "http://0.0.0.0:7331"),
description="The url for accessing the guardrails service",
)
config_id: Optional[str] = Field(default="self-check", description="Config ID to use from the config store")
@classmethod
def sample_run_config(cls, **kwargs) -> Dict[str, Any]:
return {
"guardrails_service_url": "${env.GUARDRAILS_SERVICE_URL:http://localhost:7331}",
"config_id": "self-check",
}

154
llama_stack/providers/remote/safety/nvidia/nvidia.py Normal file
View file

@ -0,0 +1,154 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import logging
from typing import Any, List, Optional
import requests
from llama_stack.apis.inference import Message
from llama_stack.apis.safety import RunShieldResponse, Safety, SafetyViolation, ViolationLevel
from llama_stack.apis.shields import Shield
from llama_stack.distribution.library_client import convert_pydantic_to_json_value
from llama_stack.providers.datatypes import ShieldsProtocolPrivate
from .config import NVIDIASafetyConfig
logger = logging.getLogger(__name__)
class NVIDIASafetyAdapter(Safety, ShieldsProtocolPrivate):
def __init__(self, config: NVIDIASafetyConfig) -> None:
"""
Initialize the NVIDIASafetyAdapter with a given safety configuration.
Args:
config (NVIDIASafetyConfig): The configuration containing the guardrails service URL and config ID.
"""
print(f"Initializing NVIDIASafetyAdapter({config.guardrails_service_url})...")
self.config = config
async def initialize(self) -> None:
pass
async def shutdown(self) -> None:
pass
async def register_shield(self, shield: Shield) -> None:
if not shield.provider_resource_id:
raise ValueError("Shield model not provided.")
async def run_shield(
self, shield_id: str, messages: List[Message], params: Optional[dict[str, Any]] = None
) -> RunShieldResponse:
"""
Run a safety shield check against the provided messages.
Args:
shield_id (str): The unique identifier for the shield to be used.
messages (List[Message]): A list of Message objects representing the conversation history.
params (Optional[dict[str, Any]]): Additional parameters for the shield check.
Returns:
RunShieldResponse: The response containing safety violation details if any.
Raises:
ValueError: If the shield with the provided shield_id is not found.
"""
shield = await self.shield_store.get_shield(shield_id)
if not shield:
raise ValueError(f"Shield {shield_id} not found")
self.shield = NeMoGuardrails(self.config, shield.shield_id)
return await self.shield.run(messages)
class NeMoGuardrails:
"""
A class that encapsulates NVIDIA's guardrails safety logic.
Sends messages to the guardrails service and interprets the response to determine
if a safety violation has occurred.
"""
def __init__(
self,
config: NVIDIASafetyConfig,
model: str,
threshold: float = 0.9,
temperature: float = 1.0,
):
"""
Initialize a NeMoGuardrails instance with the provided parameters.
Args:
config (NVIDIASafetyConfig): The safety configuration containing the config ID and guardrails URL.
model (str): The identifier or name of the model to be used for safety checks.
threshold (float, optional): The threshold for flagging violations. Defaults to 0.9.
temperature (float, optional): The temperature setting for the underlying model. Must be greater than 0. Defaults to 1.0.
Raises:
ValueError: If temperature is less than or equal to 0.
AssertionError: If config_id is not provided in the configuration.
"""
self.config_id = config.config_id
self.model = model
assert self.config_id is not None, "Must provide config id"
if temperature <= 0:
raise ValueError("Temperature must be greater than 0")
self.temperature = temperature
self.threshold = threshold
self.guardrails_service_url = config.guardrails_service_url
async def run(self, messages: List[Message]) -> RunShieldResponse:
"""
Queries the /v1/guardrails/checks endpoint of the NeMo guardrails deployed API.
Args:
messages (List[Message]): A list of Message objects to be checked for safety violations.
Returns:
RunShieldResponse: If the response indicates a violation ("blocked" status), returns a
RunShieldResponse with a SafetyViolation; otherwise, returns a RunShieldResponse with violation set to None.
Raises:
requests.HTTPError: If the POST request fails.
"""
headers = {
"Accept": "application/json",
}
request_data = {
"model": self.model,
"messages": convert_pydantic_to_json_value(messages),
"temperature": self.temperature,
"top_p": 1,
"frequency_penalty": 0,
"presence_penalty": 0,
"max_tokens": 160,
"stream": False,
"guardrails": {
"config_id": self.config_id,
},
}
response = requests.post(
url=f"{self.guardrails_service_url}/v1/guardrail/checks", headers=headers, json=request_data
)
response.raise_for_status()
if "Content-Type" in response.headers and response.headers["Content-Type"].startswith("application/json"):
response_json = response.json()
if response_json["status"] == "blocked":
user_message = "Sorry I cannot do this."
metadata = response_json["rails_status"]
return RunShieldResponse(
violation=SafetyViolation(
user_message=user_message,
violation_level=ViolationLevel.ERROR,
metadata=metadata,
)
)
return RunShieldResponse(violation=None)

17
llama_stack/providers/utils/datasetio/url_utils.py
View file

@ -10,18 +10,17 @@ from urllib.parse import unquote
import pandas
from llama_stack.apis.common.content_types import URL
from llama_stack.providers.utils.memory.vector_store import parse_data_url
def get_dataframe_from_url(url: URL):
def get_dataframe_from_uri(uri: str):
df = None
if url.uri.endswith(".csv"):
df = pandas.read_csv(url.uri)
elif url.uri.endswith(".xlsx"):
df = pandas.read_excel(url.uri)
elif url.uri.startswith("data:"):
parts = parse_data_url(url.uri)
if uri.endswith(".csv"):
df = pandas.read_csv(uri)
elif uri.endswith(".xlsx"):
df = pandas.read_excel(uri)
elif uri.startswith("data:"):
parts = parse_data_url(uri)
data = parts["data"]
if parts["is_base64"]:
data = base64.b64decode(data)
@ -39,6 +38,6 @@ def get_dataframe_from_url(url: URL):
else:
df = pandas.read_excel(data_bytes)
else:
raise ValueError(f"Unsupported file type: {url}")
raise ValueError(f"Unsupported file type: {uri}")
return df

6
llama_stack/providers/utils/inference/litellm_openai_mixin.py
View file

@ -192,7 +192,11 @@ class LiteLLMOpenAIMixin(
if request.tools:
input_dict["tools"] = [convert_tooldef_to_openai_tool(tool) for tool in request.tools]
if request.tool_config.tool_choice:
input_dict["tool_choice"] = request.tool_config.tool_choice.value
input_dict["tool_choice"] = (
request.tool_config.tool_choice.value
if isinstance(request.tool_config.tool_choice, ToolChoice)
else request.tool_config.tool_choice
)
provider_data = self.get_request_provider_data()
key_field = self.provider_data_api_key_field

186
llama_stack/providers/utils/inference/openai_compat.py
View file

@ -527,27 +527,31 @@ async def convert_message_to_openai_dict_new(
async def _convert_message_content(
content: InterleavedContent,
) -> Union[str, Iterable[OpenAIChatCompletionContentPartParam]]:
async def impl():
async def impl(
content_: InterleavedContent,
) -> Union[str, OpenAIChatCompletionContentPartParam, List[OpenAIChatCompletionContentPartParam]]:
# Llama Stack and OpenAI spec match for str and text input
if isinstance(content, str):
return content
elif isinstance(content, TextContentItem):
if isinstance(content_, str):
return content_
elif isinstance(content_, TextContentItem):
return OpenAIChatCompletionContentPartTextParam(
type="text",
text=content.text,
text=content_.text,
)
elif isinstance(content, ImageContentItem):
elif isinstance(content_, ImageContentItem):
return OpenAIChatCompletionContentPartImageParam(
type="image_url",
image_url=OpenAIImageURL(url=await convert_image_content_to_url(content)),
image_url=OpenAIImageURL(url=await convert_image_content_to_url(content_)),
)
elif isinstance(content, list):
return [await _convert_message_content(item) for item in content]
elif isinstance(content_, list):
return [await impl(item) for item in content_]
else:
raise ValueError(f"Unsupported content type: {type(content)}")
raise ValueError(f"Unsupported content type: {type(content_)}")
ret = await impl()
if isinstance(ret, str) or isinstance(ret, list) or isinstance(ret, dict):
ret = await impl(content)
# OpenAI*Message expects a str or list
if isinstance(ret, str) or isinstance(ret, list):
return ret
else:
return [ret]
@ -566,13 +570,14 @@ async def convert_message_to_openai_dict_new(
OpenAIChatCompletionMessageToolCall(
id=tool.call_id,
function=OpenAIFunction(
name=tool.tool_name,
name=tool.tool_name if not isinstance(tool.tool_name, BuiltinTool) else tool.tool_name.value,
arguments=json.dumps(tool.arguments),
),
type="function",
)
for tool in message.tool_calls
],
]
or None,
)
elif isinstance(message, ToolResponseMessage):
out = OpenAIChatCompletionToolMessage(
@ -858,7 +863,8 @@ async def convert_openai_chat_completion_stream(
event_type = ChatCompletionResponseEventType.progress
stop_reason = None
toolcall_buffer = {}
tool_call_idx_to_buffer = {}
async for chunk in stream:
choice = chunk.choices[0] # assuming only one choice per chunk
@ -868,7 +874,6 @@ async def convert_openai_chat_completion_stream(
# if there's a tool call, emit an event for each tool in the list
# if tool call and content, emit both separately
if choice.delta.tool_calls:
# the call may have content and a tool call. ChatCompletionResponseEvent
# does not support both, so we emit the content first
@ -889,44 +894,53 @@ async def convert_openai_chat_completion_stream(
)
if not enable_incremental_tool_calls:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=next(event_type),
delta=ToolCallDelta(
tool_call=_convert_openai_tool_calls(choice.delta.tool_calls)[0],
parse_status=ToolCallParseStatus.succeeded,
),
logprobs=_convert_openai_logprobs(logprobs),
for tool_call in choice.delta.tool_calls:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=event_type,
delta=ToolCallDelta(
tool_call=_convert_openai_tool_calls([tool_call])[0],
parse_status=ToolCallParseStatus.succeeded,
),
logprobs=_convert_openai_logprobs(logprobs),
)
)
)
else:
tool_call = choice.delta.tool_calls[0]
if "name" not in toolcall_buffer:
toolcall_buffer["call_id"] = tool_call.id
toolcall_buffer["name"] = None
toolcall_buffer["content"] = ""
if "arguments" not in toolcall_buffer:
toolcall_buffer["arguments"] = ""
for tool_call in choice.delta.tool_calls:
idx = tool_call.index if hasattr(tool_call, "index") else 0
if tool_call.function.name:
toolcall_buffer["name"] = tool_call.function.name
delta = f"{toolcall_buffer['name']}("
if tool_call.function.arguments:
toolcall_buffer["arguments"] += tool_call.function.arguments
delta = toolcall_buffer["arguments"]
if idx not in tool_call_idx_to_buffer:
tool_call_idx_to_buffer[idx] = {
"call_id": tool_call.id,
"name": None,
"arguments": "",
"content": "",
}
toolcall_buffer["content"] += delta
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=event_type,
delta=ToolCallDelta(
tool_call=delta,
parse_status=ToolCallParseStatus.in_progress,
),
logprobs=_convert_openai_logprobs(logprobs),
)
)
else:
buffer = tool_call_idx_to_buffer[idx]
if tool_call.function:
if tool_call.function.name:
buffer["name"] = tool_call.function.name
delta = f"{buffer['name']}("
buffer["content"] += delta
if tool_call.function.arguments:
delta = tool_call.function.arguments
buffer["arguments"] += delta
buffer["content"] += delta
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=event_type,
delta=ToolCallDelta(
tool_call=delta,
parse_status=ToolCallParseStatus.in_progress,
),
logprobs=_convert_openai_logprobs(logprobs),
)
)
elif choice.delta.content:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=event_type,
@ -935,47 +949,51 @@ async def convert_openai_chat_completion_stream(
)
)
if toolcall_buffer:
delta = ")"
toolcall_buffer["content"] += delta
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=event_type,
delta=ToolCallDelta(
tool_call=delta,
parse_status=ToolCallParseStatus.in_progress,
),
logprobs=_convert_openai_logprobs(logprobs),
)
)
try:
arguments = json.loads(toolcall_buffer["arguments"])
tool_call = ToolCall(
call_id=toolcall_buffer["call_id"],
tool_name=toolcall_buffer["name"],
arguments=arguments,
)
for idx, buffer in tool_call_idx_to_buffer.items():
logger.debug(f"toolcall_buffer[{idx}]: {buffer}")
if buffer["name"]:
delta = ")"
buffer["content"] += delta
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
event_type=event_type,
delta=ToolCallDelta(
tool_call=tool_call,
parse_status=ToolCallParseStatus.succeeded,
tool_call=delta,
parse_status=ToolCallParseStatus.in_progress,
),
stop_reason=stop_reason,
logprobs=None,
)
)
except json.JSONDecodeError:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.complete,
delta=ToolCallDelta(
tool_call=toolcall_buffer["content"],
parse_status=ToolCallParseStatus.failed,
),
stop_reason=stop_reason,
try:
arguments = json.loads(buffer["arguments"])
tool_call = ToolCall(
call_id=buffer["call_id"],
tool_name=buffer["name"],
arguments=arguments,
)
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=ToolCallDelta(
tool_call=tool_call,
parse_status=ToolCallParseStatus.succeeded,
),
stop_reason=stop_reason,
)
)
except json.JSONDecodeError as e:
print(f"Failed to parse arguments: {e}")
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=ToolCallDelta(
tool_call=buffer["content"],
parse_status=ToolCallParseStatus.failed,
),
stop_reason=stop_reason,
)
)
)
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(

15
llama_stack/scripts/run_tests.sh
View file

@ -1,15 +0,0 @@
#!/bin/bash
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
THIS_DIR="$(cd "$(dirname "$(readlink -f "${BASH_SOURCE[0]}")")" && pwd)"
set -euo pipefail
set -x
stack_dir=$(dirname $(dirname $THIS_DIR))
PYTHONPATH=$stack_dir pytest -p no:warnings --asyncio-mode auto --tb=short

11
llama_stack/templates/nvidia/build.yaml
View file

@ -1,13 +1,13 @@
version: '2'
distribution_spec:
description: Use NVIDIA NIM for running LLM inference
description: Use NVIDIA NIM for running LLM inference and safety
providers:
inference:
- remote::nvidia
vector_io:
- inline::faiss
safety:
- inline::llama-guard
- remote::nvidia
agents:
- inline::meta-reference
telemetry:
@ -15,16 +15,9 @@ distribution_spec:
eval:
- inline::meta-reference
datasetio:
- remote::huggingface
- inline::localfs
scoring:
- inline::basic
- inline::llm-as-judge
- inline::braintrust
tool_runtime:
- remote::brave-search
- remote::tavily-search
- inline::code-interpreter
- inline::rag-runtime
- remote::model-context-protocol
image_type: conda

67
llama_stack/templates/nvidia/nvidia.py
View file

@ -6,9 +6,10 @@
from pathlib import Path
from llama_stack.distribution.datatypes import Provider, ToolGroupInput
from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput, ToolGroupInput
from llama_stack.providers.remote.inference.nvidia import NVIDIAConfig
from llama_stack.providers.remote.inference.nvidia.models import MODEL_ENTRIES
from llama_stack.providers.remote.safety.nvidia import NVIDIASafetyConfig
from llama_stack.templates.template import DistributionTemplate, RunConfigSettings, get_model_registry
@ -16,19 +17,13 @@ def get_distribution_template() -> DistributionTemplate:
providers = {
"inference": ["remote::nvidia"],
"vector_io": ["inline::faiss"],
"safety": ["inline::llama-guard"],
"safety": ["remote::nvidia"],
"agents": ["inline::meta-reference"],
"telemetry": ["inline::meta-reference"],
"eval": ["inline::meta-reference"],
"datasetio": ["remote::huggingface", "inline::localfs"],
"scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
"tool_runtime": [
"remote::brave-search",
"remote::tavily-search",
"inline::code-interpreter",
"inline::rag-runtime",
"remote::model-context-protocol",
],
"datasetio": ["inline::localfs"],
"scoring": ["inline::basic"],
"tool_runtime": ["inline::rag-runtime"],
}
inference_provider = Provider(
@ -36,30 +31,35 @@ def get_distribution_template() -> DistributionTemplate:
provider_type="remote::nvidia",
config=NVIDIAConfig.sample_run_config(),
)
safety_provider = Provider(
provider_id="nvidia",
provider_type="remote::nvidia",
config=NVIDIASafetyConfig.sample_run_config(),
)
inference_model = ModelInput(
model_id="${env.INFERENCE_MODEL}",
provider_id="nvidia",
)
safety_model = ModelInput(
model_id="${env.SAFETY_MODEL}",
provider_id="nvidia",
)
available_models = {
"nvidia": MODEL_ENTRIES,
}
default_tool_groups = [
ToolGroupInput(
toolgroup_id="builtin::websearch",
provider_id="tavily-search",
),
ToolGroupInput(
toolgroup_id="builtin::rag",
provider_id="rag-runtime",
),
ToolGroupInput(
toolgroup_id="builtin::code_interpreter",
provider_id="code-interpreter",
),
]
default_models = get_model_registry(available_models)
return DistributionTemplate(
name="nvidia",
distro_type="remote_hosted",
description="Use NVIDIA NIM for running LLM inference",
description="Use NVIDIA NIM for running LLM inference and safety",
container_image=None,
template_path=Path(__file__).parent / "doc_template.md",
providers=providers,
@ -72,15 +72,34 @@ def get_distribution_template() -> DistributionTemplate:
default_models=default_models,
default_tool_groups=default_tool_groups,
),
"run-with-safety.yaml": RunConfigSettings(
provider_overrides={
"inference": [
inference_provider,
safety_provider,
]
},
default_models=[inference_model, safety_model],
default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}", provider_id="nvidia")],
default_tool_groups=default_tool_groups,
),
},
run_config_env_vars={
"LLAMASTACK_PORT": (
"5001",
"Port for the Llama Stack distribution server",
),
"NVIDIA_API_KEY": (
"",
"NVIDIA API Key",
),
"GUARDRAILS_SERVICE_URL": (
"http://0.0.0.0:7331",
"URL for the NeMo Guardrails Service",
),
"INFERENCE_MODEL": (
"Llama3.1-8B-Instruct",
"Inference model",
),
"SAFETY_MODEL": (
"meta/llama-3.1-8b-instruct",
"Name of the model to use for safety",
),
},
)

101
llama_stack/templates/nvidia/run-with-safety.yaml Normal file
View file

@ -0,0 +1,101 @@
version: '2'
image_name: nvidia
apis:
- agents
- datasetio
- eval
- inference
- safety
- scoring
- telemetry
- tool_runtime
- vector_io
providers:
inference:
- provider_id: nvidia
provider_type: remote::nvidia
config:
url: ${env.NVIDIA_BASE_URL:https://integrate.api.nvidia.com}
api_key: ${env.NVIDIA_API_KEY:}
- provider_id: nvidia
provider_type: remote::nvidia
config:
guardrails_service_url: ${env.GUARDRAILS_SERVICE_URL:http://localhost:7331}
config_id: self-check
vector_io:
- provider_id: faiss
provider_type: inline::faiss
config:
kvstore:
type: sqlite
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/faiss_store.db
safety:
- provider_id: nvidia
provider_type: remote::nvidia
config:
guardrails_service_url: ${env.GUARDRAILS_SERVICE_URL:http://localhost:7331}
config_id: self-check
agents:
- provider_id: meta-reference
provider_type: inline::meta-reference
config:
persistence_store:
type: sqlite
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/agents_store.db
telemetry:
- provider_id: meta-reference
provider_type: inline::meta-reference
config:
service_name: ${env.OTEL_SERVICE_NAME:llama-stack}
sinks: ${env.TELEMETRY_SINKS:console,sqlite}
sqlite_db_path: ${env.SQLITE_DB_PATH:~/.llama/distributions/nvidia/trace_store.db}
eval:
- provider_id: meta-reference
provider_type: inline::meta-reference
config:
kvstore:
type: sqlite
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/meta_reference_eval.db
datasetio:
- provider_id: localfs
provider_type: inline::localfs
config:
kvstore:
type: sqlite
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/localfs_datasetio.db
scoring:
- provider_id: basic
provider_type: inline::basic
config: {}
tool_runtime:
- provider_id: rag-runtime
provider_type: inline::rag-runtime
config: {}
metadata_store:
type: sqlite
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/registry.db
models:
- metadata: {}
model_id: ${env.INFERENCE_MODEL}
provider_id: nvidia
model_type: llm
- metadata: {}
model_id: ${env.SAFETY_MODEL}
provider_id: nvidia
model_type: llm
shields:
- shield_id: ${env.SAFETY_MODEL}
provider_id: nvidia
vector_dbs: []
datasets: []
scoring_fns: []
benchmarks: []
tool_groups:
- toolgroup_id: builtin::rag
provider_id: rag-runtime
server:
port: 8321

41
llama_stack/templates/nvidia/run.yaml
View file

@ -26,10 +26,11 @@ providers:
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/faiss_store.db
safety:
- provider_id: llama-guard
provider_type: inline::llama-guard
- provider_id: nvidia
provider_type: remote::nvidia
config:
excluded_categories: []
guardrails_service_url: ${env.GUARDRAILS_SERVICE_URL:http://localhost:7331}
config_id: self-check
agents:
- provider_id: meta-reference
provider_type: inline::meta-reference
@ -54,13 +55,6 @@ providers:
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/meta_reference_eval.db
datasetio:
- provider_id: huggingface
provider_type: remote::huggingface
config:
kvstore:
type: sqlite
namespace: null
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/huggingface_datasetio.db
- provider_id: localfs
provider_type: inline::localfs
config:
@ -72,33 +66,10 @@ providers:
- provider_id: basic
provider_type: inline::basic
config: {}
- provider_id: llm-as-judge
provider_type: inline::llm-as-judge
config: {}
- provider_id: braintrust
provider_type: inline::braintrust
config:
openai_api_key: ${env.OPENAI_API_KEY:}
tool_runtime:
- provider_id: brave-search
provider_type: remote::brave-search
config:
api_key: ${env.BRAVE_SEARCH_API_KEY:}
max_results: 3
- provider_id: tavily-search
provider_type: remote::tavily-search
config:
api_key: ${env.TAVILY_SEARCH_API_KEY:}
max_results: 3
- provider_id: code-interpreter
provider_type: inline::code-interpreter
config: {}
- provider_id: rag-runtime
provider_type: inline::rag-runtime
config: {}
- provider_id: model-context-protocol
provider_type: remote::model-context-protocol
config: {}
metadata_store:
type: sqlite
db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/nvidia}/registry.db
@ -227,11 +198,7 @@ datasets: []
scoring_fns: []
benchmarks: []
tool_groups:
- toolgroup_id: builtin::websearch
provider_id: tavily-search
- toolgroup_id: builtin::rag
provider_id: rag-runtime
- toolgroup_id: builtin::code_interpreter
provider_id: code-interpreter
server:
port: 8321

76
llama_stack/templates/open-benchmark/open_benchmark.py
View file

@ -6,7 +6,7 @@
from typing import Dict, List, Tuple
from llama_stack.apis.common.content_types import URL
from llama_stack.apis.datasets import DatasetPurpose, URIDataSource
from llama_stack.apis.models.models import ModelType
from llama_stack.distribution.datatypes import (
BenchmarkInput,
@ -171,76 +171,42 @@ def get_distribution_template() -> DistributionTemplate:
DatasetInput(
dataset_id="simpleqa",
provider_id="huggingface",
url=URL(uri="https://huggingface.co/datasets/llamastack/simpleqa"),
metadata={
"path": "llamastack/simpleqa",
"split": "train",
},
dataset_schema={
"input_query": {"type": "string"},
"expected_answer": {"type": "string"},
"chat_completion_input": {"type": "string"},
},
purpose=DatasetPurpose.eval_messages_answer,
source=URIDataSource(
uri="huggingface://datasets/llamastack/simpleqa?split=train",
),
),
DatasetInput(
dataset_id="mmlu_cot",
provider_id="huggingface",
url=URL(uri="https://huggingface.co/datasets/llamastack/mmlu_cot"),
metadata={
"path": "llamastack/mmlu_cot",
"name": "all",
"split": "test",
},
dataset_schema={
"input_query": {"type": "string"},
"expected_answer": {"type": "string"},
"chat_completion_input": {"type": "string"},
},
purpose=DatasetPurpose.eval_messages_answer,
source=URIDataSource(
uri="huggingface://datasets/llamastack/mmlu_cot?split=test&name=all",
),
),
DatasetInput(
dataset_id="gpqa_cot",
provider_id="huggingface",
url=URL(uri="https://huggingface.co/datasets/llamastack/gpqa_0shot_cot"),
metadata={
"path": "llamastack/gpqa_0shot_cot",
"name": "gpqa_main",
"split": "train",
},
dataset_schema={
"input_query": {"type": "string"},
"expected_answer": {"type": "string"},
"chat_completion_input": {"type": "string"},
},
purpose=DatasetPurpose.eval_messages_answer,
source=URIDataSource(
uri="huggingface://datasets/llamastack/gpqa_0shot_cot?split=test&name=gpqa_main",
),
),
DatasetInput(
dataset_id="math_500",
provider_id="huggingface",
url=URL(uri="https://huggingface.co/datasets/llamastack/math_500"),
metadata={
"path": "llamastack/math_500",
"split": "test",
},
dataset_schema={
"input_query": {"type": "string"},
"expected_answer": {"type": "string"},
"chat_completion_input": {"type": "string"},
},
purpose=DatasetPurpose.eval_messages_answer,
source=URIDataSource(
uri="huggingface://datasets/llamastack/math_500?split=test",
),
),
DatasetInput(
dataset_id="bfcl",
provider_id="huggingface",
url=URL(uri="https://huggingface.co/datasets/llamastack/bfcl_v3"),
metadata={
"path": "llamastack/bfcl_v3",
"split": "train",
},
dataset_schema={
"function": {"type": "string"},
"language": {"type": "string"},
"ground_truth": {"type": "string"},
"id": {"type": "string"},
"chat_completion_input": {"type": "string"},
},
purpose=DatasetPurpose.eval_messages_answer,
source=URIDataSource(
uri="huggingface://datasets/llamastack/bfcl_v3?split=train",
),
),
]

91
llama_stack/templates/open-benchmark/run.yaml
View file

@ -158,80 +158,39 @@ shields:
- shield_id: meta-llama/Llama-Guard-3-8B
vector_dbs: []
datasets:
- dataset_schema:
input_query:
type: string
expected_answer:
type: string
chat_completion_input:
type: string
url:
uri: https://huggingface.co/datasets/llamastack/simpleqa
metadata:
path: llamastack/simpleqa
split: train
- purpose: eval/messages-answer
source:
type: uri
uri: huggingface://datasets/llamastack/simpleqa?split=train
metadata: {}
dataset_id: simpleqa
provider_id: huggingface
- dataset_schema:
input_query:
type: string
expected_answer:
type: string
chat_completion_input:
type: string
url:
uri: https://huggingface.co/datasets/llamastack/mmlu_cot
metadata:
path: llamastack/mmlu_cot
name: all
split: test
- purpose: eval/messages-answer
source:
type: uri
uri: huggingface://datasets/llamastack/mmlu_cot?split=test&name=all
metadata: {}
dataset_id: mmlu_cot
provider_id: huggingface
- dataset_schema:
input_query:
type: string
expected_answer:
type: string
chat_completion_input:
type: string
url:
uri: https://huggingface.co/datasets/llamastack/gpqa_0shot_cot
metadata:
path: llamastack/gpqa_0shot_cot
name: gpqa_main
split: train
- purpose: eval/messages-answer
source:
type: uri
uri: huggingface://datasets/llamastack/gpqa_0shot_cot?split=test&name=gpqa_main
metadata: {}
dataset_id: gpqa_cot
provider_id: huggingface
- dataset_schema:
input_query:
type: string
expected_answer:
type: string
chat_completion_input:
type: string
url:
uri: https://huggingface.co/datasets/llamastack/math_500
metadata:
path: llamastack/math_500
split: test
- purpose: eval/messages-answer
source:
type: uri
uri: huggingface://datasets/llamastack/math_500?split=test
metadata: {}
dataset_id: math_500
provider_id: huggingface
- dataset_schema:
function:
type: string
language:
type: string
ground_truth:
type: string
id:
type: string
chat_completion_input:
type: string
url:
uri: https://huggingface.co/datasets/llamastack/bfcl_v3
metadata:
path: llamastack/bfcl_v3
split: train
- purpose: eval/messages-answer
source:
type: uri
uri: huggingface://datasets/llamastack/bfcl_v3?split=train
metadata: {}
dataset_id: bfcl
provider_id: huggingface
scoring_fns: []

2
llama_stack/templates/remote-vllm/run-with-safety.yaml
View file

@ -18,12 +18,14 @@ providers:
url: ${env.VLLM_URL}
max_tokens: ${env.VLLM_MAX_TOKENS:4096}
api_token: ${env.VLLM_API_TOKEN:fake}
tls_verify: ${env.VLLM_TLS_VERIFY:true}
- provider_id: vllm-safety
provider_type: remote::vllm
config:
url: ${env.SAFETY_VLLM_URL}
max_tokens: ${env.VLLM_MAX_TOKENS:4096}
api_token: ${env.VLLM_API_TOKEN:fake}
tls_verify: ${env.VLLM_TLS_VERIFY:true}
- provider_id: sentence-transformers
provider_type: inline::sentence-transformers
config: {}

1
llama_stack/templates/remote-vllm/run.yaml
View file

@ -18,6 +18,7 @@ providers:
url: ${env.VLLM_URL}
max_tokens: ${env.VLLM_MAX_TOKENS:4096}
api_token: ${env.VLLM_API_TOKEN:fake}
tls_verify: ${env.VLLM_TLS_VERIFY:true}
- provider_id: sentence-transformers
provider_type: inline::sentence-transformers
config: {}

3
llama_stack/templates/template.py
View file

@ -11,6 +11,7 @@ import jinja2
import yaml
from pydantic import BaseModel, Field
from llama_stack.apis.datasets import DatasetPurpose
from llama_stack.apis.models.models import ModelType
from llama_stack.distribution.datatypes import (
Api,
@ -214,7 +215,9 @@ class DistributionTemplate(BaseModel):
# Register YAML representer for ModelType
yaml.add_representer(ModelType, enum_representer)
yaml.add_representer(DatasetPurpose, enum_representer)
yaml.SafeDumper.add_representer(ModelType, enum_representer)
yaml.SafeDumper.add_representer(DatasetPurpose, enum_representer)
for output_dir in [yaml_output_dir, doc_output_dir]:
output_dir.mkdir(parents=True, exist_ok=True)

1
pyproject.toml
View file

@ -269,6 +269,7 @@ exclude = [
"^llama_stack/providers/remote/inference/together/",
"^llama_stack/providers/remote/inference/vllm/",
"^llama_stack/providers/remote/safety/bedrock/",
"^llama_stack/providers/remote/safety/nvidia/",
"^llama_stack/providers/remote/safety/sample/",
"^llama_stack/providers/remote/tool_runtime/bing_search/",
"^llama_stack/providers/remote/tool_runtime/brave_search/",

3
llama_stack/scripts/distro_codegen.py → scripts/distro_codegen.py Normal file → Executable file
View file

@ -1,3 +1,4 @@
#!/usr/bin/env python
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
@ -20,7 +21,7 @@ from llama_stack.distribution.build import (
get_provider_dependencies,
)
REPO_ROOT = Path(__file__).parent.parent.parent
REPO_ROOT = Path(__file__).parent.parent
class ChangedPathTracker:

1
scripts/gen-changelog.py Normal file → Executable file
View file

@ -1,3 +1,4 @@
#!/usr/bin/env python
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#

5
llama_stack/scripts/generate_prompt_format.py → scripts/generate_prompt_format.py Normal file → Executable file
View file

@ -1,3 +1,4 @@
#!/usr/bin/env python
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
@ -18,7 +19,7 @@ import fire
from llama_stack.models.llama.sku_list import resolve_model
from llama_stack.providers.inline.inference.meta_reference.config import MetaReferenceInferenceConfig
from llama_stack.providers.inline.inference.meta_reference.generation import Llama
from llama_stack.providers.inline.inference.meta_reference.llama3.generation import Llama3
THIS_DIR = Path(__file__).parent.resolve()
@ -41,7 +42,7 @@ def run_main(
llama_model = resolve_model(model_id)
if not llama_model:
raise ValueError(f"Model {model_id} not found")
generator = Llama.build(
generator = Llama3.build(
config=config,
model_id=model_id,
llama_model=llama_model,

8
llama_stack/scripts/run_client_sdk_tests.py → scripts/run_client_sdk_tests.py Normal file → Executable file
View file

@ -1,3 +1,4 @@
#!/usr/bin/env python
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
@ -15,8 +16,7 @@ Script for running api on AsyncLlamaStackAsLibraryClient with templates
Assuming directory structure:
- llama-stack
- llama_stack
- scripts
- scripts
- tests
- api
@ -25,10 +25,10 @@ Example command:
cd llama-stack
EXPORT TOGETHER_API_KEY=<..>
EXPORT FIREWORKS_API_KEY=<..>
python llama_stack/scripts/run_client_sdk_tests.py --templates together fireworks --report
./scripts/run_client_sdk_tests.py --templates together fireworks --report
"""
REPO_ROOT = Path(__file__).parent.parent.parent
REPO_ROOT = Path(__file__).parent.parent
CLIENT_SDK_TESTS_RELATIVE_PATH = "tests/api/"

118
tests/integration/agents/test_agents.py
View file

@ -271,7 +271,7 @@ def test_custom_tool(llama_stack_client_with_mocked_inference, agent_config):
client_tool = get_boiling_point
agent_config = {
**agent_config,
"tools": ["builtin::websearch", client_tool],
"tools": [client_tool],
}
agent = Agent(llama_stack_client_with_mocked_inference, **agent_config)
@ -320,42 +320,55 @@ def test_custom_tool_infinite_loop(llama_stack_client_with_mocked_inference, age
assert num_tool_calls <= 5
def test_tool_choice(llama_stack_client_with_mocked_inference, agent_config):
def run_agent(tool_choice):
client_tool = get_boiling_point
test_agent_config = {
**agent_config,
"tool_config": {"tool_choice": tool_choice},
"tools": [client_tool],
}
agent = Agent(llama_stack_client_with_mocked_inference, **test_agent_config)
session_id = agent.create_session(f"test-session-{uuid4()}")
response = agent.create_turn(
messages=[
{
"role": "user",
"content": "What is the boiling point of polyjuice?",
},
],
session_id=session_id,
stream=False,
)
return [step for step in response.steps if step.step_type == "tool_execution"]
tool_execution_steps = run_agent("required")
def test_tool_choice_required(llama_stack_client_with_mocked_inference, agent_config):
tool_execution_steps = run_agent_with_tool_choice(
llama_stack_client_with_mocked_inference, agent_config, "required"
)
assert len(tool_execution_steps) > 0
tool_execution_steps = run_agent("none")
def test_tool_choice_none(llama_stack_client_with_mocked_inference, agent_config):
tool_execution_steps = run_agent_with_tool_choice(llama_stack_client_with_mocked_inference, agent_config, "none")
assert len(tool_execution_steps) == 0
tool_execution_steps = run_agent("get_boiling_point")
def test_tool_choice_get_boiling_point(llama_stack_client_with_mocked_inference, agent_config):
if "llama" not in agent_config["model"].lower():
pytest.xfail("NotImplemented for non-llama models")
tool_execution_steps = run_agent_with_tool_choice(
llama_stack_client_with_mocked_inference, agent_config, "get_boiling_point"
)
assert len(tool_execution_steps) >= 1 and tool_execution_steps[0].tool_calls[0].tool_name == "get_boiling_point"
def run_agent_with_tool_choice(client, agent_config, tool_choice):
client_tool = get_boiling_point
test_agent_config = {
**agent_config,
"tool_config": {"tool_choice": tool_choice},
"tools": [client_tool],
"max_infer_iters": 2,
}
agent = Agent(client, **test_agent_config)
session_id = agent.create_session(f"test-session-{uuid4()}")
response = agent.create_turn(
messages=[
{
"role": "user",
"content": "What is the boiling point of polyjuice?",
},
],
session_id=session_id,
stream=False,
)
return [step for step in response.steps if step.step_type == "tool_execution"]
@pytest.mark.parametrize("rag_tool_name", ["builtin::rag/knowledge_search", "builtin::rag"])
def test_rag_agent(llama_stack_client_with_mocked_inference, agent_config, rag_tool_name):
urls = ["chat.rst", "llama3.rst", "memory_optimizations.rst", "lora_finetune.rst"]
@ -571,7 +584,7 @@ def test_rag_and_code_agent(llama_stack_client_with_mocked_inference, agent_conf
[(get_boiling_point, False), (get_boiling_point_with_metadata, True)],
)
def test_create_turn_response(llama_stack_client_with_mocked_inference, agent_config, client_tools):
client_tool, expectes_metadata = client_tools
client_tool, expects_metadata = client_tools
agent_config = {
**agent_config,
"input_shields": [],
@ -597,7 +610,7 @@ def test_create_turn_response(llama_stack_client_with_mocked_inference, agent_co
assert steps[0].step_type == "inference"
assert steps[1].step_type == "tool_execution"
assert steps[1].tool_calls[0].tool_name.startswith("get_boiling_point")
if expectes_metadata:
if expects_metadata:
assert steps[1].tool_responses[0].metadata["source"] == "https://www.google.com"
assert steps[2].step_type == "inference"
@ -609,3 +622,44 @@ def test_create_turn_response(llama_stack_client_with_mocked_inference, agent_co
assert last_step_completed_at < step.started_at
assert step.started_at < step.completed_at
last_step_completed_at = step.completed_at
def test_multi_tool_calls(llama_stack_client_with_mocked_inference, agent_config):
if "gpt" not in agent_config["model"]:
pytest.xfail("Only tested on GPT models")
agent_config = {
**agent_config,
"tools": [get_boiling_point],
}
agent = Agent(llama_stack_client_with_mocked_inference, **agent_config)
session_id = agent.create_session(f"test-session-{uuid4()}")
response = agent.create_turn(
messages=[
{
"role": "user",
"content": "Call get_boiling_point twice to answer: What is the boiling point of polyjuice in both celsius and fahrenheit?",
},
],
session_id=session_id,
stream=False,
)
steps = response.steps
assert len(steps) == 7
assert steps[0].step_type == "shield_call"
assert steps[1].step_type == "inference"
assert steps[2].step_type == "shield_call"
assert steps[3].step_type == "tool_execution"
assert steps[4].step_type == "shield_call"
assert steps[5].step_type == "inference"
assert steps[6].step_type == "shield_call"
tool_execution_step = steps[3]
assert len(tool_execution_step.tool_calls) == 2
assert tool_execution_step.tool_calls[0].tool_name.startswith("get_boiling_point")
assert tool_execution_step.tool_calls[1].tool_name.startswith("get_boiling_point")
output = response.output_message.content.lower()
assert "-100" in output and "-212" in output

5
tests/integration/datasetio/__init__.py
View file

@ -1,5 +0,0 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.

114
tests/integration/datasetio/test_datasetio.py
View file

@ -1,114 +0,0 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import base64
import mimetypes
import os
from pathlib import Path
import pytest
# How to run this test:
#
# LLAMA_STACK_CONFIG="template-name" pytest -v tests/integration/datasetio
@pytest.fixture
def dataset_for_test(llama_stack_client):
dataset_id = "test_dataset"
register_dataset(llama_stack_client, dataset_id=dataset_id)
yield
# Teardown - this always runs, even if the test fails
try:
llama_stack_client.datasets.unregister(dataset_id)
except Exception as e:
print(f"Warning: Failed to unregister test_dataset: {e}")
def data_url_from_file(file_path: str) -> str:
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
with open(file_path, "rb") as file:
file_content = file.read()
base64_content = base64.b64encode(file_content).decode("utf-8")
mime_type, _ = mimetypes.guess_type(file_path)
data_url = f"data:{mime_type};base64,{base64_content}"
return data_url
def register_dataset(llama_stack_client, for_generation=False, for_rag=False, dataset_id="test_dataset"):
if for_rag:
test_file = Path(os.path.abspath(__file__)).parent / "test_rag_dataset.csv"
else:
test_file = Path(os.path.abspath(__file__)).parent / "test_dataset.csv"
test_url = data_url_from_file(str(test_file))
if for_generation:
dataset_schema = {
"expected_answer": {"type": "string"},
"input_query": {"type": "string"},
"chat_completion_input": {"type": "chat_completion_input"},
}
elif for_rag:
dataset_schema = {
"expected_answer": {"type": "string"},
"input_query": {"type": "string"},
"generated_answer": {"type": "string"},
"context": {"type": "string"},
}
else:
dataset_schema = {
"expected_answer": {"type": "string"},
"input_query": {"type": "string"},
"generated_answer": {"type": "string"},
}
dataset_providers = [x for x in llama_stack_client.providers.list() if x.api == "datasetio"]
dataset_provider_id = dataset_providers[0].provider_id
llama_stack_client.datasets.register(
dataset_id=dataset_id,
dataset_schema=dataset_schema,
url=dict(uri=test_url),
provider_id=dataset_provider_id,
)
def test_register_unregister_dataset(llama_stack_client):
register_dataset(llama_stack_client)
response = llama_stack_client.datasets.list()
assert isinstance(response, list)
assert len(response) == 1
assert response[0].identifier == "test_dataset"
llama_stack_client.datasets.unregister("test_dataset")
response = llama_stack_client.datasets.list()
assert isinstance(response, list)
assert len(response) == 0
def test_get_rows_paginated(llama_stack_client, dataset_for_test):
response = llama_stack_client.datasetio.get_rows_paginated(
dataset_id="test_dataset",
rows_in_page=3,
)
assert isinstance(response.rows, list)
assert len(response.rows) == 3
assert response.next_page_token == "3"
# iterate over all rows
response = llama_stack_client.datasetio.get_rows_paginated(
dataset_id="test_dataset",
rows_in_page=2,
page_token=response.next_page_token,
)
assert isinstance(response.rows, list)
assert len(response.rows) == 2
assert response.next_page_token == "5"

0
llama_stack/scripts/__init__.py → tests/integration/datasets/__init__.py
View file

0
tests/integration/datasetio/test_dataset.csv → tests/integration/datasets/test_dataset.csv
View file

95
tests/integration/datasets/test_datasets.py Normal file
View file

@ -0,0 +1,95 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import base64
import mimetypes
import os
import pytest
# How to run this test:
#
# LLAMA_STACK_CONFIG="template-name" pytest -v tests/integration/datasets
def data_url_from_file(file_path: str) -> str:
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
with open(file_path, "rb") as file:
file_content = file.read()
base64_content = base64.b64encode(file_content).decode("utf-8")
mime_type, _ = mimetypes.guess_type(file_path)
data_url = f"data:{mime_type};base64,{base64_content}"
return data_url
@pytest.mark.parametrize(
"purpose, source, provider_id, limit",
[
(
"eval/messages-answer",
{
"type": "uri",
"uri": "huggingface://datasets/llamastack/simpleqa?split=train",
},
"huggingface",
10,
),
(
"eval/messages-answer",
{
"type": "rows",
"rows": [
{
"messages": [{"role": "user", "content": "Hello, world!"}],
"answer": "Hello, world!",
},
{
"messages": [
{
"role": "user",
"content": "What is the capital of France?",
}
],
"answer": "Paris",
},
],
},
"localfs",
2,
),
(
"eval/messages-answer",
{
"type": "uri",
"uri": data_url_from_file(os.path.join(os.path.dirname(__file__), "test_dataset.csv")),
},
"localfs",
5,
),
],
)
def test_register_and_iterrows(llama_stack_client, purpose, source, provider_id, limit):
dataset = llama_stack_client.datasets.register(
purpose=purpose,
source=source,
)
assert dataset.identifier is not None
assert dataset.provider_id == provider_id
iterrow_response = llama_stack_client.datasets.iterrows(dataset.identifier, limit=limit)
assert len(iterrow_response.data) == limit
dataset_list = llama_stack_client.datasets.list()
assert dataset.identifier in [d.identifier for d in dataset_list]
llama_stack_client.datasets.unregister(dataset.identifier)
dataset_list = llama_stack_client.datasets.list()
assert dataset.identifier not in [d.identifier for d in dataset_list]

0
tests/integration/datasetio/test_rag_dataset.csv → tests/integration/datasets/test_rag_dataset.csv
View file

39
tests/integration/eval/test_eval.py
View file

@ -4,10 +4,11 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import uuid
from pathlib import Path
import pytest
from ..datasetio.test_datasetio import register_dataset
from ..datasets.test_datasets import data_url_from_file
# How to run this test:
#
@ -16,15 +17,21 @@ from ..datasetio.test_datasetio import register_dataset
@pytest.mark.parametrize("scoring_fn_id", ["basic::equality"])
def test_evaluate_rows(llama_stack_client, text_model_id, scoring_fn_id):
register_dataset(llama_stack_client, for_generation=True, dataset_id="test_dataset_for_eval")
response = llama_stack_client.datasets.list()
assert any(x.identifier == "test_dataset_for_eval" for x in response)
rows = llama_stack_client.datasetio.get_rows_paginated(
dataset_id="test_dataset_for_eval",
rows_in_page=3,
dataset = llama_stack_client.datasets.register(
purpose="eval/messages-answer",
source={
"type": "uri",
"uri": data_url_from_file(Path(__file__).parent.parent / "datasets" / "test_dataset.csv"),
},
)
assert len(rows.rows) == 3
response = llama_stack_client.datasets.list()
assert any(x.identifier == dataset.identifier for x in response)
rows = llama_stack_client.datasets.iterrows(
dataset_id=dataset.identifier,
limit=3,
)
assert len(rows.data) == 3
scoring_functions = [
scoring_fn_id,
@ -32,7 +39,7 @@ def test_evaluate_rows(llama_stack_client, text_model_id, scoring_fn_id):
benchmark_id = str(uuid.uuid4())
llama_stack_client.benchmarks.register(
benchmark_id=benchmark_id,
dataset_id="test_dataset_for_eval",
dataset_id=dataset.identifier,
scoring_functions=scoring_functions,
)
list_benchmarks = llama_stack_client.benchmarks.list()
@ -40,7 +47,7 @@ def test_evaluate_rows(llama_stack_client, text_model_id, scoring_fn_id):
response = llama_stack_client.eval.evaluate_rows(
benchmark_id=benchmark_id,
input_rows=rows.rows,
input_rows=rows.data,
scoring_functions=scoring_functions,
benchmark_config={
"eval_candidate": {
@ -59,11 +66,17 @@ def test_evaluate_rows(llama_stack_client, text_model_id, scoring_fn_id):
@pytest.mark.parametrize("scoring_fn_id", ["basic::subset_of"])
def test_evaluate_benchmark(llama_stack_client, text_model_id, scoring_fn_id):
register_dataset(llama_stack_client, for_generation=True, dataset_id="test_dataset_for_eval_2")
dataset = llama_stack_client.datasets.register(
purpose="eval/messages-answer",
source={
"type": "uri",
"uri": data_url_from_file(Path(__file__).parent.parent / "datasets" / "test_dataset.csv"),
},
)
benchmark_id = str(uuid.uuid4())
llama_stack_client.benchmarks.register(
benchmark_id=benchmark_id,
dataset_id="test_dataset_for_eval_2",
dataset_id=dataset.identifier,
scoring_functions=[scoring_fn_id],
)

89
tests/integration/scoring/test_scoring.py
View file

@ -5,23 +5,11 @@
# the root directory of this source tree.
from pathlib import Path
import pandas as pd
import pytest
from ..datasetio.test_datasetio import register_dataset
@pytest.fixture
def rag_dataset_for_test(llama_stack_client):
dataset_id = "test_dataset"
register_dataset(llama_stack_client, for_rag=True, dataset_id=dataset_id)
yield # This is where the test function will run
# Teardown - this always runs, even if the test fails
try:
llama_stack_client.datasets.unregister(dataset_id)
except Exception as e:
print(f"Warning: Failed to unregister test_dataset: {e}")
@pytest.fixture
def sample_judge_prompt_template():
@ -92,49 +80,34 @@ def test_scoring_functions_register(
# TODO: add unregister api for scoring functions
def test_scoring_score(llama_stack_client, rag_dataset_for_test):
@pytest.mark.parametrize("scoring_fn_id", ["basic::equality"])
def test_scoring_score(llama_stack_client, scoring_fn_id):
# scoring individual rows
rows = llama_stack_client.datasetio.get_rows_paginated(
dataset_id="test_dataset",
rows_in_page=3,
)
assert len(rows.rows) == 3
df = pd.read_csv(Path(__file__).parent.parent / "datasets" / "test_dataset.csv")
rows = df.to_dict(orient="records")
scoring_fns_list = llama_stack_client.scoring_functions.list()
scoring_functions = {
scoring_fns_list[0].identifier: None,
scoring_fn_id: None,
}
response = llama_stack_client.scoring.score(
input_rows=rows.rows,
input_rows=rows,
scoring_functions=scoring_functions,
)
assert len(response.results) == len(scoring_functions)
for x in scoring_functions:
assert x in response.results
assert len(response.results[x].score_rows) == len(rows.rows)
# score batch
response = llama_stack_client.scoring.score_batch(
dataset_id="test_dataset",
scoring_functions=scoring_functions,
save_results_dataset=False,
)
assert len(response.results) == len(scoring_functions)
for x in scoring_functions:
assert x in response.results
assert len(response.results[x].score_rows) == 5
assert len(response.results[x].score_rows) == len(rows)
def test_scoring_score_with_params_llm_as_judge(
llama_stack_client, sample_judge_prompt_template, judge_model_id, rag_dataset_for_test
llama_stack_client,
sample_judge_prompt_template,
judge_model_id,
):
# scoring individual rows
rows = llama_stack_client.datasetio.get_rows_paginated(
dataset_id="test_dataset",
rows_in_page=3,
)
assert len(rows.rows) == 3
df = pd.read_csv(Path(__file__).parent.parent / "datasets" / "test_dataset.csv")
rows = df.to_dict(orient="records")
scoring_functions = {
"llm-as-judge::base": dict(
@ -149,24 +122,13 @@ def test_scoring_score_with_params_llm_as_judge(
}
response = llama_stack_client.scoring.score(
input_rows=rows.rows,
input_rows=rows,
scoring_functions=scoring_functions,
)
assert len(response.results) == len(scoring_functions)
for x in scoring_functions:
assert x in response.results
assert len(response.results[x].score_rows) == len(rows.rows)
# score batch
response = llama_stack_client.scoring.score_batch(
dataset_id="test_dataset",
scoring_functions=scoring_functions,
save_results_dataset=False,
)
assert len(response.results) == len(scoring_functions)
for x in scoring_functions:
assert x in response.results
assert len(response.results[x].score_rows) == 5
assert len(response.results[x].score_rows) == len(rows)
@pytest.mark.parametrize(
@ -178,13 +140,14 @@ def test_scoring_score_with_params_llm_as_judge(
],
)
def test_scoring_score_with_aggregation_functions(
llama_stack_client, sample_judge_prompt_template, judge_model_id, provider_id, rag_dataset_for_test
llama_stack_client,
sample_judge_prompt_template,
judge_model_id,
provider_id,
rag_dataset_for_test,
):
rows = llama_stack_client.datasetio.get_rows_paginated(
dataset_id="test_dataset",
rows_in_page=3,
)
assert len(rows.rows) == 3
df = pd.read_csv(Path(__file__).parent.parent / "datasets" / "test_dataset.csv")
rows = df.to_dict(orient="records")
scoring_fns_list = [x for x in llama_stack_client.scoring_functions.list() if x.provider_id == provider_id]
if len(scoring_fns_list) == 0:
@ -224,12 +187,12 @@ def test_scoring_score_with_aggregation_functions(
scoring_functions[scoring_fn.identifier] = None
response = llama_stack_client.scoring.score(
input_rows=rows.rows,
input_rows=rows,
scoring_functions=scoring_functions,
)
assert len(response.results) == len(scoring_functions)
for x in scoring_functions:
assert x in response.results
assert len(response.results[x].score_rows) == len(rows.rows)
assert len(response.results[x].score_rows) == len(rows)
assert len(response.results[x].aggregated_results) == len(aggr_fns)

4
tests/unit/providers/inference/test_remote_vllm.py
View file

@ -187,8 +187,8 @@ def test_chat_completion_doesnt_block_event_loop(caplog):
loop.set_debug(True)
caplog.set_level(logging.WARNING)
# Log when event loop is blocked for more than 100ms
loop.slow_callback_duration = 0.1
# Log when event loop is blocked for more than 200ms
loop.slow_callback_duration = 0.2
# Sleep for 500ms in our delayed http response
sleep_time = 0.5