Gahow Wang ec8114ecbc train: --eval-ckpt eval-only mode (v0-vs-v1 same-set val loss) ...

Expose eval_loss() and add a --eval-ckpt <path> branch to bin/train: load an
existing checkpoint into a model of the given arch and score it on the held-out
val split, then exit. Lets v0 and v1 be measured on the identical validation set
(the acceptance metric) without a separate eval binary.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

2026-06-15 18:44:40 +08:00

..

bin

train: --eval-ckpt eval-only mode (v0-vs-v1 same-set val loss)

2026-06-15 18:44:40 +08:00

checkpoint.rs

train: loop + checkpoint save/load + sampler + train binary

2026-06-15 16:29:58 +08:00

clip.rs

perf: GPU AdamW + grad-norm

2026-06-15 16:53:09 +08:00

data.rs

data: full TinyStories + tokenized-id cache, val loss, CLI arch

2026-06-15 18:34:48 +08:00

lib.rs

train: --eval-ckpt eval-only mode (v0-vs-v1 same-set val loss)

2026-06-15 18:44:40 +08:00

sample.rs

train: loop + checkpoint save/load + sampler + train binary

2026-06-15 16:29:58 +08:00

schedule.rs

data: gpt2 bpe via xserv-tokenizer + TinyStories corpus + lr schedule + grad clip

2026-06-15 16:29:32 +08:00

train_loop.rs

train: --eval-ckpt eval-only mode (v0-vs-v1 same-set val loss)

2026-06-15 18:44:40 +08:00