phoenix-oss/llama-stack
1
0
Fork 0
forked from phoenix-oss/llama-stack-mirror
Code Pull requests Releases Packages 2 Activity Actions

Remove "routing_table" and "routing_key" concepts for the user (#201)

Browse source 
This PR makes several core changes to the developer experience surrounding Llama Stack.

Background: PR #92 introduced the notion of "routing" to the Llama Stack. It introduces three object types: (1) models, (2) shields and (3) memory banks. Each of these objects can be associated with a distinct provider. So you can get model A to be inferenced locally while model B, C can be inference remotely (e.g.)

However, this had a few drawbacks:

you could not address the provider instances -- i.e., if you configured "meta-reference" with a given model, you could not assign an identifier to this instance which you could re-use later.
the above meant that you could not register a "routing_key" (e.g. model) dynamically and say "please use this existing provider I have already configured" for a new model.
the terms "routing_table" and "routing_key" were exposed directly to the user. in my view, this is way too much overhead for a new user (which almost everyone is.) people come to the stack wanting to do ML and encounter a completely unexpected term.
What this PR does: This PR structures the run config with only a single prominent key:

- providers
Providers are instances of configured provider types. Here's an example which shows two instances of the remote::tgi provider which are serving two different models.

providers:
  inference:
  - provider_id: foo
    provider_type: remote::tgi
    config: { ... }
  - provider_id: bar
    provider_type: remote::tgi
    config: { ... }
Secondly, the PR adds dynamic registration of { models | shields | memory_banks } to the API surface. The distribution still acts like a "routing table" (as previously) except that it asks the backing providers for a listing of these objects. For example it asks a TGI or Ollama inference adapter what models it is serving. Only the models that are being actually served can be requested by the user for inference. Otherwise, the Stack server will throw an error.

When dynamically registering these objects, you can use the provider IDs shown above. Info about providers can be obtained using the Api.inspect set of endpoints (/providers, /routes, etc.)

The above examples shows the correspondence between inference providers and models registry items. Things work similarly for the safety <=> shields and memory <=> memory_banks pairs.

Registry: This PR also makes it so that Providers need to implement additional methods for registering and listing objects. For example, each Inference provider is now expected to implement the ModelsProtocolPrivate protocol (naming is not great!) which consists of two methods

register_model
list_models
The goal is to inform the provider that a certain model needs to be supported so the provider can make any relevant backend changes if needed (or throw an error if the model cannot be supported.)

There are many other cleanups included some of which are detailed in a follow-up comment.
This commit is contained in:
Ashwin Bharambe 2024-10-10 10:24:13 -07:00 • committed by GitHub
parent 8c3010553f
commit 6bb57e72a7
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
93 changed files with 4697 additions and 4457 deletions
Download patch file Download diff file Expand all files Collapse all files

227
llama_stack/providers/adapters/inference/fireworks/fireworks.py
View file

@ -10,14 +10,19 @@ from fireworks.client import Fireworks
from llama_models.llama3.api.chat_format import ChatFormat
from llama_models.llama3.api.datatypes import Message, StopReason
from llama_models.llama3.api.datatypes import Message
from llama_models.llama3.api.tokenizer import Tokenizer
from llama_stack.providers.utils.inference.routable import RoutableProviderForModels
from llama_stack.apis.inference import * # noqa: F403
from llama_stack.providers.utils.inference.augment_messages import (
augment_messages_for_tools,
from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
from llama_stack.providers.utils.inference.openai_compat import (
get_sampling_options,
process_chat_completion_response,
process_chat_completion_stream_response,
)
from llama_stack.providers.utils.inference.prompt_adapter import (
chat_completion_request_to_prompt,
)
from .config import FireworksImplConfig
@ -27,21 +32,18 @@ FIREWORKS_SUPPORTED_MODELS = {
"Llama3.1-8B-Instruct": "fireworks/llama-v3p1-8b-instruct",
"Llama3.1-70B-Instruct": "fireworks/llama-v3p1-70b-instruct",
"Llama3.1-405B-Instruct": "fireworks/llama-v3p1-405b-instruct",
"Llama3.2-1B-Instruct": "fireworks/llama-v3p2-1b-instruct",
"Llama3.2-3B-Instruct": "fireworks/llama-v3p2-3b-instruct",
}
class FireworksInferenceAdapter(Inference, RoutableProviderForModels):
class FireworksInferenceAdapter(ModelRegistryHelper, Inference):
def __init__(self, config: FireworksImplConfig) -> None:
RoutableProviderForModels.__init__(
ModelRegistryHelper.__init__(
self, stack_to_provider_models_map=FIREWORKS_SUPPORTED_MODELS
)
self.config = config
tokenizer = Tokenizer.get_instance()
self.formatter = ChatFormat(tokenizer)
@property
def client(self) -> Fireworks:
return Fireworks(api_key=self.config.api_key)
self.formatter = ChatFormat(Tokenizer.get_instance())
async def initialize(self) -> None:
return
@ -49,7 +51,7 @@ class FireworksInferenceAdapter(Inference, RoutableProviderForModels):
async def shutdown(self) -> None:
pass
async def completion(
def completion(
self,
model: str,
content: InterleavedTextMedia,
@ -59,27 +61,7 @@ class FireworksInferenceAdapter(Inference, RoutableProviderForModels):
) -> AsyncGenerator:
raise NotImplementedError()
def _messages_to_fireworks_messages(self, messages: list[Message]) -> list:
fireworks_messages = []
for message in messages:
if message.role == "ipython":
role = "tool"
else:
role = message.role
fireworks_messages.append({"role": role, "content": message.content})
return fireworks_messages
def get_fireworks_chat_options(self, request: ChatCompletionRequest) -> dict:
options = {}
if request.sampling_params is not None:
for attr in {"temperature", "top_p", "top_k", "max_tokens"}:
if getattr(request.sampling_params, attr):
options[attr] = getattr(request.sampling_params, attr)
return options
async def chat_completion(
def chat_completion(
self,
model: str,
messages: List[Message],
@ -101,147 +83,48 @@ class FireworksInferenceAdapter(Inference, RoutableProviderForModels):
logprobs=logprobs,
)
messages = augment_messages_for_tools(request)
# accumulate sampling params and other options to pass to fireworks
options = self.get_fireworks_chat_options(request)
fireworks_model = self.map_to_provider_model(request.model)
if not request.stream:
r = await self.client.chat.completions.acreate(
model=fireworks_model,
messages=self._messages_to_fireworks_messages(messages),
stream=False,
**options,
)
stop_reason = None
if r.choices[0].finish_reason:
if r.choices[0].finish_reason == "stop":
stop_reason = StopReason.end_of_turn
elif r.choices[0].finish_reason == "length":
stop_reason = StopReason.out_of_tokens
completion_message = self.formatter.decode_assistant_message_from_content(
r.choices[0].message.content, stop_reason
)
yield ChatCompletionResponse(
completion_message=completion_message,
logprobs=None,
)
client = Fireworks(api_key=self.config.api_key)
if stream:
return self._stream_chat_completion(request, client)
else:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.start,
delta="",
)
)
return self._nonstream_chat_completion(request, client)
buffer = ""
ipython = False
stop_reason = None
async def _nonstream_chat_completion(
self, request: ChatCompletionRequest, client: Fireworks
) -> ChatCompletionResponse:
params = self._get_params(request)
r = await client.completion.acreate(**params)
return process_chat_completion_response(request, r, self.formatter)
async for chunk in self.client.chat.completions.acreate(
model=fireworks_model,
messages=self._messages_to_fireworks_messages(messages),
stream=True,
**options,
):
if chunk.choices[0].finish_reason:
if stop_reason is None and chunk.choices[0].finish_reason == "stop":
stop_reason = StopReason.end_of_turn
elif (
stop_reason is None
and chunk.choices[0].finish_reason == "length"
):
stop_reason = StopReason.out_of_tokens
break
async def _stream_chat_completion(
self, request: ChatCompletionRequest, client: Fireworks
) -> AsyncGenerator:
params = self._get_params(request)
text = chunk.choices[0].delta.content
if text is None:
continue
stream = client.completion.acreate(**params)
async for chunk in process_chat_completion_stream_response(
request, stream, self.formatter
):
yield chunk
# check if its a tool call ( aka starts with <|python_tag|> )
if not ipython and text.startswith("<|python_tag|>"):
ipython = True
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=ToolCallDelta(
content="",
parse_status=ToolCallParseStatus.started,
),
)
)
buffer += text
continue
def _get_params(self, request: ChatCompletionRequest) -> dict:
prompt = chat_completion_request_to_prompt(request, self.formatter)
# Fireworks always prepends with BOS
if prompt.startswith("<|begin_of_text|>"):
prompt = prompt[len("<|begin_of_text|>") :]
if ipython:
if text == "<|eot_id|>":
stop_reason = StopReason.end_of_turn
text = ""
continue
elif text == "<|eom_id|>":
stop_reason = StopReason.end_of_message
text = ""
continue
options = get_sampling_options(request)
options.setdefault("max_tokens", 512)
return {
"model": self.map_to_provider_model(request.model),
"prompt": prompt,
"stream": request.stream,
**options,
}
buffer += text
delta = ToolCallDelta(
content=text,
parse_status=ToolCallParseStatus.in_progress,
)
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=delta,
stop_reason=stop_reason,
)
)
else:
buffer += text
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=text,
stop_reason=stop_reason,
)
)
# parse tool calls and report errors
message = self.formatter.decode_assistant_message_from_content(
buffer, stop_reason
)
parsed_tool_calls = len(message.tool_calls) > 0
if ipython and not parsed_tool_calls:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=ToolCallDelta(
content="",
parse_status=ToolCallParseStatus.failure,
),
stop_reason=stop_reason,
)
)
for tool_call in message.tool_calls:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
delta=ToolCallDelta(
content=tool_call,
parse_status=ToolCallParseStatus.success,
),
stop_reason=stop_reason,
)
)
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.complete,
delta="",
stop_reason=stop_reason,
)
)
async def embeddings(
self,
model: str,
contents: List[InterleavedTextMedia],
) -> EmbeddingsResponse:
raise NotImplementedError()