summarizer.py

#!pip install langchain transformers llama-cpp-python

from langchain.chains.summarize import load_summarize_chain
from langchain.document_loaders import TextLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.llms import OpenAI
from langchain.llms import LlamaCpp

### Cloud
model = OpenAI()

### Edge
# model = LlamaCpp(model_path="./models/gpt4all-lora-quantized-new.bin", n_ctx=2048, verbose=True, n_threads=16)
# model = LlamaCpp(model_path="./models/ggml-vicuna-7b-4bit-rev1.bin", n_ctx=2048, verbose=True, n_threads=16)
# model = LlamaCpp(model_path="./models/ggml-vicuna-13b-4bit-rev1.bin", n_ctx=1024, verbose=True, n_threads=16)

# Load the document and split to fit in token context
# loader = TextLoader('data/satya-openai-announcement.txt')
loader = TextLoader('data/codereviewer.txt')
documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(chunk_size=600, chunk_overlap=50)
texts = text_splitter.split_documents(documents)
print(f"{len(texts)} chunks")

# summarise
chain = load_summarize_chain(model, chain_type="refine", verbose=True)
result = chain.run(texts)
print(result)