Wenjia LiandHongsheng Liu 38a2500bdf [Perf] MiniMax H3 Qwen3-VL support fused RMSNorm on NPU (#5915) ...

Signed-off-by: Wenjia Li <wjialish@gmail.com>
Signed-off-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

2026-08-12 19:12:54 +08:00

..

__init__.py

[Model] Add MiniMax H3 diffusion support (#5691)

2026-08-03 14:10:40 +08:00

condition_noise.py

[Feature] Align MiniMax H3 official input matrix (#5752)

2026-08-05 07:12:01 +00:00

denoise_loop.py

[Attention] Refine TRTLLM attention support for MiniMax H3 (#5779)

2026-08-06 09:38:17 +00:00

encoder.py

[Perf] MiniMax H3 Qwen3-VL support fused RMSNorm on NPU (#5915)

2026-08-12 19:12:54 +08:00

minimax_h3_transformer.py

[Perf][Diffusion] MiniMax-H3: opt into packed varlen attention on NPU to eliminate quadratic mask materialization (#5891)

2026-08-10 19:47:03 +08:00

packed_sequence.py

[NPU][Quantization] Add RainFusion attention and INT8 online quantization for MiniMax H3 (#5706)

2026-08-06 14:15:21 +08:00

packed_tokens.py

[Model] Add MiniMax H3 diffusion support (#5691)

2026-08-03 14:10:40 +08:00

pipeline_minimax_h3.py

[NPU][Model] Add distilled 4-step sigma schedule support for MiniMax H3 t2va (#5991)

2026-08-11 13:46:10 +08:00

presentation.py

[Model] Add MiniMax H3 diffusion support (#5691)

2026-08-03 14:10:40 +08:00

quality_policy.py

[Model][Feat] Support Minimax-H3 quality grading requests through dynamic loading/unloading with Cache-DiT (#5853)

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

reference_video.py

[CI] Align MiniMax H3 Ref2VA inputs and add I2VA/Ref2VA accuracy coverage (#5978)

2026-08-11 13:36:37 +08:00

scheduling_minimax_h3_euler_ancestral.py

[Model] Add MiniMax H3 diffusion support (#5691)

2026-08-03 14:10:40 +08:00

time_request.py

[NPU][Model] Add distilled 4-step sigma schedule support for MiniMax H3 t2va (#5991)

2026-08-11 13:46:10 +08:00

vae.py

[NPU][Quantization] Add RainFusion attention and INT8 online quantization for MiniMax H3 (#5706)

2026-08-06 14:15:21 +08:00