Andreas Karatzas c4041f37a4 [ROCm][LoRA] Fix MoE accuracy regression by preserving float32 router weight scaling (#31931) ...

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

2026-01-08 04:17:56 +00:00

..

layers

[ROCm][LoRA] Fix MoE accuracy regression by preserving float32 router weight scaling (#31931)

2026-01-08 04:17:56 +00:00

model_loader

Enable quantized attention in NemotronH models (#31898)

2026-01-07 17:37:19 +00:00

models

Enable quantized attention in NemotronH models (#31898)

2026-01-07 17:37:19 +00:00

warmup

[UX] Reduce DeepGEMM warmup log output to single progress bar (#30903)

2025-12-17 20:21:51 -08:00

__init__.py

[Platform] Deprecate seed_everything (#31659)

2026-01-04 18:34:04 -08:00

custom_op.py

[Bugfix][CPU] Fix RotaryEmbedding fallback causing gibberish with --enforce-eager (#31643)

2026-01-06 01:25:38 +08:00

parameter.py

[Docs] Replace rst style double-backtick with md single-backtick (#27091)

2025-10-17 02:47:34 -07:00

utils.py

[Platform] Deprecate seed_everything (#31659)

2026-01-04 18:34:04 -08:00