Skip to content

GEMINI.md

Latest commit

 

History

History
42 lines (28 loc) · 1.25 KB

GEMINI.md

File metadata and controls

42 lines (28 loc) · 1.25 KB
Raw

Gemini Project Configuration

About This Project

use conda venv genAI

This project is an implementation of a decoder-only Transformer model from scratch in Python. The primary goal is to generate coherent text, initially based on the WikiHow dataset.

<<<<<<< HEAD

Commands

  • Run Training: python src/train.py --data data/processed/wikihow.txt --model_path models/
  • Run Inference: python src/generate.py --model_path models/model.pt
  • Run Tests: pytest
  • Lint & Format: ruff check . && black .

Dependencies

Install dependencies using pip: pip install -r requirements.txt

Coding Style

  • Follow the PEP 8 style guide.
  • Use black for code formatting.
  • All functions and classes should have docstrings.
  • Use type hints for all function signatures.

Key Files

  • src/train.py: Main script for training the model.
  • src/model.py: Contains the Transformer model architecture.
  • src/dataset.py: Data loading and preprocessing logic. =======

Present status of the project

Trained model weight: models/transformer.pt

WIP

Prepare the model for huggingface-hub. The end user must be able to use the model ruchir99/transformer-101 with AutoModelForCausalLM.from_pretrained() function

06fb81e (restarting git)