llama-stack/docs/cli_reference.md

Llama CLI Reference

The llama CLI tool helps you setup and use the Llama toolchain & agentic systems. It should be available on your path after installing the llama-stack package.

Subcommands

1. download: llama cli tools supports downloading the model from Meta or HuggingFace.
2. model: Lists available models and their properties.
3. stack: Allows you to build and run a Llama Stack server. You can read more about this here.

Sample Usage

llama --help

usage: llama [-h] {download,model,stack} ...

Welcome to the Llama CLI

options:
  -h, --help            show this help message and exit

subcommands:
  {download,model,stack}

Step 1. Get the models

You first need to have models downloaded locally.

To download any model you need the Model Descriptor. This can be obtained by running the command

llama model list

You should see a table like this:

+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Model Descriptor                      | HuggingFace Repo                            | Context Length | Hardware Requirements      |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-8B                      | meta-llama/Meta-Llama-3.1-8B                | 128K           | 1 GPU, each >= 20GB VRAM   |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-70B                     | meta-llama/Meta-Llama-3.1-70B               | 128K           | 8 GPUs, each >= 20GB VRAM  |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-405B:bf16-mp8           |                                             | 128K           | 8 GPUs, each >= 120GB VRAM |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-405B                    | meta-llama/Meta-Llama-3.1-405B-FP8          | 128K           | 8 GPUs, each >= 70GB VRAM  |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-405B:bf16-mp16          | meta-llama/Meta-Llama-3.1-405B              | 128K           | 16 GPUs, each >= 70GB VRAM |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-8B-Instruct             | meta-llama/Meta-Llama-3.1-8B-Instruct       | 128K           | 1 GPU, each >= 20GB VRAM   |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-70B-Instruct            | meta-llama/Meta-Llama-3.1-70B-Instruct      | 128K           | 8 GPUs, each >= 20GB VRAM  |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-405B-Instruct:bf16-mp8  |                                             | 128K           | 8 GPUs, each >= 120GB VRAM |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-405B-Instruct           | meta-llama/Meta-Llama-3.1-405B-Instruct-FP8 | 128K           | 8 GPUs, each >= 70GB VRAM  |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Meta-Llama3.1-405B-Instruct:bf16-mp16 | meta-llama/Meta-Llama-3.1-405B-Instruct     | 128K           | 16 GPUs, each >= 70GB VRAM |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Llama-Guard-3-8B                      | meta-llama/Llama-Guard-3-8B                 | 128K           | 1 GPU, each >= 20GB VRAM   |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Llama-Guard-3-8B:int8-mp1             | meta-llama/Llama-Guard-3-8B-INT8            | 128K           | 1 GPU, each >= 10GB VRAM   |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+
| Prompt-Guard-86M                      | meta-llama/Prompt-Guard-86M                 | 128K           | 1 GPU, each >= 1GB VRAM    |
+---------------------------------------+---------------------------------------------+----------------+----------------------------+

To download models, you can use the llama download command.

Downloading from Meta

Here is an example download command to get the 8B/70B Instruct model. You will need META_URL which can be obtained from here

Download the required checkpoints using the following commands:

# download the 8B model, this can be run on a single GPU
llama download --source meta --model-id Meta-Llama3.1-8B-Instruct --meta-url META_URL

# you can also get the 70B model, this will require 8 GPUs however
llama download --source meta --model-id Meta-Llama3.1-70B-Instruct --meta-url META_URL

# llama-agents have safety enabled by default. For this, you will need
# safety models -- Llama-Guard and Prompt-Guard
llama download --source meta --model-id Prompt-Guard-86M --meta-url META_URL
llama download --source meta --model-id Llama-Guard-3-8B --meta-url META_URL

Downloading from Huggingface

Essentially, the same commands above work, just replace --source meta with --source huggingface.

llama download --source huggingface --model-id  Meta-Llama3.1-8B-Instruct --hf-token <HF_TOKEN>

llama download --source huggingface --model-id Meta-Llama3.1-70B-Instruct --hf-token <HF_TOKEN>

llama download --source huggingface --model-id Llama-Guard-3-8B --ignore-patterns *original*
llama download --source huggingface --model-id Prompt-Guard-86M --ignore-patterns *original*

Important: Set your environment variable HF_TOKEN or pass in --hf-token to the command to validate your access. You can find your token at https://huggingface.co/settings/tokens.

Tip: Default for llama download is to run with --ignore-patterns *.safetensors since we use the .pth files in the original folder. For Llama Guard and Prompt Guard, however, we need safetensors. Hence, please run with --ignore-patterns original so that safetensors are downloaded and .pth files are ignored.

Downloading via Ollama

If you're already using ollama, we also have a supported Llama Stack distribution local-ollama and you can continue to use ollama for managing model downloads.

ollama pull llama3.1:8b-instruct-fp16
ollama pull llama3.1:70b-instruct-fp16

Note

Only the above two models are currently supported by Ollama.

Step 2: Understand the models

The llama model command helps you explore the model’s interface.

2.1 Subcommands

1. download: Download the model from different sources. (meta, huggingface)
2. list: Lists all the models available for download with hardware requirements to deploy the models.
3. template: <TODO: What is a template?>
4. describe: Describes all the properties of the model.

2.2 Sample Usage

llama model <subcommand> <options>

llama model --help

usage: llama model [-h] {download,list,template,describe} ...

Work with llama models

options:
  -h, --help            show this help message and exit

model_subcommands:
  {download,list,template,describe}

You can use the describe command to know more about a model:

llama model describe -m Meta-Llama3.1-8B-Instruct

2.3 Describe

+-----------------------------+---------------------------------------+
| Model                       | Meta-                                 |
|                             | Llama3.1-8B-Instruct                  |
+-----------------------------+---------------------------------------+
| HuggingFace ID              | meta-llama/Meta-Llama-3.1-8B-Instruct |
+-----------------------------+---------------------------------------+
| Description                 | Llama 3.1 8b instruct model           |
+-----------------------------+---------------------------------------+
| Context Length              | 128K tokens                           |
+-----------------------------+---------------------------------------+
| Weights format              | bf16                                  |
+-----------------------------+---------------------------------------+
| Model params.json           | {                                     |
|                             |     "dim": 4096,                      |
|                             |     "n_layers": 32,                   |
|                             |     "n_heads": 32,                    |
|                             |     "n_kv_heads": 8,                  |
|                             |     "vocab_size": 128256,             |
|                             |     "ffn_dim_multiplier": 1.3,        |
|                             |     "multiple_of": 1024,              |
|                             |     "norm_eps": 1e-05,                |
|                             |     "rope_theta": 500000.0,           |
|                             |     "use_scaled_rope": true           |
|                             | }                                     |
+-----------------------------+---------------------------------------+
| Recommended sampling params | {                                     |
|                             |     "strategy": "top_p",              |
|                             |     "temperature": 1.0,               |
|                             |     "top_p": 0.9,                     |
|                             |     "top_k": 0                        |
|                             | }                                     |
+-----------------------------+---------------------------------------+

2.4 Template

You can even run llama model template see all of the templates and their tokens:

llama model template

+-----------+---------------------------------+
| Role      | Template Name                   |
+-----------+---------------------------------+
| user      | user-default                    |
| assistant | assistant-builtin-tool-call     |
| assistant | assistant-custom-tool-call      |
| assistant | assistant-default               |
| system    | system-builtin-and-custom-tools |
| system    | system-builtin-tools-only       |
| system    | system-custom-tools-only        |
| system    | system-default                  |
| tool      | tool-success                    |
| tool      | tool-failure                    |
+-----------+---------------------------------+

And fetch an example by passing it to --name:

llama model template --name tool-success

+----------+----------------------------------------------------------------+
| Name     | tool-success                                                   |
+----------+----------------------------------------------------------------+
| Template | <|start_header_id|>ipython<|end_header_id|>                    |
|          |                                                                |
|          | completed                                                      |
|          | [stdout]{"results":["something                                 |
|          | something"]}[/stdout]<|eot_id|>                                |
|          |                                                                |
+----------+----------------------------------------------------------------+
| Notes    | Note ipython header and [stdout]                               |
+----------+----------------------------------------------------------------+

Or:

llama model template --name system-builtin-tools-only

+----------+--------------------------------------------+
| Name     | system-builtin-tools-only                  |
+----------+--------------------------------------------+
| Template | <|start_header_id|>system<|end_header_id|> |
|          |                                            |
|          | Environment: ipython                       |
|          | Tools: brave_search, wolfram_alpha         |
|          |                                            |
|          | Cutting Knowledge Date: December 2023      |
|          | Today Date: 21 August 2024                 |
|          | <|eot_id|>                                 |
|          |                                            |
+----------+--------------------------------------------+
| Notes    |                                            |
+----------+--------------------------------------------+

These commands can help understand the model interface and how prompts / messages are formatted for various scenarios.

NOTE: Outputs in terminal are color printed to show special tokens.

Step 3: Building, and Configuring Llama Stack Distributions

• Please see our Getting Started guide for details.

Step 3.1. Build

In the following steps, imagine we'll be working with a Meta-Llama3.1-8B-Instruct model. We will name our build 8b-instruct to help us remember the config. We will start build our distribution (in the form of a Conda environment, or Docker image). In this step, we will specify:

• name: the name for our distribution (e.g. 8b-instruct)
• image_type: our build image type (conda | docker)
• distribution_spec: our distribution specs for specifying API providers
  • description: a short description of the configurations for the distribution
  • providers: specifies the underlying implementation for serving each API endpoint
  • image_type: conda | docker to specify whether to build the distribution in the form of Docker image or Conda environment.

Build a local distribution with conda

The following command and specifications allows you to get started with building.

llama stack build <path/to/config>

• You will be required to pass in a file path to the build.config file (e.g. ./llama_stack/distribution/example_configs/conda/local-conda-example-build.yaml). We provide some example build config files for configuring different types of distributions in the ./llama_stack/distribution/example_configs/ folder.

The file will be of the contents

$ cat ./llama_stack/distribution/example_configs/conda/local-conda-example-build.yaml

name: 8b-instruct
distribution_spec:
  distribution_type: local
  description: Use code from `llama_stack` itself to serve all llama stack APIs
  docker_image: null
  providers:
    inference: meta-reference
    memory: meta-reference-faiss
    safety: meta-reference
    agentic_system: meta-reference
    telemetry: console
image_type: conda

You may run the llama stack build command to generate your distribution with --name to override the name for your distribution.

$ llama stack build ~/.llama/distributions/conda/8b-instruct-build.yaml --name 8b-instruct
...
...
Build spec configuration saved at ~/.llama/distributions/conda/8b-instruct-build.yaml

After this step is complete, a file named 8b-instruct-build.yaml will be generated and saved at ~/.llama/distributions/conda/8b-instruct-build.yaml.

How to build distribution with different API providers using configs

To specify a different API provider, we can change the distribution_spec in our <name>-build.yaml config. For example, the following build spec allows you to build a distribution using TGI as the inference API provider.

$ cat ./llama_stack/distribution/example_configs/conda/local-tgi-conda-example-build.yaml

name: local-tgi-conda-example
distribution_spec:
  description: Use TGI (local or with Hugging Face Inference Endpoints for running LLM inference. When using HF Inference Endpoints, you must provide the name of the endpoint).
  docker_image: null
  providers:
    inference: remote::tgi
    memory: meta-reference-faiss
    safety: meta-reference
    agentic_system: meta-reference
    telemetry: console
image_type: conda

The following command allows you to build a distribution with TGI as the inference API provider, with the name tgi.

llama stack build --config ./llama_stack/distribution/example_configs/conda/local-tgi-conda-example-build.yaml --name tgi

We provide some example build configs to help you get started with building with different API providers.

How to build distribution with Docker image

To build a docker image, simply change the image_type to docker in our <name>-build.yaml file, and run llama stack build --config <name>-build.yaml.

$ cat ./llama_stack/distribution/example_configs/docker/local-docker-example-build.yaml

name: local-docker-example
distribution_spec:
  description: Use code from `llama_stack` itself to serve all llama stack APIs
  docker_image: null
  providers:
    inference: meta-reference
    memory: meta-reference-faiss
    safety: meta-reference
    agentic_system: meta-reference
    telemetry: console
image_type: docker

The following command allows you to build a Docker image with the name docker-local

llama stack build --config ./llama_stack/distribution/example_configs/docker/local-docker-example-build.yaml --name docker-local

Dockerfile created successfully in /tmp/tmp.I0ifS2c46A/DockerfileFROM python:3.10-slim
WORKDIR /app
...
...
You can run it with: podman run -p 8000:8000 llamastack-docker-local
Build spec configuration saved at /home/xiyan/.llama/distributions/docker/docker-local-build.yaml

Step 3.2. Configure

After our distribution is built (either in form of docker or conda environment), we will run the following command to

llama stack configure [<path/to/name.build.yaml> | <docker-image-name>]

• For conda environments: <path/to/name.build.yaml> would be the generated build spec saved from Step 1.
• For docker images downloaded from Dockerhub, you could also use as the argument.
  • Run docker images to check list of available images on your machine.

$ llama stack configure ~/.llama/distributions/conda/8b-instruct-build.yaml

Configuring API: inference (meta-reference)
Enter value for model (existing: Meta-Llama3.1-8B-Instruct) (required):
Enter value for quantization (optional):
Enter value for torch_seed (optional):
Enter value for max_seq_len (existing: 4096) (required):
Enter value for max_batch_size (existing: 1) (required):

Configuring API: memory (meta-reference-faiss)

Configuring API: safety (meta-reference)
Do you want to configure llama_guard_shield? (y/n): y
Entering sub-configuration for llama_guard_shield:
Enter value for model (default: Llama-Guard-3-8B) (required):
Enter value for excluded_categories (default: []) (required):
Enter value for disable_input_check (default: False) (required):
Enter value for disable_output_check (default: False) (required):
Do you want to configure prompt_guard_shield? (y/n): y
Entering sub-configuration for prompt_guard_shield:
Enter value for model (default: Prompt-Guard-86M) (required):

Configuring API: agentic_system (meta-reference)
Enter value for brave_search_api_key (optional):
Enter value for bing_search_api_key (optional):
Enter value for wolfram_api_key (optional):

Configuring API: telemetry (console)

YAML configuration has been written to ~/.llama/builds/conda/8b-instruct-run.yaml

After this step is successful, you should be able to find a run configuration spec in ~/.llama/builds/conda/8b-instruct-run.yaml with the following contents. You may edit this file to change the settings.

As you can see, we did basic configuration above and configured:

• inference to run on model Meta-Llama3.1-8B-Instruct (obtained from llama model list)
• Llama Guard safety shield with model Llama-Guard-3-8B
• Prompt Guard safety shield with model Prompt-Guard-86M

For how these configurations are stored as yaml, checkout the file printed at the end of the configuration.

Note that all configurations as well as models are stored in ~/.llama

Step 3.2.1 API Keys for Tools

API key configuration for the Agentic System will be asked by the llama stack build script when you install a Llama Stack distribution.

Tools that the model supports and which need API Keys --

• Brave for web search (https://api.search.brave.com/register)
• Wolfram for math operations (https://developer.wolframalpha.com/)

Tip If you do not have API keys, you can still run the app without model having access to the tools.

Step 3.3. Run

Now, let's start the Llama Stack Distribution Server. You will need the YAML configuration file which was written out at the end by the llama stack configure step.

llama stack run ~/.llama/builds/conda/8b-instruct-run.yaml

You should see the Llama Stack server start and print the APIs that it is supporting

$ llama stack run ~/.llama/builds/local/conda/8b-instruct.yaml

> initializing model parallel with size 1
> initializing ddp with size 1
> initializing pipeline with size 1
Loaded in 19.28 seconds
NCCL version 2.20.5+cuda12.4
Finished model load YES READY
Serving POST /inference/batch_chat_completion
Serving POST /inference/batch_completion
Serving POST /inference/chat_completion
Serving POST /inference/completion
Serving POST /safety/run_shields
Serving POST /agentic_system/memory_bank/attach
Serving POST /agentic_system/create
Serving POST /agentic_system/session/create
Serving POST /agentic_system/turn/create
Serving POST /agentic_system/delete
Serving POST /agentic_system/session/delete
Serving POST /agentic_system/memory_bank/detach
Serving POST /agentic_system/session/get
Serving POST /agentic_system/step/get
Serving POST /agentic_system/turn/get
Listening on :::5000
INFO:     Started server process [453333]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://[::]:5000 (Press CTRL+C to quit)

Note

Configuration is in ~/.llama/builds/local/conda/8b-instruct.yaml. Feel free to increase max_seq_len.

Important

The "local" distribution inference server currently only supports CUDA. It will not work on Apple Silicon machines.

This server is running a Llama model locally.

Step 3.4 Test with Client

Once the server is setup, we can test it with a client to see the example outputs.

cd /path/to/llama-stack
conda activate <env>  # any environment containing the llama-stack pip package will work

python -m llama_stack.apis.inference.client localhost 5000

This will run the chat completion client and query the distribution’s /inference/chat_completion API.

Here is an example output:

Initializing client for http://localhost:5000
User>hello world, troll me in two-paragraphs about 42

Assistant> You think you're so smart, don't you? You think you can just waltz in here and ask about 42, like it's some kind of trivial matter. Well, let me tell you, 42 is not just a number, it's a way of life. It's the answer to the ultimate question of life, the universe, and everything, according to Douglas Adams' magnum opus, "The Hitchhiker's Guide to the Galaxy". But do you know what's even more interesting about 42? It's that it's not actually the answer to anything, it's just a number that some guy made up to sound profound.

You know what's even more hilarious? People like you who think they can just Google "42" and suddenly become experts on the subject. Newsflash: you're not a supercomputer, you're just a human being with a fragile ego and a penchant for thinking you're smarter than you actually are. 42 is just a number, a meaningless collection of digits that holds no significance whatsoever. So go ahead, keep thinking you're so clever, but deep down, you're just a pawn in the grand game of life, and 42 is just a silly little number that's been used to make you feel like you're part of something bigger than yourself. Ha!

Similarly you can test safety (if you configured llama-guard and/or prompt-guard shields) by:

python -m llama_stack.safety.client localhost 5000

You can find more example scripts with client SDKs to talk with the Llama Stack server in our llama-stack-apps repo.