chen 29a313a402 [Optimization] Support FA2/FA3/FA4 with attn_mask_q (#6354) ...

* support FA4 sm100

* flash attn backend support mask

* flash attn backend run flashmask correct

* add test for flash_attn_backend and flash_attn_func

* check

* add test for fa4

* requirements.txt add fa4 whl

* check test on sm100

* fix CI conflict

* add enable_torch_proxy for flash_mask

* lazy import fa4

* check

* fix tests import

* check test_load_mpt import

2026-02-05 14:39:00 +08:00

..

test_air_top_p_sampling.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_cutlass_fp8_fp8_fp8_dual_gemm_fused.py

【Hackathon 9th No.28】add test_cutlass_fp8_fp8_fp8_dual_gemm_fused (#3935)

2025-09-10 14:57:49 +08:00

test_cutlass_scaled_mm.py

fix typos (#3633)

2025-08-28 14:42:24 +08:00

test_deqant_int8_cpp_extension.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_dequant.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_draft_model_postprocess.py

【Hackathon 9th No.63】add test_draft_model_postprocess.py (#3757)

2025-09-04 15:00:48 +08:00

test_draft_model_preprocess.py

[Speculative Decoding] Fix attn_mask_offset for multi-step MTP in mixed and PD-split modes (#5738)

2025-12-25 01:54:59 -08:00

test_draft_model_set_value_by_flags.py

[MTP] add test_draft_model_set_value_by_flags.py (#3741)

2025-09-02 19:33:33 +08:00

test_draft_model_update.py

【Hackathon 9th No.61、65】add test_draft_model_update (#3940)

2025-09-15 11:19:50 +08:00

test_dynamic_per_token_scaled_fp8_quant.py

[Docs] Add License in Unittest (#4957)

2025-11-12 10:44:09 +08:00

test_eagle_get_hidden_states.py

[UnitTest][MTP]add test_eagle_get_hidden_states (#3876)

2025-09-08 17:10:01 +08:00

test_eagle_get_self_hidden_states.py

[UnitTest][MTP]add test_eagle_get_hidden_states (#3876)

2025-09-08 17:10:01 +08:00

test_flash_mask_attn.py

[Optimization] Support FA2/FA3/FA4 with attn_mask_q (#6354)

2026-02-05 14:39:00 +08:00

test_fp8_fp8_half_cuda_core_gemm.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_fused_get_rotary_embedding.py

【Hackathon 9th No.25】add test_fused_get_rotary_embedding (#3892)

2025-09-12 15:38:43 +08:00

test_fused_hadamard_quant_fp8.py

[Docs] Add License in Unittest (#4957)

2025-11-12 10:44:09 +08:00

test_fused_masked_swiglu_quant.py

[Optimize] optimize mask_quant & swiglu (#6222)

2026-02-02 13:52:38 +08:00

test_fused_moe.py

[BugFix] fix wint2 (#6109)

2026-01-20 21:46:21 +08:00

test_fused_neox_rope_embedding.py

[Feature] Optim PaddleOCR-VL (#4873)

2025-11-07 14:56:44 +08:00

test_fused_rotary_position_encoding.py

【Hackathon 9th No.46】add test_fused_rotary_position_encoding (#3848)

2025-09-19 17:50:19 +08:00

test_gelu_tanh.py

[Feature] Optim PaddleOCR-VL (#4873)

2025-11-07 14:56:44 +08:00

test_get_padding_offset.py

remove speculate_get_padding_offset op (#6308)

2026-02-03 15:18:12 +08:00

test_get_position_ids_and_mask_encoder_batch.py

【Hackathon 9th No.34】add test_get_position_ids_and_mask_encoder_batch (#3739)

2025-09-04 14:54:30 +08:00

test_get_token_penalty_multi_scores.py

[BugFix] Fix token_penalty kernel (#6069)

2026-01-28 12:03:05 +08:00

test_group_swiglu_with_masked.py

【Hackathon 9th No.32】add unit tests for group_swiglu_with_masked (#3748)

2025-09-05 11:53:47 +08:00

test_hybrid_mtp_ngram.py

[Docs] Add License in Unittest (#4957)

2025-11-12 10:44:09 +08:00

test_limit_thinking_content_length.py

[BugFix] fix thinking bug (#4710)

2025-10-31 22:00:31 +08:00

test_machete_mm.py

[Optimize] Machete using group scale default (#4121)

2025-09-18 13:51:11 +08:00

test_moe_redundant_topk_select.py

【Hackathon 9th No.35】add test_moe_redundant_topk_select (#3867)

2025-09-05 11:29:02 +08:00

test_moe_top_k_select.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_ngram_match.py

[UnitTest][MTP] supplementary unit test for ngram_match (#3732)

2025-09-08 17:06:06 +08:00

test_noaux_tc_redundant.py

[Feature] Support noaux for eplb (#5143)

2025-11-21 14:10:32 +08:00

test_noaux_tc.py

[Docs] Add License in Unittest (#4957)

2025-11-12 10:44:09 +08:00

test_per_token_quant.py

[Feature] FD_USE_PHI_FP8_QUANT (#6320)

2026-02-03 22:33:03 -08:00

test_pre_cache_len_concat.py

[Others] Remove useless code (#5404)

2025-12-08 13:59:46 +08:00

test_qk_rmsnorm_fused.py

[Optimization] Support FA2/FA3/FA4 with attn_mask_q (#6354)

2026-02-05 14:39:00 +08:00

test_reasoning_phase_token_constraint.py

[Others] Support constrained decoding when enable_thinking is false (#6248)

2026-01-28 00:05:17 -08:00

test_rebuild_padding.py

add PADDLE_ENFORCE (#6321)

2026-02-04 10:47:19 +08:00

test_rejection_top_p_sampling.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_scaled_gemm_f8_i4_f16.py

Add stable ci (#3460)

2025-08-20 08:57:17 +08:00

test_set_value_by_flags_and_idx.py

Optimizing the performance of think length limit using custom operators (#4279)

2025-10-20 21:09:13 +08:00

test_share_external_data.py

【Hackathon 9th No.22】add unit tests for share_external_data (#3744)

2025-09-08 17:05:48 +08:00

test_speculate_get_output_padding_offset.py

[UnitTest][MTP]add test_speculate_get_output_padding_offset (#3740)

2025-09-03 22:21:21 +08:00

test_speculate_get_seq_lens_output.py

[UnitTest][MTP]add test_speculate_get_padding_offset (#3730)

2025-09-03 22:21:02 +08:00

test_speculate_get_target_logits.py

Add ut for speculative sampler (#4650)

2025-10-30 10:37:49 +08:00

test_speculate_get_token_penalty_multi_scores.py

[BugFix] Fix token_penalty kernel (#6069)

2026-01-28 12:03:05 +08:00

test_speculate_insert_first_token.py

Add ut for speculative sampler (#4650)

2025-10-30 10:37:49 +08:00

test_speculate_limit_thinking_content_length.py

[BugFix] fix speculate_limit_thinking_content_length (#5590)

2025-12-16 04:31:45 -08:00

test_speculate_set_stop_value_multi_seqs.py

[Feature] support stop_token_ids (#5399)

2025-12-09 17:49:12 +08:00

test_speculate_update.py

[Others] remove stop_nums (#6182)

2026-01-26 12:12:47 +08:00

test_speculate_verify.py

[Feature]Support tag phase token enforce generation (#6034)

2026-01-15 03:59:55 -08:00

test_speculative_schedule_cache.py

[Others] remove stop_nums (#6182)

2026-01-26 12:12:47 +08:00

test_split_fuse.py

[CI] Standard unittest (#3606)

2025-08-26 19:03:11 +08:00

test_stop_generation_multi_ends.py

[Feature] support stop_token_ids (#5399)

2025-12-09 17:49:12 +08:00

test_swap_layout.py

[CI] Add unit test for swap_layout && remove unit test of splitwise_scheduler (#6250)

2026-01-28 19:20:20 +08:00

test_token_penalty.py

[CI] Standard unittest (#3606)

2025-08-26 19:03:11 +08:00

test_top_k_renorm_probs.py

【Hackathon 9th No.37】add test_top_k_renorm_probs (#3755)

2025-09-16 11:12:46 +08:00

test_top_p_candidates.py

[Docs] Add License in Unittest (#4957)

2025-11-12 10:44:09 +08:00

test_tree_mask.py

add more cases for attention unit test (#5931)

2026-01-15 19:52:35 +08:00

test_tritonmoe_preprocess.py

【Hackathon 9th No.30】add test_tritonmoe_preprocess (#3891)

2025-09-22 15:31:32 +08:00

test_update_attn_mask.py

[Docs] Add License in Unittest (#4957)

2025-11-12 10:44:09 +08:00

test_update_inputs_v1.py

[Model Runner] Refactor execute_model for GPU async scheduling (#6176)

2026-01-28 14:19:33 +08:00

test_w4afp8_gemm.py

fix w4afp8 (#5634)

2025-12-22 13:39:41 +08:00

test_wfp8afp8_sparse_gemm.py

fix typos (#3633)

2025-08-28 14:42:24 +08:00