Skip to content

Commit

Permalink
unpin huggingface hub deps (#17433)
Browse files Browse the repository at this point in the history
  • Loading branch information
@logan-markewich
logan-markewich authored Jan 6, 2025
1 parent 2ea7c51 commit 1209ccc
Show file tree
Hide file tree
Showing 10 changed files with 92 additions and 102 deletions.
4 changes: 2 additions & 2 deletions llama-index-integrations/extractors/llama-index-extractors-entity/pyproject.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,12 +27,12 @@ exclude = ["**/BUILD"]
license = "MIT"
name = "llama-index-extractors-entity"
readme = "README.md"
version = "0.3.0"
version = "0.3.1"

[tool.poetry.dependencies]
python = ">=3.9,<4.0"
span-marker = ">=1.5.0"
huggingface-hub = "<0.24.0"
huggingface-hub = ">=0.23.0"
llama-index-core = "^0.12.0"

[tool.poetry.group.dev.dependencies]
Expand Down
4 changes: 2 additions & 2 deletions llama-index-integrations/extractors/llama-index-extractors-relik/pyproject.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,12 +27,12 @@ exclude = ["**/BUILD"]
license = "MIT"
name = "llama-index-extractors-relik"
readme = "README.md"
version = "0.3.0"
version = "0.3.1"

[tool.poetry.dependencies]
python = ">=3.10,<3.12"
relik = "^1.0.3"
huggingface-hub = "<0.24.0" # TODO: relik breaks on newer versions
huggingface-hub = ">=0.23.0"
spacy = "*"
llama-index-core = "^0.12.0"

Expand Down
4 changes: 2 additions & 2 deletions llama-index-integrations/llms/llama-index-llms-gaudi/pyproject.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,11 +30,11 @@ license = "MIT"
name = "llama-index-llms-gaudi"
packages = [{include = "llama_index/"}]
readme = "README.md"
version = "0.2.0"
version = "0.2.1"

[tool.poetry.dependencies]
python = ">=3.9,<4.0"
huggingface-hub = "^0.23.0"
huggingface-hub = ">=0.23.0"
torch = "^2.1.2"
text-generation = "^0.7.0"
llama-index-core = "^0.12.0"
Expand Down
63 changes: 30 additions & 33 deletions ...tegrations/llms/llama-index-llms-huggingface-api/llama_index/llms/huggingface_api/base.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,6 @@

from huggingface_hub import AsyncInferenceClient, InferenceClient, model_info
from huggingface_hub.hf_api import ModelInfo
from huggingface_hub.inference._types import ConversationalOutput
from llama_index.core.base.llms.types import (
ChatMessage,
ChatResponse,
Expand All @@ -26,34 +25,6 @@
logger = logging.getLogger(__name__)


def chat_messages_to_conversational_kwargs(
messages: Sequence[ChatMessage],
) -> Dict[str, Any]:
"""Convert ChatMessages to keyword arguments for Inference API conversational."""
if len(messages) % 2 != 1:
raise NotImplementedError("Messages passed in must be of odd length.")
last_message = messages[-1]
kwargs: Dict[str, Any] = {
"text": last_message.content,
**last_message.additional_kwargs,
}
if len(messages) != 1:
kwargs["past_user_inputs"] = []
kwargs["generated_responses"] = []
for user_msg, assistant_msg in zip(messages[::2], messages[1::2]):
if (
user_msg.role != MessageRole.USER
or assistant_msg.role != MessageRole.ASSISTANT
):
raise NotImplementedError(
"Didn't handle when messages aren't ordered in alternating"
f" pairs of {(MessageRole.USER, MessageRole.ASSISTANT)}."
)
kwargs["past_user_inputs"].append(user_msg.content)
kwargs["generated_responses"].append(assistant_msg.content)
return kwargs


class HuggingFaceInferenceAPI(CustomLLM):
"""
Wrapper on the Hugging Face's Inference API.
Expand Down Expand Up @@ -225,12 +196,17 @@ def metadata(self) -> LLMMetadata:
def chat(self, messages: Sequence[ChatMessage], **kwargs: Any) -> ChatResponse:
# default to conversational task as that was the previous functionality
if self.task == "conversational" or self.task is None:
output: "ConversationalOutput" = self._sync_client.conversational(
**{**chat_messages_to_conversational_kwargs(messages), **kwargs}
output = self._sync_client.chat_completion(
messages=[
{"role": m.role.value, "content": m.content} for m in messages
],
model=self.model_name,
**kwargs,
)
return ChatResponse(
message=ChatMessage(
role=MessageRole.ASSISTANT, content=output["generated_text"]
role=MessageRole.ASSISTANT,
content=output["choices"][0]["message"]["content"] or "",
)
)
else:
Expand Down Expand Up @@ -276,7 +252,28 @@ async def acomplete(
async def astream_chat(
self, messages: Sequence[ChatMessage], **kwargs: Any
) -> ChatResponseAsyncGen:
raise NotImplementedError
# default to conversational task as that was the previous functionality
if self.task == "conversational" or self.task is None:
output = await self._async_client.chat_completion(
messages=[
{"role": m.role.value, "content": m.content} for m in messages
],
model=self.model_name,
**kwargs,
)
return ChatResponse(
message=ChatMessage(
role=MessageRole.ASSISTANT,
content=output["choices"][0]["message"]["content"] or "",
)
)
else:
# try and use text generation
prompt = self.messages_to_prompt(messages)
completion = await self.acomplete(prompt)
return ChatResponse(
message=ChatMessage(role=MessageRole.ASSISTANT, content=completion.text)
)

async def astream_complete(
self, prompt: str, formatted: bool = False, **kwargs: Any
Expand Down
4 changes: 2 additions & 2 deletions llama-index-integrations/llms/llama-index-llms-huggingface-api/pyproject.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,11 +27,11 @@ exclude = ["**/BUILD"]
license = "MIT"
name = "llama-index-llms-huggingface-api"
readme = "README.md"
version = "0.3.0"
version = "0.3.1"

[tool.poetry.dependencies]
python = ">=3.9,<4.0"
huggingface-hub = "^0.23.0"
huggingface-hub = ">=0.23.0"
llama-index-core = "^0.12.0"

[tool.poetry.group.dev.dependencies]
Expand Down
22 changes: 10 additions & 12 deletions llama-index-integrations/llms/llama-index-llms-huggingface-api/tests/test_huggingface_api.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,29 +46,27 @@ def test_chat(self, hf_inference_api: HuggingFaceInferenceAPI) -> None:
" It's based on the book of the same name by James Fenimore Cooper."
)
conversational_return = {
"generated_text": generated_response,
"conversation": {
"generated_responses": ["It's Die Hard for sure.", generated_response],
"past_user_inputs": [
"Which movie is the best?",
"Can you explain why?",
],
},
"choices": [
{
"message": {
"content": generated_response,
}
}
],
}

with patch.object(
hf_inference_api._sync_client,
"conversational",
"chat_completion",
return_value=conversational_return,
) as mock_conversational:
response = hf_inference_api.chat(messages=messages)

assert response.message.role == MessageRole.ASSISTANT
assert response.message.content == generated_response
mock_conversational.assert_called_once_with(
text="Can you explain why?",
past_user_inputs=["Which movie is the best?"],
generated_responses=["It's Die Hard for sure."],
messages=[{"role": m.role.value, "content": m.content} for m in messages],
model=STUB_MODEL_NAME,
)

def test_chat_text_generation(
Expand Down
63 changes: 30 additions & 33 deletions ...index-integrations/llms/llama-index-llms-huggingface/llama_index/llms/huggingface/base.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,6 @@
import torch
from huggingface_hub import AsyncInferenceClient, InferenceClient, model_info
from huggingface_hub.hf_api import ModelInfo
from huggingface_hub.inference._types import ConversationalOutput
from llama_index.core.base.llms.types import (
ChatMessage,
ChatResponse,
Expand Down Expand Up @@ -434,34 +433,6 @@ def stream_chat(
return stream_completion_response_to_chat_response(completion_response)


def chat_messages_to_conversational_kwargs(
messages: Sequence[ChatMessage],
) -> Dict[str, Any]:
"""Convert ChatMessages to keyword arguments for Inference API conversational."""
if len(messages) % 2 != 1:
raise NotImplementedError("Messages passed in must be of odd length.")
last_message = messages[-1]
kwargs: Dict[str, Any] = {
"text": last_message.content,
**last_message.additional_kwargs,
}
if len(messages) != 1:
kwargs["past_user_inputs"] = []
kwargs["generated_responses"] = []
for user_msg, assistant_msg in zip(messages[::2], messages[1::2]):
if (
user_msg.role != MessageRole.USER
or assistant_msg.role != MessageRole.ASSISTANT
):
raise NotImplementedError(
"Didn't handle when messages aren't ordered in alternating"
f" pairs of {(MessageRole.USER, MessageRole.ASSISTANT)}."
)
kwargs["past_user_inputs"].append(user_msg.content)
kwargs["generated_responses"].append(assistant_msg.content)
return kwargs


@deprecated(
"Deprecated in favor of `HuggingFaceInferenceAPI` from `llama-index-llms-huggingface-api` which should be used instead.",
action="always",
Expand Down Expand Up @@ -637,12 +608,17 @@ def metadata(self) -> LLMMetadata:
def chat(self, messages: Sequence[ChatMessage], **kwargs: Any) -> ChatResponse:
# default to conversational task as that was the previous functionality
if self.task == "conversational" or self.task is None:
output: "ConversationalOutput" = self._sync_client.conversational(
**{**chat_messages_to_conversational_kwargs(messages), **kwargs}
output = self._sync_client.chat_completion(
messages=[
{"role": m.role.value, "content": m.content} for m in messages
],
model=self.model_name,
**kwargs,
)
return ChatResponse(
message=ChatMessage(
role=MessageRole.ASSISTANT, content=output["generated_text"]
role=MessageRole.ASSISTANT,
content=output["choices"][0]["message"]["content"] or "",
)
)
else:
Expand Down Expand Up @@ -675,7 +651,28 @@ def stream_complete(
async def achat(
self, messages: Sequence[ChatMessage], **kwargs: Any
) -> ChatResponse:
raise NotImplementedError
# default to conversational task as that was the previous functionality
if self.task == "conversational" or self.task is None:
output = await self._async_client.chat_completion(
messages=[
{"role": m.role.value, "content": m.content} for m in messages
],
model=self.model_name,
**kwargs,
)
return ChatResponse(
message=ChatMessage(
role=MessageRole.ASSISTANT,
content=output["choices"][0]["message"]["content"] or "",
)
)
else:
# try and use text generation
prompt = self.messages_to_prompt(messages)
completion = await self.acomplete(prompt)
return ChatResponse(
message=ChatMessage(role=MessageRole.ASSISTANT, content=completion.text)
)

async def acomplete(
self, prompt: str, formatted: bool = False, **kwargs: Any
Expand Down
4 changes: 2 additions & 2 deletions llama-index-integrations/llms/llama-index-llms-huggingface/pyproject.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -28,11 +28,11 @@ exclude = ["**/BUILD"]
license = "MIT"
name = "llama-index-llms-huggingface"
readme = "README.md"
version = "0.4.1"
version = "0.4.2"

[tool.poetry.dependencies]
python = ">=3.9,<4.0"
huggingface-hub = "^0.23.0"
huggingface-hub = ">=0.23.0"
torch = "^2.1.2"
text-generation = "^0.7.0"
llama-index-core = "^0.12.0"
Expand Down
22 changes: 10 additions & 12 deletions llama-index-integrations/llms/llama-index-llms-huggingface/tests/test_huggingface.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,29 +46,27 @@ def test_chat(self, hf_inference_api: HuggingFaceInferenceAPI) -> None:
" It's based on the book of the same name by James Fenimore Cooper."
)
conversational_return = {
"generated_text": generated_response,
"conversation": {
"generated_responses": ["It's Die Hard for sure.", generated_response],
"past_user_inputs": [
"Which movie is the best?",
"Can you explain why?",
],
},
"choices": [
{
"message": {
"content": generated_response,
}
}
]
}

with patch.object(
hf_inference_api._sync_client,
"conversational",
"chat_completion",
return_value=conversational_return,
) as mock_conversational:
response = hf_inference_api.chat(messages=messages)

assert response.message.role == MessageRole.ASSISTANT
assert response.message.content == generated_response
mock_conversational.assert_called_once_with(
text="Can you explain why?",
past_user_inputs=["Which movie is the best?"],
generated_responses=["It's Die Hard for sure."],
messages=[{"role": m.role.value, "content": m.content} for m in messages],
model=STUB_MODEL_NAME,
)

def test_chat_text_generation(
Expand Down
4 changes: 2 additions & 2 deletions llama-index-integrations/readers/llama-index-readers-huggingface-fs/pyproject.toml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,11 @@ license = "MIT"
maintainers = ["jerryjliu"]
name = "llama-index-readers-huggingface-fs"
readme = "README.md"
version = "0.3.0"
version = "0.3.1"

[tool.poetry.dependencies]
python = ">=3.9,<4.0"
huggingface-hub = "^0.20.3"
huggingface-hub = ">=0.20.3"
pandas = "*"
llama-index-core = "^0.12.0"

Expand Down

0 comments on commit 1209ccc

Please sign in to comment.