Spaces:
Sleeping
Sleeping
File size: 3,740 Bytes
19f4fce 6f97473 19f4fce 183c15c 19f4fce 183c15c 19f4fce 183c15c 19f4fce 183c15c 19f4fce 183c15c 13c5bb4 183c15c 19f4fce 183c15c 19f4fce 183c15c 19f4fce 183c15c 21fbffd 13c5bb4 183c15c 19f4fce 6f97473 6f6f650 13c5bb4 6f97473 183c15c 19f4fce ad2d0d7 19f4fce ad2d0d7 19f4fce 4335fe0 19f4fce 183c15c |
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 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 |
import os
import pickle
from json import dumps, loads
from typing import Any, List, Mapping, Optional
import numpy as np
import openai
import streamlit as st
import pandas as pd
from dotenv import load_dotenv
from huggingface_hub import HfFileSystem
from langchain.llms.base import LLM
from llama_index import (
Document,
GPTVectorStoreIndex,
LLMPredictor,
PromptHelper,
ServiceContext,
SimpleDirectoryReader,
StorageContext,
load_index_from_storage,
)
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
# from utils.customLLM import CustomLLM
load_dotenv()
# openai.api_key = os.getenv("OPENAI_API_KEY")
fs = HfFileSystem()
# define prompt helper
# set maximum input size
CONTEXT_WINDOW = 2048
# set number of output tokens
NUM_OUTPUT = 525
# set maximum chunk overlap
CHUNK_OVERLAP_RATION = 0.2
llm_model_name = "bigscience/bloom-560m"
tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
model = AutoModelForCausalLM.from_pretrained(llm_model_name, config="T5Config")
model_pipeline = pipeline(
model=model,
tokenizer=tokenizer,
task="text-generation",
# device=0, # GPU device number
# max_length=512,
do_sample=True,
top_p=0.95,
top_k=50,
temperature=0.7,
)
class CustomLLM(LLM):
pipeline = model_pipeline
def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str:
prompt_length = len(prompt)
response = self.pipeline(prompt, max_new_tokens=525)[0]["generated_text"]
# only return newly generated tokens
return response[prompt_length:]
@property
def _identifying_params(self) -> Mapping[str, Any]:
return {"name_of_model": llm_model_name}
@property
def _llm_type(self) -> str:
return "custom"
class LlamaCustom:
def __init__(self, model_name: str) -> None:
self.vector_index = self.initialize_index(model_name=model_name)
def initialize_index(self, model_name: str):
index_name = model_name.split("/")[-1]
file_path = f"./vectorStores/{index_name}"
if os.path.exists(path=file_path):
# rebuild storage context
storage_context = StorageContext.from_defaults(persist_dir=file_path)
# local load index access
index = load_index_from_storage(storage_context)
# huggingface repo load access
# with fs.open(file_path, "r") as file:
# index = pickle.loads(file.readlines())
return index
else:
# define llm
prompt_helper = PromptHelper(
context_window=CONTEXT_WINDOW,
num_output=NUM_OUTPUT,
chunk_overlap_ratio=CHUNK_OVERLAP_RATION,
)
llm_predictor = LLMPredictor(llm=CustomLLM())
service_context = ServiceContext.from_defaults(
llm_predictor=llm_predictor, prompt_helper=prompt_helper
)
# documents = prepare_data(r"./assets/regItems.json")
documents = SimpleDirectoryReader(input_dir="./assets/pdf").load_data()
index = GPTVectorStoreIndex.from_documents(
documents, service_context=service_context
)
# local write access
index.storage_context.persist(file_path)
# huggingface repo write access
# with fs.open(file_path, "w") as file:
# file.write(pickle.dumps(index))
return index
def get_response(self, query_str):
print("query_str: ", query_str)
query_engine = self.vector_index.as_query_engine()
response = query_engine.query(query_str)
return str(response) |