Stefano CastagnettaandKartikey Vyas 4e0681ed56 [Quantization] Support packed NVFP4 Qwen4Exp PLE embeddings (#56273) ...

Signed-off-by: Stefano Castagnetta <scastagnetta@nvidia.com>
Co-authored-by: Kartikey Vyas <kartikey@artificialanalysis.ai>

2026-10-14 08:24:55 +08:00

..

attention

[HiSparse] Read GPU-resident KV directly for mixed and BF16 prefills (#60417)

2026-10-13 16:40:20 -04:00

fused_moe

[Bugfix] Fix DSpark graph-padding handling in the DeepSeek-V4.1 MoE fused top-k (#59166)

2026-10-14 07:49:52 +08:00

fusion

[XPU] fix fused_input_norm for MM (#59865)

2026-10-09 20:16:35 +08:00

hpc

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

mamba

[Bugfix] Fix ReplaySSM ring shape assertion when TP does not divide Mamba2 n_groups (#60704)

2026-10-13 11:31:53 +00:00

minimax_rms_norm

[Fix] MiniMax: fix CUDA IPC handle exchange on cuda-bindings >= 13.4 (#60917)

2026-10-11 20:38:56 -04:00

pooler

[Feature][Pooling] Add offline fixed-length late chunking (#60164)

2026-10-13 11:16:21 +00:00

quantization

[Quantization] Support packed NVFP4 Qwen4Exp PLE embeddings (#56273)

2026-10-14 08:24:55 +08:00

rotary_embedding

[Model] Drop the 4x M-RoPE cache headroom for Qwen3-VL, Qwen3.5 and Qwen3.8 (#59853)

2026-10-10 06:51:52 +00:00

__init__.py

…

activation.py

[ROCm] Use silu_and_mul_with_clamp's torch._C op (#52052)

2026-09-22 18:30:14 -05:00

attention_layer_base.py

[Bugfix][HiSparse] Fix MTP acceptance collapse under FULL graphs with a saturated GPU pool (#59309)

2026-10-01 23:04:21 -07:00

conv.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

fused_allreduce_gemma_rms_norm.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

fused_embed_norm.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

fused_qk_norm_rope.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

indexer_topk.py

[Misc] Report a missing DeepSelect extension only when it is requested (#60242)

2026-10-07 09:53:17 +00:00

layernorm.py

[Model Loader] Keep norm layers neutral under dummy weight init (#61323)

2026-10-13 08:54:24 -07:00

lightning_attn.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

linear.py

[ROCm][Bugfix] Revert #56433 + #51692 to fix accuracy breakdown for DeepSeek-V4 (#57132)

2026-09-16 18:58:23 +00:00

logits_processor.py

[SpecDecode] Add LiLiCorr drafter (#57934)

2026-09-25 10:08:28 -04:00

mhc.py

[ROCm][DSv4.1][Perf] Run the delayed mHC seams through aiter's fused Triton kernel (#58655)

2026-09-29 10:19:10 +08:00

mla.py

[3/N] HiSparse: host-resident sparse-MLA decode hot-buffering (#53781)

2026-09-12 00:36:47 -04:00

resampler.py

Add pydocstyle to the ruff rules (#52136)

2026-09-16 16:44:29 +00:00

rocm_paged_mxfp4_indexer.py

[ROCm][DSv4.1] Paged MXFP4 sparse indexer on aiter's MQA-logits kernel (#58671)

2026-09-29 14:37:57 +08:00

sparse_attn_indexer.py

[Kernel][Build] Bump DeepGEMM to the 26/09/30 release and adapt the sparse indexer (#61149)

2026-10-12 23:24:43 -07:00

sparse_mqa_indexer.py

[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

2026-09-16 10:57:24 -07:00

utils.py

[Feature] Support batch invariance on ROCm (#52231)

2026-10-12 10:53:26 -05:00

vocab_parallel_embedding.py

[Qwen4Exp] support Sequence Parallelism for Qwen3.8-Flash-Next (#56322)

2026-10-12 20:56:44 +08:00