LLM/DocsGPT
1
0
Fork 0
mirror of https://github.com/arc53/DocsGPT.git synced 2025-11-30 00:53:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: pytest issues

Browse Source
This commit is contained in:
Siddhant Rai
2024-04-15 19:35:59 +05:30
parent d71c74c6fb
commit c1c69ed22b
6 changed files with 155 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
application/llm/huggingface.py
View File

@@ -1,44 +1,57 @@
from application.llm.base import BaseLLM
class HuggingFaceLLM(BaseLLM):
def __init__(self, api_key, llm_name='Arc53/DocsGPT-7B',q=False):
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
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)
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
"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, messages, stream=False, **kwargs):
context = messages[0]['content']
user_question = messages[-1]['content']
def _raw_gen(self, model, 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, messages, stream=True, **kwargs):
def _raw_gen_stream(self, model, messages, stream=True, **kwargs):
raise NotImplementedError("HuggingFaceLLM Streaming is not implemented yet.")