my-project-public

repository

loading code, commits, and activity

repositories

loading repo index

#1	import importlib
#2	import logging
#3	import os
#4	from typing import Optional
#5
#6	from langchain_community.llms.huggingface_endpoint import HuggingFaceEndpoint
#7	from langchain_community.llms.huggingface_hub import HuggingFaceHub
#8	from langchain_community.llms.huggingface_pipeline import HuggingFacePipeline
#9
#10	from embedchain.config import BaseLlmConfig
#11	from embedchain.helpers.json_serializable import register_deserializable
#12	from embedchain.llm.base import BaseLlm
#13
#14	logger = logging.getLogger(__name__)
#15
#16
#17	@register_deserializable
#18	class HuggingFaceLlm(BaseLlm):
#19	def __init__(self, config: Optional[BaseLlmConfig] = None):
#20	try:
#21	importlib.import_module("huggingface_hub")
#22	except ModuleNotFoundError:
#23	raise ModuleNotFoundError(
#24	"The required dependencies for HuggingFaceHub are not installed."
#25	"Please install with `pip install huggingface-hub==0.23.0`"
#26	) from None
#27
#28	super().__init__(config=config)
#29	if not self.config.api_key and "HUGGINGFACE_ACCESS_TOKEN" not in os.environ:
#30	raise ValueError("Please set the HUGGINGFACE_ACCESS_TOKEN environment variable or pass it in the config.")
#31
#32	def get_llm_model_answer(self, prompt):
#33	if self.config.system_prompt:
#34	raise ValueError("HuggingFaceLlm does not support `system_prompt`")
#35	return HuggingFaceLlm._get_answer(prompt=prompt, config=self.config)
#36
#37	@staticmethod
#38	def _get_answer(prompt: str, config: BaseLlmConfig) -> str:
#39	# If the user wants to run the model locally, they can do so by setting the `local` flag to True
#40	if config.model and config.local:
#41	return HuggingFaceLlm._from_pipeline(prompt=prompt, config=config)
#42	elif config.model:
#43	return HuggingFaceLlm._from_model(prompt=prompt, config=config)
#44	elif config.endpoint:
#45	return HuggingFaceLlm._from_endpoint(prompt=prompt, config=config)
#46	else:
#47	raise ValueError("Either `model` or `endpoint` must be set in config")
#48
#49	@staticmethod
#50	def _from_model(prompt: str, config: BaseLlmConfig) -> str:
#51	model_kwargs = {
#52	"temperature": config.temperature or 0.1,
#53	"max_new_tokens": config.max_tokens,
#54	}
#55
#56	if 0.0 < config.top_p < 1.0:
#57	model_kwargs["top_p"] = config.top_p
#58	else:
#59	raise ValueError("`top_p` must be > 0.0 and < 1.0")
#60
#61	model = config.model
#62	api_key = config.api_key or os.getenv("HUGGINGFACE_ACCESS_TOKEN")
#63	logger.info(f"Using HuggingFaceHub with model {model}")
#64	llm = HuggingFaceHub(
#65	huggingfacehub_api_token=api_key,
#66	repo_id=model,
#67	model_kwargs=model_kwargs,
#68	)
#69	return llm.invoke(prompt)
#70
#71	@staticmethod
#72	def _from_endpoint(prompt: str, config: BaseLlmConfig) -> str:
#73	api_key = config.api_key or os.getenv("HUGGINGFACE_ACCESS_TOKEN")
#74	llm = HuggingFaceEndpoint(
#75	huggingfacehub_api_token=api_key,
#76	endpoint_url=config.endpoint,
#77	task="text-generation",
#78	model_kwargs=config.model_kwargs,
#79	)
#80	return llm.invoke(prompt)
#81
#82	@staticmethod
#83	def _from_pipeline(prompt: str, config: BaseLlmConfig) -> str:
#84	model_kwargs = {
#85	"temperature": config.temperature or 0.1,
#86	"max_new_tokens": config.max_tokens,
#87	}
#88
#89	if 0.0 < config.top_p < 1.0:
#90	model_kwargs["top_p"] = config.top_p
#91	else:
#92	raise ValueError("`top_p` must be > 0.0 and < 1.0")
#93
#94	llm = HuggingFacePipeline.from_model_id(
#95	model_id=config.model,
#96	task="text-generation",
#97	pipeline_kwargs=model_kwargs,
#98	)
#99	return llm.invoke(prompt)
#100

z6Mkq5mY3JWtxoxUobWcfNHm7AkRubgSWEZTkBVqZXJviFZ5/my-project-public