match reference gpt2 config (scale_attn_weights, no js_config)

force onnx/model.onnx via transformers.js_config

use_cache=true (graph is with-past)

remove stale root onnx (keep onnx/model.onnx)

no-constant-fold with-past onnx

add ONNX export for browser inference

add ONNX export for browser inference

add ONNX export for browser inference

use_cache=false to match no-cache ONNX graph

custom-GPT onnx in onnx/ subdir

add ONNX export for browser inference

fix tokenizer_config (add_bos/eos false)

put onnx in onnx/ subdir for transformers.js

add ONNX export for browser inference

add tokenizer.json for transformers.js

add ONNX export for browser inference

GPT-15M: full 300M-token training, val loss 1.75

initial commit