Aryan Kumar 21e749156c [fix]: address QAD review findings ...

- Include V2 width/height/num_frames and the train manifest in the step-0
  fingerprint so a dimension change cannot reuse a stale baseline.
- Honor CompressionSpec.first_steps in the V2 rung scope.
- Mark tagged checkpoints weights-only in metadata and teach the exporter
  to resolve and load them, so best/ can be exported.
- Add a tile-64 backward comparison against the STE reference.

2026-10-19 12:16:45 -07:00

..

attn_qat.md

[perf] Default SM100 long-sequence Attn-QAT tiling (#1899)

2026-10-07 15:14:38 -07:00

data_preprocess.md

[docs] Update docs and README (#975)

2026-01-04 14:59:19 -06:00

finetune.md

[feat] Add streaming and GPU-accelerated LoRA extraction (#1784)

2026-10-07 04:48:35 -07:00

hunyuan15_overfit.md

[feat] Add HunyuanVideo 1.5 embedding preprocessing pipeline (#1663)

2026-10-07 04:20:58 -07:00

minimax_h3_fp4_qad.md

[fix]: address QAD review findings

2026-10-19 12:16:45 -07:00

overview.md

[misc]: consolidate dataset download scripts under examples/datasets/ (#1667)

2026-08-02 14:21:24 -07:00

trackers.md

[feat]: add SwanLab tracker (#1461)

2026-07-07 08:35:38 +00:00

train_infra.md

[docs] Migrate documentation and cookbook to Blume (#1969)

2026-10-16 16:09:04 -07:00