Andrej Karpathy 92d63d4e8b clean up fragile code

2026-07-03 22:54:57 +00:00

..

base_eval.py

nuke huggingface and its giant dependency footprint from orbit. datasets still todo

2026-07-03 18:59:45 +00:00

base_train.py

fix deprecated commit on torchao

2026-07-03 19:27:43 +00:00

chat_cli.py

delete autocast, an unnecessary thorn in my side, manage dtypes directly

2026-03-04 23:55:30 +00:00

chat_eval.py

delete spurious 'cute' bloat that doesn't matter to a serious research stack

2026-07-03 18:48:32 +00:00

chat_rl.py

delete the whole report thing i think it was a bad idea it just bloats the code. doing this subtracts 540 LOC

2026-07-02 02:46:01 +00:00

chat_sft.py

delete spurious 'cute' bloat that doesn't matter to a serious research stack

2026-07-03 18:48:32 +00:00

infer_bench.py

clean up fragile code

2026-07-03 22:54:57 +00:00

tok_eval.py

rename fwe to climbmix

2026-07-03 00:43:50 +02:00

tok_train.py

fix token_bytes calculation to use raw token bytes instead of round-tripping through str, which corrupts tokens that are not valid standalone UTF-8 (e.g. bytes >= 0x80 became 3-byte U+FFFD). 193/32768 tokens were affected, so bpb of new runs will read ~0.05% higher than old runs. Thank you @ellenjxu #693

2026-07-03 17:30:47 +00:00