fix: set kv_cache_dtype q4->float16 and torch_dtype float16

fix: add tokenizer_config with chat_template from base model

add model_q4.onnx_data

add model_q4.onnx with correct external data ref

fix: config with q4 dtype mapping

int4 WebGPU ONNX with patched external data refs

cleanup: remove fp16 files

fix: use full model config from merged model

fix: add fp16 dtype mapping to transformers.js_config

fp16 WebGPU ONNX (no quantization, compatible ops)

cleanup: remove old/duplicate ONNX files

fix: patch external data refs to underscore format (model.onnx_data)

fix: keep original ONNX filenames to match internal data references

v3: WebGPU int4 ONNX from 6-epoch fine-tune (loss 1.059)

Add generation_config.json

Restructure ONNX for Transformers.js compatibility (onnx/ subdir + config)

Add model card

Upload ONNX int4 model via Xenova's LFM2 builder

initial commit