hexone2086 56aa10a503 feat: per-step supervision over full sequence ...

- model outputs (B, S, 2) instead of (B, 2) — GRU output at every timestep
- train/val loss computed over all S timesteps with reduction=none
- benchmark/evaluate.py takes pred[:, -1, :] for final-step evaluation
- added per-step loss logging (8 evenly spaced steps) to TensorBoard

Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>

2026-07-09 15:49:23 +08:00

..

__init__.py

initial commit

2026-05-29 18:49:01 +08:00

benchmark.py

initial commit

2026-05-29 18:49:01 +08:00

config.py

initial commit

2026-05-29 18:49:01 +08:00

evaluate.py

feat: per-step supervision over full sequence

2026-07-09 15:49:23 +08:00