GitHub · Where software is built

Labels Milestones New issue

[BUG] download_file_with_lock crashes multi-GPU training on transient network failures (no retry/timeout)

code robustness

#554

· icenfly opened

on Feb 22, 2026

[BUG] core_eval.py throws ValueError when core-metric-max-per-task < num_fewshot

code robustness

#550

· suraj-self opened

on Feb 21, 2026

Add --model-tag to scripts.chat_cli call in runcpu.sh/speedrun.sh example

#542

· obriensystems opened

on Feb 19, 2026

make validation loss reproducible

#541

· karpathy opened

on Feb 18, 2026

base_train.py crashes with raw AssertionError on small batch sizes instead of auto-aligning or providing actionable error

#532

· suraj-self opened

on Feb 15, 2026

speedrun.sh updates

#529

· chrisjmccormick opened

on Feb 14, 2026

[BUG] AssertionError in gpt.py: Evaluation prompts exceed static RoPE cache (seq_len=256)

code robustness

#514

· suraj-self opened

on Feb 8, 2026

Add an easy way to centralize diverse trained gpt models (different seeds,arch etc). They can be used later on to develop new learning paradigms.

#507

· hhroberthdaniel opened

on Feb 5, 2026

Move <code>save_checkpoint()</code> before <code>evaluate_model()</code>

#446

· jolonf opened

on Jan 18, 2026

base_eval.py: hellaswag gets progressively slower and leaks memory on small models (Mac Studio)

#427

· pkrefta opened

on Jan 9, 2026

Add support for TPU?

#284

· 1-alexkarter opened

on Nov 13, 2025

Question: Are there plans to implement TorchScript/TorchTrace/ONNX inference?

#73

· i-zaitsev opened

on Oct 15, 2025