Waterhorse/chessgpt-base-v1

Chessgpt-Base-3B-v1

Chessgpt-Base-v1 is the base model of Chessgpt.

• Base Model: Chessgpt-base-v1
• Chat Version: chessgpt-chat-v1

Also, we are actively working on the development of the next-generation model, ChessGPT-V2. We welcome any contribution, especially on chess related dataset. For related matters, please contact xidong.feng.20@ucl.ac.uk.

Model Details

• Model type: Language Model
• Language(s): English
• License: Apache 2.0
• Model Description: A 2.8B parameter pretrained language model in Chess.

GPU Inference

This requires a GPU with 8GB memory.

import torch
import transformers
from transformers import AutoTokenizer, AutoModelForCausalLM

MIN_TRANSFORMERS_VERSION = '4.25.1'

# check transformers version
assert transformers.__version__ >= MIN_TRANSFORMERS_VERSION, f'Please upgrade transformers to version {MIN_TRANSFORMERS_VERSION} or higher.'

# init
tokenizer = AutoTokenizer.from_pretrained("Waterhorse/chessgpt-base-v1")
model = AutoModelForCausalLM.from_pretrained("Waterhorse/chessgpt-base-v1", torch_dtype=torch.float16)
model = model.to('cuda:0')

# infer
# Conversation between two
prompt = "Q: 1.e4 c5, what is the name of this opening?A:"

inputs = tokenizer(prompt, return_tensors='pt').to(model.device)
input_length = inputs.input_ids.shape[1]
outputs = model.generate(
    **inputs, max_new_tokens=128, do_sample=True, temperature=0.7, top_p=0.7, top_k=50, return_dict_in_generate=True,
)
token = outputs.sequences[0, input_length:]
output_str = tokenizer.decode(token)
print(output_str)

Uses

Excluded uses are described below.

Direct Use

chessgpt-base-v1 is mainly for research on large language model, especially for those research about policy learning and language modeling.

Out-of-Scope Use

chessgpt-base-v1 is a language model trained on chess related data and may not perform well for other use cases beyond chess domain.

Bias, Risks, and Limitations

Just as with any language model, chessgpt-base-v1 carries inherent limitations that necessitate careful consideration. Specifically, it may occasionally generate responses that are irrelevant or incorrect, particularly when tasked with interpreting complex or ambiguous queries. Additionally, given that its training is rooted in online data, the model may inadvertently reflect and perpetuate common online stereotypes and biases.

Evaluation

Please refer to our paper and codefor benchmark results.

Citation Information

@article{feng2023chessgpt,
  title={ChessGPT: Bridging Policy Learning and Language Modeling},
  author={Feng, Xidong and Luo, Yicheng and Wang, Ziyan and Tang, Hongrui and Yang, Mengyue and Shao, Kun and Mguni, David and Du, Yali and Wang, Jun},
  journal={arXiv preprint arXiv:2306.09200},
  year={2023}
}