📝 Article • 💻 Demo • 🤗 Model

LFM2.5-230M-Chess-ONNX

ONNX build of LFM2.5-230M-Chess in the transformers.js layout, for running the model in the browser on WebGPU with a WASM fallback.

This build is vocab-pruned to the 2,106 chess tokens, which cuts the full-precision embedding table out of the file.

Usage

import { AutoModelForCausalLM } from "@huggingface/transformers";

const model = await AutoModelForCausalLM.from_pretrained(
  "mlabonne/LFM2.5-230M-Chess-ONNX",
  { device: "webgpu", dtype: "q4" },
);

Prompts are plain token id sequences, so no text tokenizer is needed.

Files

Precision	Size (MiB)	Notes
fp32	629	reference graph
q4	212	what the Space loads