phi3-7b-chess-beta / README.md
bhuvanmdev's picture
Update README.md
022b9e0 verified
metadata
language:
  - en
license: mit
tags:
  - game
  - experimetal
  - chess
datasets:
  - bhuvanmdev/chess-causal-formatted

Experimental Chess Model (Causal)

Overview

This model is an experimental fine-tuned variant designed for causal inference on a very small subset of chess games. It leverages the base model obtained from Microsoft(phi-3-mini-4k-instruct) and has been fine-tuned using Hugging Face Transformers with the Accelerate library.

Key Details

  • Task: Causal inference on chess games
  • Base Model: phi-3-mini-4k-instruct
  • Fine-Tuning Framework: Hugging Face Transformers with Accelerate and peft
  • License: MIT

Description

The primary purpose of this model is to explore causal relationships within chess games. It was trained on a limited dataset, making it suitable for experimentation and research. While its performance may not match larger-scale models, it serves as a starting point for causal analysis in the chess games. It also gives us an insight on how causal models react to high level chess games (2000> ELO).

Limitations

  • Small Dataset: Due to the limited data, the model's generalization capabilities are restricted.
  • Experimental Nature: This model is not production-ready and should be used for research purposes only.
  • Causal Interpretation: Interpretation of causal effects requires careful consideration and domain expertise.

Usage

will be updated shortly !!!

Metrics

global_step=2795, training_loss=0.15753029228749557, metrics={'train_runtime': 7548.9262, 'train_samples_per_second': 0.37, 'train_steps_per_second': 0.37, 'total_flos': 4.255669870466458e+16, 'train_loss': 0.15753029228749557, 'epoch': 1.0, 'num_input_tokens_seen': 1892547} will be updated shortly !!!

Author

The main authors of the base model can be found Here

Consider having a read at the original model card to understand the biases,limitations and other necessary details.

It's one of my first systematically fine-tuned model, Feel free to experiment with this model and contribute to its development! ;) THANK YOU