JILM

Jilm enhances your AI capabilities.

Installation

First download a GPTj model, for instance ggml-gpt4all-j-v1.3-groovy.bin that can be installed here for instance.

Then copy the .env.template file into .env and modify it. Only the line with the model path should be modified:

JILM_LLM_MODEL_PATH="/path/to/ggml-gpt4all-j-v1.3-groovy.bin"

To install from the github repo

git clone https://github.com/jpoullet2000/jilm.git
pip install poetry
poetry install

Get started

Start using JILM

doc = DocumentLoader.load_single_document(data_folder / "paul_graham_essays_worked.txt")
splitter = TextSplitter()
docs = splitter.split([doc])
res = DocQuery("What does the author describe as good work?", docs=docs)
answer, docs = res.run()

The displayed result:

Question: What does the author describe as good work?
Helpful Answer: The author does not explicitly describe what constitutes good work in the given context. However, it can be inferred that good work is related to the core of the work, as it involves a significant contribution to the overall project. The author suggests that good work should at least be something close to the core of the work. The lack of specific examples in the given context may suggest that it is a difficult or subjective matter.<|endoftext|>

This is not limited to txt files. Suppose, you want to ask questions to a book in a pdf format.

import requests

# Download the article "Talking About Large Language Models"
url = 'https://arxiv.org/pdf/2212.03551.pdf'
response = requests.get(url)
with open('article.pdf', 'wb') as f:
    f.write(response.content)

# Query it
doc = DocumentLoader.load_single_document("article.pdf")
splitter = TextSplitter(chunk_overlap=100, chunk_size=1000, max_tokens=1000)
docs = splitter.split([doc])
res = DocQuery("According to this article can Language Models reason? Answer by 'yes' or 'no'.", docs=docs)
answer, docs = res.run()

Returns the answer No.