API Updates: fleshing out RAG APIs, introduce "llama stack" CLI command (#51)

* add tools to chat completion request

* use templates for generating system prompts

* Moved ToolPromptFormat and jinja templates to llama_models.llama3.api

* <WIP> memory changes

- inlined AgenticSystemInstanceConfig so API feels more ergonomic
- renamed it to AgentConfig, AgentInstance -> Agent
- added a MemoryConfig and `memory` parameter
- added `attachments` to input and `output_attachments` to the response

- some naming changes

* InterleavedTextAttachment -> InterleavedTextMedia, introduce memory tool

* flesh out memory banks API

* agentic loop has a RAG implementation

* faiss provider implementation

* memory client works

* re-work tool definitions, fix FastAPI issues, fix tool regressions

* fix agentic_system utils

* basic RAG seems to work

* small bug fixes for inline attachments

* Refactor custom tool execution utilities

* Bug fix, show memory retrieval steps in EventLogger

* No need for api_key for Remote providers

* add special unicode character ↵ to showcase newlines in model prompt templates

* remove api.endpoints imports

* combine datatypes.py and endpoints.py into api.py

* Attachment / add TTL api

* split batch_inference from inference

* minor import fixes

* use a single impl for ChatFormat.decode_assistant_mesage

* use interleaved_text_media_as_str() utilityt

* Fix api.datatypes imports

* Add blobfile for tiktoken

* Add ToolPromptFormat to ChatFormat.encode_message so that tools are encoded properly

* templates take optional --format={json,function_tag}

* Rag Updates

* Add `api build` subcommand -- WIP

* fix

* build + run image seems to work

* <WIP> adapters

* bunch more work to make adapters work

* api build works for conda now

* ollama remote adapter works

* Several smaller fixes to make adapters work

Also, reorganized the pattern of __init__ inside providers so
configuration can stay lightweight

* llama distribution -> llama stack + containers (WIP)

* All the new CLI for api + stack work

* Make Fireworks and Together into the Adapter format

* Some quick fixes to the CLI behavior to make it consistent

* Updated README phew

* Update cli_reference.md

* llama_toolchain/distribution -> llama_toolchain/core

* Add termcolor

* update paths

* Add a log just for consistency

* chmod +x scripts

* Fix api dependencies not getting added to configuration

* missing import lol

* Delete utils.py; move to agentic system

* Support downloading of URLs for attachments for code interpreter

* Simplify and generalize `llama api build` yay

* Update `llama stack configure` to be very simple also

* Fix stack start

* Allow building an "adhoc" distribution

* Remote `llama api []` subcommands

* Fixes to llama stack commands and update docs

* Update documentation again and add error messages to llama stack start

* llama stack start -> llama stack run

* Change name of build for less confusion

* Add pyopenapi fork to the repository, update RFC assets

* Remove conflicting annotation

* Added a "--raw" option for model template printing

---------

Co-authored-by: Hardik Shah <hjshah@fb.com>
Co-authored-by: Ashwin Bharambe <ashwin@meta.com>
Co-authored-by: Dalton Flanagan <6599399+dltn@users.noreply.github.com>

llama_toolchain/safety/api/__init__.py

@@ -4,5 +4,4 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from .datatypes import *  # noqa
-from .endpoints import *  # noqa
+from .api import *  # noqa: F401 F403

llama_toolchain/safety/api/api.py

@@ -5,13 +5,12 @@
 # the root directory of this source tree.
 
 from enum import Enum
-from typing import Dict, Optional, Union
-
-from llama_models.llama3.api.datatypes import ToolParamDefinition
-from llama_models.schema_utils import json_schema_type
+from typing import Dict, List, Optional, Protocol, Union
 
+from llama_models.schema_utils import json_schema_type, webmethod
 from pydantic import BaseModel, validator
 
+from llama_models.llama3.api.datatypes import *  # noqa: F403
 from llama_toolchain.common.deployment_types import RestAPIExecutionConfig
 
 
@@ -70,3 +69,22 @@ class ShieldResponse(BaseModel):
             except ValueError:
                 return v
         return v
+
+
+@json_schema_type
+class RunShieldRequest(BaseModel):
+    messages: List[Message]
+    shields: List[ShieldDefinition]
+
+
+@json_schema_type
+class RunShieldResponse(BaseModel):
+    responses: List[ShieldResponse]
+
+
+class Safety(Protocol):
+    @webmethod(route="/safety/run_shields")
+    async def run_shields(
+        self,
+        request: RunShieldRequest,
+    ) -> RunShieldResponse: ...

llama_toolchain/safety/api/endpoints.py

@@ -1,32 +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.
-
-from .datatypes import *  # noqa: F403
-from typing import List, Protocol
-
-from llama_models.llama3.api.datatypes import Message
-
-# this dependency is annoying and we need a forked up version anyway
-from llama_models.schema_utils import webmethod
-
-
-@json_schema_type
-class RunShieldRequest(BaseModel):
-    messages: List[Message]
-    shields: List[ShieldDefinition]
-
-
-@json_schema_type
-class RunShieldResponse(BaseModel):
-    responses: List[ShieldResponse]
-
-
-class Safety(Protocol):
-    @webmethod(route="/safety/run_shields")
-    async def run_shields(
-        self,
-        request: RunShieldRequest,
-    ) -> RunShieldResponse: ...

llama_toolchain/safety/client.py

@@ -5,29 +5,32 @@
 # the root directory of this source tree.
 
 import asyncio
+import json
+
+from typing import Any
 
 import fire
 import httpx
 
 from llama_models.llama3.api.datatypes import UserMessage
+from pydantic import BaseModel
 from termcolor import cprint
 
-from .api import (
-    BuiltinShield,
-    RunShieldRequest,
-    RunShieldResponse,
-    Safety,
-    ShieldDefinition,
-)
+from llama_toolchain.core.datatypes import RemoteProviderConfig
+
+from .api import *  # noqa: F403
 
 
-async def get_client_impl(base_url: str):
-    return SafetyClient(base_url)
+async def get_client_impl(config: RemoteProviderConfig, _deps: Any) -> Safety:
+    return SafetyClient(config.url)
+
+
+def encodable_dict(d: BaseModel):
+    return json.loads(d.json())
 
 
 class SafetyClient(Safety):
     def __init__(self, base_url: str):
-        print(f"Initializing client for {base_url}")
         self.base_url = base_url
 
     async def initialize(self) -> None:
@@ -40,7 +43,9 @@ class SafetyClient(Safety):
         async with httpx.AsyncClient() as client:
             response = await client.post(
                 f"{self.base_url}/safety/run_shields",
-                data=request.json(),
+                json={
+                    "request": encodable_dict(request),
+                },
                 headers={"Content-Type": "application/json"},
                 timeout=20,
             )

llama_toolchain/safety/meta_reference/__init__.py

@@ -4,5 +4,14 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from .config import SafetyConfig  # noqa
-from .safety import get_provider_impl  # noqa
+from .config import SafetyConfig
+
+
+async def get_provider_impl(config: SafetyConfig, _deps):
+    from .safety import MetaReferenceSafetyImpl
+
+    assert isinstance(config, SafetyConfig), f"Unexpected config type: {type(config)}"
+
+    impl = MetaReferenceSafetyImpl(config)
+    await impl.initialize()
+    return impl

llama_toolchain/safety/meta_reference/safety.py

@@ -5,12 +5,10 @@
 # the root directory of this source tree.
 
 import asyncio
-from typing import Dict
 
 from llama_models.sku_list import resolve_model
 
 from llama_toolchain.common.model_utils import model_local_dir
-from llama_toolchain.distribution.datatypes import Api, ProviderSpec
 from llama_toolchain.safety.api import *  # noqa
 
 from .config import SafetyConfig
@@ -25,14 +23,6 @@ from .shields import (
 )
 
 
-async def get_provider_impl(config: SafetyConfig, _deps: Dict[Api, ProviderSpec]):
-    assert isinstance(config, SafetyConfig), f"Unexpected config type: {type(config)}"
-
-    impl = MetaReferenceSafetyImpl(config)
-    await impl.initialize()
-    return impl
-
-
 def resolve_and_get_path(model_name: str) -> str:
     model = resolve_model(model_name)
     assert model is not None, f"Could not resolve model {model_name}"

llama_toolchain/safety/meta_reference/shields/base.py

@@ -5,10 +5,10 @@
 # the root directory of this source tree.
 
 from abc import ABC, abstractmethod
-from typing import List, Union
+from typing import List
 
-from llama_models.llama3.api.datatypes import Attachment, Message
-from llama_toolchain.safety.api.datatypes import *  # noqa: F403
+from llama_models.llama3.api.datatypes import interleaved_text_media_as_str, Message
+from llama_toolchain.safety.api import *  # noqa: F403
 
 CANNED_RESPONSE_TEXT = "I can't answer that. Can I help with something else?"
 
@@ -30,18 +30,7 @@ class ShieldBase(ABC):
 
 
 def message_content_as_str(message: Message) -> str:
-    def _to_str(content: Union[str, Attachment]) -> str:
-        if isinstance(content, str):
-            return content
-        elif isinstance(content, Attachment):
-            return f"File: {str(content.url)}"
-        else:
-            raise
-
-    if isinstance(message.content, list) or isinstance(message.content, tuple):
-        return "\n".join([_to_str(c) for c in message.content])
-    else:
-        return _to_str(message.content)
+    return interleaved_text_media_as_str(message.content)
 
 
 # For shields that operate on simple strings

llama_toolchain/safety/meta_reference/shields/code_scanner.py

@@ -8,7 +8,7 @@ from codeshield.cs import CodeShield
 from termcolor import cprint
 
 from .base import ShieldResponse, TextShield
-from llama_toolchain.safety.api.datatypes import *  # noqa: F403
+from llama_toolchain.safety.api import *  # noqa: F403
 
 
 class CodeScannerShield(TextShield):

llama_toolchain/safety/meta_reference/shields/llama_guard.py

@@ -14,7 +14,7 @@ from llama_models.llama3.api.datatypes import Message, Role
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 from .base import CANNED_RESPONSE_TEXT, OnViolationAction, ShieldBase, ShieldResponse
-from llama_toolchain.safety.api.datatypes import *  # noqa: F403
+from llama_toolchain.safety.api import *  # noqa: F403
 
 SAFE_RESPONSE = "safe"
 _INSTANCE = None

llama_toolchain/safety/meta_reference/shields/prompt_guard.py

@@ -14,7 +14,7 @@ from termcolor import cprint
 from transformers import AutoModelForSequenceClassification, AutoTokenizer
 
 from .base import message_content_as_str, OnViolationAction, ShieldResponse, TextShield
-from llama_toolchain.safety.api.datatypes import *  # noqa: F403
+from llama_toolchain.safety.api import *  # noqa: F403
 
 
 class PromptGuardShield(TextShield):

llama_toolchain/safety/providers.py

@@ -6,7 +6,7 @@
 
 from typing import List
 
-from llama_toolchain.distribution.datatypes import Api, InlineProviderSpec, ProviderSpec
+from llama_toolchain.core.datatypes import Api, InlineProviderSpec, ProviderSpec
 
 
 def available_safety_providers() -> List[ProviderSpec]: