Aryan KumarandAryan Kumar 4b4df1a94e [perf] Fused Q/K preconditioning + NVFP4 quantize for FA4 NVFP4 attention (bit-identical) (#2052) ...

Co-authored-by: Aryan Kumar <aryan5v@users.noreply.github.com>

2026-11-03 01:07:11 -08:00

..

test_attention_metadata_base.py

[refactor]: shared attention infra additions for QAT-compat (Attn-QAT 5/12) (#1383)

2026-05-23 16:19:24 -07:00

test_attn_qat_infer_arch_gate.py

[feat]: fp8 PV mode for the FA4-FP4 attention path (#1654)

2026-10-07 03:36:54 -07:00

test_attn_qat_train_kernel.py

[feat] Port NVFP4 QAT/QAD to modular train framework (#1619)

2026-07-18 15:01:09 -07:00

test_compile_policy.py

[perf]: QAT/QAD add safe regional compile for modular training (#1718)

2026-10-07 05:24:24 -07:00

test_cudnn_sdpa_dense.py

[bugfix]: get the full CI suite green on main (pytest 9 flag teardown, Qwen golden env) (#2059)

2026-11-02 23:32:40 -08:00

test_fa4_env_cache.py

[bugfix]: get the full CI suite green on main (pytest 9 flag teardown, Qwen golden env) (#2059)

2026-11-02 23:32:40 -08:00

test_flash_attn_capacity.py

[feat] Add H3 PDD/DMPDD training and packed Ref2VA preprocessing (#1984)

2026-10-20 11:15:57 -07:00

test_flash_attn_cute_custom_op.py

[perf]: QAT/QAD add safe regional compile for modular training (#1718)

2026-10-07 05:24:24 -07:00

test_flash_attn_cute_varlen_kwargs.py

[bugfix] FA4: stop forwarding max_seqlen to the varlen kernel (#2043)

2026-11-01 15:21:38 -08:00

test_flash_attn_default_custom_op.py

[misc] yapf pass under CI's interpreter (3.12) + pin hook language_version (#1702)

2026-08-08 22:24:17 -07:00

test_flash_attn_h3_packed_varlen.py

[perf] Add opt-in packed-varlen FA4 for MiniMax-H3 (#1742)

2026-08-22 17:16:47 -07:00

test_flash_attn_no_pad_custom_op.py

[new-model] Add V2A: native MMAudio inference pipeline (#1622)

2026-08-09 17:29:27 -07:00

test_flash_attn_no_pad_resolver.py

[misc] Move test-suite environment access into the registry and allowlists (#1898)

2026-09-30 00:15:39 -07:00

test_flash_attn_nvfp4_opt_out.py

[misc] Move test-suite environment access into the registry and allowlists (#1898)

2026-09-30 00:15:39 -07:00

test_forward_stacked.py

[perf] MiniMax H3: opt-in faster kernel tier for the DiT fused ops (#2042)

2026-11-01 15:22:09 -08:00

test_fp4_vsa_qat_tile64.py

[feat]: shared FastH3 NVFP4 FP4-attention QAD recipe (OmniRef + V2 T2AV) (#1989)

2026-11-03 00:54:28 -08:00

test_lingbotworld_fast_backend.py

[new-model] Added LingBot-World-Fast image-to-video support (#1665)

2026-10-07 03:32:38 -07:00

test_minimax_h3_fine_override.py

[feat]: FastH3 OmniRef inference on consumer devices (RTX 5090 / PRO 6000 / DGX Spark / 4090) (#1980)

2026-10-19 14:57:36 -07:00

test_minimax_h3_sparse_int8.py

[feat]: FastH3 on single RTX GPUs (#1919)

2026-10-06 11:37:01 -07:00

test_nvfp4_fa4_pad.py

[perf] MiniMax H3: faster FP4 attention padding, uint8 VAE pixel path and cuDNN FP4 GEMMs (#2046)

2026-11-01 22:23:10 -08:00

test_nvfp4_fused_qk_quant.py

[perf] Fused Q/K preconditioning + NVFP4 quantize for FA4 NVFP4 attention (bit-identical) (#2052)

2026-11-03 01:07:11 -08:00

test_nvfp4_qk_transform_deploy.py

[feat] Pika H3 Flex NVFP4 QAD training: calibration, QAD, FP4 attention and export (#2051)

2026-11-03 01:04:26 -08:00

test_packed_dense_attention.py

[feat] Add H3 PDD/DMPDD training and packed Ref2VA preprocessing (#1984)

2026-10-20 11:15:57 -07:00

test_qat_vsa_inference_memory.py

[feat]: FastH3 OmniRef inference on consumer devices (RTX 5090 / PRO 6000 / DGX Spark / 4090) (#1980)

2026-10-19 14:57:36 -07:00

test_sdpa_head_sizes.py

[bugfix]: allow unrestricted head sizes in SDPA (#1596)

2026-07-13 16:04:51 -07:00

test_sdpa_metadata_mask_contract.py

[feat] Add Kandinsky-5 T2V/I2V pipeline support (#1471)

2026-07-07 14:43:30 -07:00

test_selector_role_override.py

[misc] yapf pass under CI's interpreter (3.12) + pin hook language_version (#1702)

2026-08-08 22:24:17 -07:00

test_video_sparse_attention_metadata.py

[bugfix]: compute VSA topk from padded blocks (#1517)

2026-07-03 14:16:51 -07:00

test_vsa64_fused_layout.py

[perf] Fuse VSA64 tile layout for SM100 inference (#1900)

2026-10-07 15:31:58 -07:00

test_vsa_fp4_prefix_split.py

[feat]: FastH3 OmniRef inference on consumer devices (RTX 5090 / PRO 6000 / DGX Spark / 4090) (#1980)

2026-10-19 14:57:36 -07:00

test_vsa_h3_backward.py

[ci] Run single-GPU CI lanes on a DGX Spark runner; weekly full suite on main (#2007)

2026-10-22 13:44:55 -07:00

test_vsa_h3_blackwell_route.py

[ci] Run single-GPU CI lanes on a DGX Spark runner; weekly full suite on main (#2007)

2026-10-22 13:44:55 -07:00

test_vsa_h3_heads_first_tile.py

[refactor] VSA: rename the sm100a block-sparse kernel to ptx_blackwell (#1956)

2026-10-17 21:10:19 -07:00

test_vsa_h3_metadata.py

[feat] FastH3 Ref2VA PDD inference with reference-video VSA sparsity (#1907)

2026-10-02 11:28:07 -07:00

test_vsa_h3_ref2va_regions.py

[feat] Add H3 PDD/DMPDD training and packed Ref2VA preprocessing (#1984)

2026-10-20 11:15:57 -07:00

test_vsa_h3_tile_grad_safety.py

[feat] FastH3 Ref2VA PDD inference with reference-video VSA sparsity (#1907)

2026-10-02 11:28:07 -07:00

test_vsa_h3_training_capacity.py

[feat] Add H3 PDD/DMPDD training and packed Ref2VA preprocessing (#1984)

2026-10-20 11:15:57 -07:00