phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-07-29 15:23:51 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Use huggingface_hub inference client for TGI inference

Browse source
This commit is contained in:
Celina Hanouti 2024-09-05 18:29:04 +02:00
parent 21bedc1596
commit e5bcfdac21
6 changed files with 179 additions and 142 deletions
Download patch file Download diff file Expand all files Collapse all files

21
docs/cli_reference.md
View file

@ -248,44 +248,51 @@ llama stack list-distributions
```
<pre style="font-family: monospace;">
i+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| Distribution ID | Providers | Description |
+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| local | { | Use code from `llama_toolchain` itself to serve all llama stack APIs |
| | "inference": "meta-reference", | |
| | "memory": "meta-reference-faiss", | |
| | "safety": "meta-reference", | |
| | "agentic_system": "meta-reference" | |
| | } | |
+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| remote | { | Point to remote services for all llama stack APIs |
| | "inference": "remote", | |
| | "safety": "remote", | |
| | "agentic_system": "remote", | |
| | "memory": "remote" | |
| | } | |
+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| local-ollama | { | Like local, but use ollama for running LLM inference |
| | "inference": "remote::ollama", | |
| | "safety": "meta-reference", | |
| | "agentic_system": "meta-reference", | |
| | "memory": "meta-reference-faiss" | |
| | } | |
+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| local-plus-fireworks-inference | { | Use Fireworks.ai for running LLM inference |
| | "inference": "remote::fireworks", | |
| | "safety": "meta-reference", | |
| | "agentic_system": "meta-reference", | |
| | "memory": "meta-reference-faiss" | |
| | } | |
+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| local-plus-together-inference | { | Use Together.ai for running LLM inference |
| | "inference": "remote::together", | |
| | "safety": "meta-reference", | |
| | "agentic_system": "meta-reference", | |
| | "memory": "meta-reference-faiss" | |
| | } | |
+--------------------------------+---------------------------------------+----------------------------------------------------------------------+
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
| local-plus-tgi-inference | { | Use TGI (local or with <a href="https://huggingface.co/inference-endpoints/dedicated"> |
| | "inference": "remote::tgi", | Hugging Face Inference Endpoints</a>) for running LLM inference |
| | "safety": "meta-reference", | |
| | "agentic_system": "meta-reference", | |
| | "memory": "meta-reference-faiss" | |
| | } | |
+--------------------------------+---------------------------------------+-------------------------------------------------------------------------------------------+
</pre>
As you can see above, each “distribution” details the “providers” it is composed of. For example, `local` uses the “meta-reference” provider for inference while local-ollama relies on a different provider (Ollama) for inference. Similarly, you can use Fireworks or Together.AI for running inference as well.

10
llama_toolchain/core/distribution_registry.py
View file

@ -58,6 +58,16 @@ def available_distribution_specs() -> List[DistributionSpec]:
Api.memory: "meta-reference-faiss",
},
),
DistributionSpec(
distribution_id="local-plus-tgi-inference",
description="Use TGI for running LLM inference",
providers={
Api.inference: remote_provider_id("tgi"),
Api.safety: "meta-reference",
Api.agentic_system: "meta-reference",
Api.memory: "meta-reference-faiss",
},
),
]

11
llama_toolchain/inference/adapters/tgi/__init__.py
View file

@ -4,12 +4,15 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from llama_toolchain.core.datatypes import RemoteProviderConfig
from .config import TGIImplConfig
async def get_adapter_impl(config: RemoteProviderConfig, _deps):
from .tgi import TGIInferenceAdapter
async def get_adapter_impl(config: TGIImplConfig, _deps):
from .tgi import TGIAdapter
impl = TGIInferenceAdapter(config.url)
assert isinstance(
config, TGIImplConfig
), f"Unexpected config type: {type(config)}"
impl = TGIAdapter(config)
await impl.initialize()
return impl

22
llama_toolchain/inference/adapters/tgi/config.py Normal file
View file

@ -0,0 +1,22 @@
# 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 Optional
from llama_models.schema_utils import json_schema_type
from pydantic import BaseModel, Field, field_validator
@json_schema_type
class TGIImplConfig(BaseModel):
url: str = Field(
default="https://api-inference.huggingface.co",
description="The URL for the TGI endpoint",
)
api_token: Optional[str] = Field(
default="",
description="The HF token for Hugging Face Inference Endpoints",
)

184
llama_toolchain/inference/adapters/tgi/tgi.py
View file

@ -4,63 +4,44 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from typing import AsyncGenerator, List
import httpx
from typing import AsyncGenerator
from huggingface_hub import InferenceClient
from llama_models.llama3.api.chat_format import ChatFormat
from llama_models.llama3.api.datatypes import Message, StopReason
from llama_models.llama3.api.tokenizer import Tokenizer
from llama_models.sku_list import resolve_model
from text_generation import Client
from llama_toolchain.inference.api import *
from llama_toolchain.inference.api.api import ( # noqa: F403
ChatCompletionRequest,
ChatCompletionResponse,
ChatCompletionResponseStreamChunk,
)
from llama_toolchain.inference.api import * # noqa: F403
from llama_toolchain.inference.prepare_messages import prepare_messages
from .config import TGIImplConfig
SUPPORTED_MODELS = {
HF_SUPPORTED_MODELS = {
"Meta-Llama3.1-8B-Instruct": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"Meta-Llama3.1-70B-Instruct": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"Meta-Llama3.1-405B-Instruct": "meta-llama/Meta-Llama-3.1-405B-Instruct",
}
class TGIInferenceAdapter(Inference):
def __init__(self, url: str) -> None:
self.url = url.rstrip("/")
class TGIAdapter(Inference):
def __init__(self, config: TGIImplConfig) -> None:
self.config = config
self.tokenizer = Tokenizer.get_instance()
self.formatter = ChatFormat(self.tokenizer)
self.model = None
self.max_tokens = None
@property
def client(self) -> InferenceClient:
return InferenceClient(base_url=self.config.url, token=self.config.api_token)
async def initialize(self) -> None:
hf_models = {v: k for k, v in SUPPORTED_MODELS.items()}
try:
print(f"Connecting to TGI server at: {self.url}")
async with httpx.AsyncClient() as client:
response = await client.get(f"{self.url}/info")
response.raise_for_status()
info = response.json()
if "model_id" not in info:
raise RuntimeError("Missing model_id in model info")
if "max_total_tokens" not in info:
raise RuntimeError("Missing max_total_tokens in model info")
self.max_tokens = info["max_total_tokens"]
model_id = info["model_id"]
if model_id not in hf_models:
raise RuntimeError(
f"TGI is serving model: {model_id}, use one of the supported models: {','.join(hf_models.keys())}"
)
self.model = hf_models[model_id]
except Exception as e:
import traceback
traceback.print_exc()
raise RuntimeError("Could not connect to TGI server") from e
pass
async def shutdown(self) -> None:
pass
@ -68,15 +49,25 @@ class TGIInferenceAdapter(Inference):
async def completion(self, request: CompletionRequest) -> AsyncGenerator:
raise NotImplementedError()
def _convert_messages(self, messages: List[Message]) -> List[Message]:
ret = []
def _convert_messages(self, messages: list[Message]) -> List[Message]: # type: ignore
tgi_messages = []
for message in messages:
if message.role == "ipython":
role = "tool"
else:
role = message.role
ret.append({"role": role, "content": message.content})
return ret
tgi_messages.append({"role": role, "content": message.content})
return tgi_messages
def resolve_hf_model(self, model_name: str) -> str:
model = resolve_model(model_name)
assert (
model is not None
and model.descriptor(shorten_default_variant=True) in HF_SUPPORTED_MODELS
), f"Unsupported model: {model_name}, use one of the supported models: {','.join(HF_SUPPORTED_MODELS.keys())}"
return HF_SUPPORTED_MODELS.get(model.descriptor(shorten_default_variant=True))
def get_chat_options(self, request: ChatCompletionRequest) -> dict:
options = {}
@ -88,48 +79,34 @@ class TGIInferenceAdapter(Inference):
return options
async def chat_completion(self, request: ChatCompletionRequest) -> AsyncGenerator:
messages = prepare_messages(request)
model_input = self.formatter.encode_dialog_prompt(messages)
prompt = self.tokenizer.decode(model_input.tokens)
max_new_tokens = min(
request.sampling_params.max_tokens or self.max_tokens,
self.max_tokens - len(model_input.tokens) - 1,
)
if request.model != self.model:
raise ValueError(
f"Model mismatch, expected: {self.model}, got: {request.model}"
)
options = self.get_chat_options(request)
messages = self._convert_messages(request.messages)
client = Client(base_url=self.url)
if not request.stream:
r = client.generate(
prompt,
max_new_tokens=max_new_tokens,
stop_sequences=["<|eom_id|>", "<|eot_id|>"],
response = self.client.chat_completion(
messages=messages,
stream=False,
**options,
)
if r.details.finish_reason:
if r.details.finish_reason == "stop":
stop_reason = None
if response.choices[0].finish_reason:
if (
response.choices[0].finish_reason == "stop_sequence"
or response.choices[0].finish_reason == "eos_token"
):
stop_reason = StopReason.end_of_turn
elif r.details.finish_reason == "length":
stop_reason = StopReason.out_of_tokens
else:
stop_reason = StopReason.end_of_message
else:
elif response.choices[0].finish_reason == "length":
stop_reason = StopReason.out_of_tokens
completion_message = self.formatter.decode_assistant_message_from_content(
r.generated_text, stop_reason
response.choices[0].message.content,
stop_reason,
)
yield ChatCompletionResponse(
completion_message=completion_message,
logprobs=None,
)
else:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
@ -137,24 +114,35 @@ class TGIInferenceAdapter(Inference):
delta="",
)
)
buffer = ""
ipython = False
stop_reason = None
tokens = []
for response in client.generate_stream(
prompt,
max_new_tokens=max_new_tokens,
stop_sequences=["<|eom_id|>", "<|eot_id|>"],
**options,
for chunk in self.client.chat_completion(
messages=messages, stream=True, **options
):
token_result = response.token
if chunk.choices[0].finish_reason:
if (
stop_reason is None
and chunk.choices[0].finish_reason == "stop_sequence"
) or (
stop_reason is None
and chunk.choices[0].finish_reason == "eos_token"
):
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
buffer += token_result.text
tokens.append(token_result.id)
text = chunk.choices[0].delta.content
if text is None:
continue
if not ipython and buffer.startswith("<|python_tag|>"):
# 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(
@ -165,27 +153,25 @@ class TGIInferenceAdapter(Inference):
),
)
)
buffer = buffer[len("<|python_tag|>") :]
buffer += text
continue
if token_result.text == "<|eot_id|>":
if ipython:
if text == "<|eot_id|>":
stop_reason = StopReason.end_of_turn
text = ""
elif token_result.text == "<|eom_id|>":
continue
elif text == "<|eom_id|>":
stop_reason = StopReason.end_of_message
text = ""
else:
text = token_result.text
continue
if ipython:
buffer += text
delta = ToolCallDelta(
content=text,
parse_status=ToolCallParseStatus.in_progress,
)
else:
delta = text
if stop_reason is None:
yield ChatCompletionResponseStreamChunk(
event=ChatCompletionResponseEvent(
event_type=ChatCompletionResponseEventType.progress,
@ -193,12 +179,20 @@ class TGIInferenceAdapter(Inference):
stop_reason=stop_reason,
)
)
if stop_reason is None:
stop_reason = StopReason.out_of_tokens
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(tokens, stop_reason)
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(

3
llama_toolchain/inference/providers.py
View file

@ -39,8 +39,9 @@ def available_inference_providers() -> List[ProviderSpec]:
api=Api.inference,
adapter=AdapterSpec(
adapter_id="tgi",
pip_packages=["text-generation"],
pip_packages=["huggingface_hub"],
module="llama_toolchain.inference.adapters.tgi",
config_class="llama_toolchain.inference.adapters.tgi.TGIImplConfig",
),
),
remote_provider_spec(