ChatGPT4All

gistfile1.txt

wget https://the-eye.eu/public/AI/models/nomic-ai/gpt4all/gpt4all-lora-quantized.bin


pip install pyllama
mkdir llama
python -m llama.download --model_size 7B --folder llama/


# https://github.com/nomic-ai/pygpt4all/pyllamacpp
# pip install pyllamacpp fails and so directly download it from github
git clone --recursive https://github.com/nomic-ai/pygpt4all/ && cd pyllamacpp
pip install .


pyllamacpp-convert-gpt4all gpt4all-lora-quantized.bin llama/tokenizer.model gpt4all-lora-q-converted.bin


GPT4ALL_MODEL_PATH = "/root/gpt4all-lora-q-converted.bin"

main.py

from langchain.llms import LlamaCpp
from langchain import PromptTemplate, LLMChain
from langchain.document_loaders import TextLoader
from langchain.indexes import VectorstoreIndexCreator
from langchain.vectorstores import Chroma
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import RetrievalQA
from langchain.embeddings import LlamaCppEmbeddings


print("running")
GPT4ALL_MODEL_PATH="/root/gpt4all-lora-q-converted.bin"

print("loading llm")
llm = LlamaCpp(model_path=GPT4ALL_MODEL_PATH,max_tokens=128)

print("creating fragment")
def search_context(src, phrase, buffer=100):
  with open(src, 'r') as f:
    txt=f.read()
    words = txt.split()
    index = words.index(phrase)
    start_index = max(0, index - buffer)
    end_index = min(len(words), index + buffer+1)
    return ' '.join(words[start_index:end_index])

fragment = './fragment.txt'
with open(fragment, 'w') as fo:
    _txt = search_context('./state_of_the_union.txt', "Ketanji")
    fo.write(_txt)


print("loading fragment")
loader = TextLoader('./fragment.txt')
documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=0)
texts = text_splitter.split_documents(documents)


print("llama embeddings")
llama_embeddings = LlamaCppEmbeddings(model_path=GPT4ALL_MODEL_PATH, n_batch=512)

persist_directory = 'db_2'

docsearch = Chroma.from_documents(documents=texts, embedding=llama_embeddings, persist_directory=persist_directory)

MIN_DOCS = 1

qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=docsearch.as_retriever(search_kwargs={"k": MIN_DOCS}))
query = "What did the president say about Ketanji Brown Jackson"

print(query)

output = qa.run(query)
print(output)

"""
index = VectorstoreIndexCreator(embedding=llama_embeddings,
                                vectorstore_kwargs={"persist_directory": "db"}
                               ).from_loaders([loader])

query = "What did the president say about Ketanji Brown Jackson"

index.query(query, llm=llm)
"""


template = """
Question: {question}
Answer: 
"""

"""
prompt = PromptTemplate(template=template, input_variables=["question"])

llm = LlamaCpp(model_path=GPT4ALL_MODEL_PATH)

llm_chain = LLMChain(prompt=prompt, llm=llm)

question = "What did the president say about Ketanji Brown Jackson"

resp = llm_chain.run(question)
print(resp)

"""

meta.py

# pip install unstructured
# wget wget https://s21.q4cdn.com/399680738/files/doc_financials/2022/q4/Meta-12.31.2022-Exhibit-99.1-FINAL.pdf
# mkdir docs
# mv *.pdf docs/
# pip install pdf2image
# pip install pytesseract
# apt-get install poppler-utils
# pip install 'git+https://github.com/facebookresearch/detectron2.git'
# pip install Pillow==9.0.0

import os
from langchain.llms import LlamaCpp
from langchain import PromptTemplate, LLMChain
from langchain.document_loaders import TextLoader
from langchain.indexes import VectorstoreIndexCreator
from langchain.vectorstores import Chroma
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains import RetrievalQA
from langchain.embeddings import LlamaCppEmbeddings
from langchain.document_loaders import UnstructuredPDFLoader
from detectron2.config import get_cfg
cfg = get_cfg()    
cfg.MODEL.DEVICE = 'gpu'

print("running")
GPT4ALL_MODEL_PATH="/root/gpt4all-lora-q-converted.bin"

print("loading llm")
llm = LlamaCpp(model_path=GPT4ALL_MODEL_PATH,max_tokens=128)


print("loading fragment")
#loader = TextLoader('./fragment.txt')
text_folder = 'docs'
#loaders = [UnstructuredPDFLoader(os.path.join(text_folder, fn)) for fn in os.listdir(text_folder)]

loader = UnstructuredPDFLoader("docs/Meta-12.31.2022-Exhibit-99.1-FINAL.pdf")


documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=512, chunk_overlap=0)
texts = text_splitter.split_documents(documents)


print("llama embeddings")
llama_embeddings = LlamaCppEmbeddings(model_path=GPT4ALL_MODEL_PATH, n_batch=512)

persist_directory = 'db_2'

docsearch = Chroma.from_documents(documents=texts, embedding=llama_embeddings, persist_directory=persist_directory)

MIN_DOCS = 1

qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=docsearch.as_retriever(search_kwargs={"k": MIN_DOCS}))
query = "How much revenue did Meta make in 2022?"

print(query)

output = qa.run(query)
print(output)