File size: 1,795 Bytes
8a41f4d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 |
from application.llm.base import BaseLLM
class HuggingFaceLLM(BaseLLM):
def __init__(self, api_key, llm_name='Arc53/DocsGPT-7B',q=False):
global hf
from langchain.llms import HuggingFacePipeline
if q:
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
tokenizer = AutoTokenizer.from_pretrained(llm_name)
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16
)
model = AutoModelForCausalLM.from_pretrained(llm_name,quantization_config=bnb_config)
else:
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
tokenizer = AutoTokenizer.from_pretrained(llm_name)
model = AutoModelForCausalLM.from_pretrained(llm_name)
pipe = pipeline(
"text-generation", model=model,
tokenizer=tokenizer, max_new_tokens=2000,
device_map="auto", eos_token_id=tokenizer.eos_token_id
)
hf = HuggingFacePipeline(pipeline=pipe)
def gen(self, model, engine, messages, stream=False, **kwargs):
context = messages[0]['content']
user_question = messages[-1]['content']
prompt = f"### Instruction \n {user_question} \n ### Context \n {context} \n ### Answer \n"
result = hf(prompt)
return result.content
def gen_stream(self, model, engine, messages, stream=True, **kwargs):
raise NotImplementedError("HuggingFaceLLM Streaming is not implemented yet.")
|