zl_bench/ops/inventory_vllm.csv at da2288564584039366e7be07fe4e94539a560c81

Files

T

zhoulin da22885645 Add multi-op batch screening layer; fix cudagraph fallback regressions

run_batch.py drives run_pytest.sh per operator across GPUs (stable-hash
sharding, per-op subprocess isolation, process-group timeouts, retry with
deterministic-failure cutoff, two-level dtype fallback, .complete resume,
per-op REPLAY_FROM). batch_summary.py aggregates run.log tables into
summary.csv. ops/ holds the curated assets: dual-repo inventories rebuilt
via AST scan + pytest collect verification, shape sets migrated from the
old regression harness and merged with upstream core_shapes class-name
keys (upstream's set_shapes falls back op_name -> MRO class name ->
1-D DEFAULT_SHAPES, so replacing the shape file without class keys
crashes the BLAS family), and a dismiss list where all 76 entries carry
verified reasons. Validated end to end: 1036-op full screen with zero
failures.

Also fix two cudagraph plugin regressions: newer torch appends "enable
device-side assertions" to every CUDA error, so the loose fatal-error
marker disabled the documented do_bench fallback entirely; and an aborted
graph capture can leave the default CUDA RNG generator stuck in capturing
state, poisoning every later torch.randn - captures now run under a
throwaway RNG state. run_pytest.sh gains an optional DTYPES passthrough.

2026-08-12 19:04:09 +00:00

11 KiB

Raw Blame History

1	repo	op	op_file	file	marked_skip	collected	op_names
2	vllm	perf_chunk_gated_delta_rule_fwd	FLA/test_chunk_gated_delta_rule_fwd	benchmark/test_FLA/test_chunk_gated_delta_rule_fwd.py		yes	chunk_gated_delta_rule_fwd
3	vllm	perf_chunk_gated_delta_rule	FLA/test_chunk_gated_delta_rule_perf	benchmark/test_FLA/test_chunk_gated_delta_rule_perf.py		yes	chunk_gated_delta_rule
4	vllm	chunk_gdn2	FLA/test_chunk_gdn2	benchmark/test_FLA/test_chunk_gdn2.py		yes	chunk_gdn2
5	vllm	perf_chunk_gla	FLA/test_chunk_gla_perf	benchmark/test_FLA/test_chunk_gla_perf.py		yes
6	vllm	chunk_kda	FLA/test_chunk_kda	benchmark/test_FLA/test_chunk_kda.py		yes	chunk_kda
7	vllm	perf_fused_recurrent_gated_delta_rule	FLA/test_fused_recurrent_gated_delta_rule_perf	benchmark/test_FLA/test_fused_recurrent_gated_delta_rule_perf.py		yes	fused_recurrent_gated_delta_rule
8	vllm	act_quant_perf	act_quant	benchmark/test_act_quant.py		yes	act_quant_triton
9	vllm	add_rms_norm	add_rms_norm	benchmark/test_add_rms_norm.py		yes	add_rms_norm
10	vllm	apply_repetition_penalties	apply_repetition_penalties	benchmark/test_apply_repetition_penalties.py		yes	apply_repetition_penalties
11	vllm	apply_rotary_pos_emb	apply_rotary_pos_emb	benchmark/test_apply_rotary_pos_emb.py		yes	apply_rotary_pos_emb
12	vllm	beam_search_score	beam_search_score	benchmark/test_beam_search_score.py		yes	beam_search_score
13	vllm	beam_search_score_	beam_search_score	benchmark/test_beam_search_score.py		yes	beam_search_score_
14	vllm	bincount	bincount	benchmark/test_bincount.py		yes	bincount
15	vllm	bincount_weighted	bincount	benchmark/test_bincount.py		yes	bincount_weighted
16	vllm	blas_benchmark	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes
17	vllm	perf_w8a8_block_fp8_matmul	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes	w8a8_block_fp8_matmul
18	vllm	perf_w8a8_block_fp8_matmul_deepgemm	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes	w8a8_block_fp8_matmul_deepgemm
19	vllm	perf_sparse_attention	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes	sparse_attention
20	vllm	mv_and_outer_benchmark	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes
21	vllm	addmv_benchmark	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes
22	vllm	vdot_benchmark	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes	vdot
23	vllm	addr_benchmark	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes	addr
24	vllm	perf_router_gemm	blas_perf_parallel	benchmark/test_blas_perf_parallel.py		yes	router_gemm
25	vllm	concat_and_cache_mla	concat_and_cache_mla	benchmark/test_concat_and_cache_mla.py		yes	concat_and_cache_mla
26	vllm	cp_gather_indexer_k_quant_cache_benchmark	cp_gather_indexer_k_quant_cache	benchmark/test_cp_gather_indexer_k_quant_cache.py		yes
27	vllm	cross_entropy_loss	cross_entropy_loss	benchmark/test_cross_entropy_loss.py		yes	cross_entropy_loss
28	vllm	cutlass_scaled_mm_benchmark	cutlass_scaled_mm	benchmark/test_cutlass_scaled_mm.py		yes
29	vllm	combine_topk_swa_indices_benchmark	deepseek_v4_attention_combine_topk_swa_indices	benchmark/test_deepseek_v4_attention_combine_topk_swa_indices.py		yes
30	vllm	compute_global_topk_indices_and_lens_benchmark	deepseek_v4_attention_compute_global_topk_indices_and_lens	benchmark/test_deepseek_v4_attention_compute_global_topk_indices_and_lens.py		yes
31	vllm	dequantize_and_gather_k_cache_benchmark	deepseek_v4_attention_dequantize_and_gather_k_cache	benchmark/test_deepseek_v4_attention_dequantize_and_gather_k_cache.py		yes
32	vllm	fused_q_kv_rmsnorm_benchmark	deepseek_v4_attention_fused_q_kv_rmsnorm	benchmark/test_deepseek_v4_attention_fused_q_kv_rmsnorm.py		yes
33	vllm	dgeglu	dgeglu	benchmark/test_dgeglu.py		yes	dgeglu
34	vllm	dreglu	dreglu	benchmark/test_dreglu.py		yes	dreglu
35	vllm	dswiglu	dswiglu	benchmark/test_dswiglu.py		yes	dswiglu
36	vllm	flash_attention_forward	flash_attention_forward	benchmark/test_flash_attention_forward.py		yes	flash_attention_forward
37	vllm	flash_attn_varlen_func	flash_attn_varlen_func	benchmark/test_flash_attn_varlen_func.py		yes	flash_attn_varlen_func
38	vllm	flash_attn_varlen_opt_func	flash_attn_varlen_opt_init_func	benchmark/test_flash_attn_varlen_opt_init_func.py		yes	flash_attn_varlen_func
39	vllm	flash_mla	flash_mla	benchmark/test_flash_mla.py		yes	flash_mla
40	vllm	flash_mla_sparse_fwd	flash_mla_sparse_fwd	benchmark/test_flash_mla_sparse_fwd.py		yes
41	vllm	flash_mla_with_kvcache	flash_mla_with_kvcache	benchmark/test_flash_mla_with_kvcache.py		yes
42	vllm	fp8_fp4_mqa_logits	fp8_fp4_mqa_logits	benchmark/test_fp8_fp4_mqa_logits.py		yes	fp8_fp4_mqa_logits
43	vllm	fp8_fp4_paged_mqa_logits	fp8_fp4_paged_mqa_logits	benchmark/test_fp8_fp4_paged_mqa_logits.py		yes	fp8_fp4_paged_mqa_logits
44	vllm	fused_add_rms_norm	fused_add_rms_norm	benchmark/test_fused_add_rms_norm.py		yes	fused_add_rms_norm
45	vllm	fused_deepseek_v4_qnorm_rope_kv_rope_quant_insert	fused_deepseek_v4_qnorm_rope_kv_rope_quant_insert	benchmark/test_fused_deepseek_v4_qnorm_rope_kv_rope_quant_insert.py		yes
46	vllm	fused_indexer_q_rope_quant	fused_indexer_q_rope_quant	benchmark/test_fused_indexer_q_rope_quant.py		yes	fused_indexer_q_rope_quant
47	vllm	fused_inv_rope_fp8_quant	fused_inv_rope_fp8_quant	benchmark/test_fused_inv_rope_fp8_quant.py		yes	fused_inv_rope_fp8_quant
48	vllm	fused_marlin_moe	fused_marlin_moe	benchmark/test_fused_marlin_moe.py		yes	fused_marlin_moe
49	vllm	fused_moe_impl_gems_vs_vllm	fused_moe	benchmark/test_fused_moe.py		yes	fused_experts_impl
50	vllm	fused_moe_fp8	fused_moe_fp8	benchmark/test_fused_moe_fp8.py		yes	fused_experts_impl
51	vllm	fused_moe_fp8_blockwise	fused_moe_fp8_blockwise	benchmark/test_fused_moe_fp8_blockwise.py		yes	fused_experts_impl
52	vllm	fused_moe_int4_w4a16	fused_moe_int4_w4a16	benchmark/test_fused_moe_int4_w4a16.py		yes	fused_experts_impl
53	vllm	fused_experts_impl_int8	fused_moe_int8	benchmark/test_fused_moe_int8.py		yes	fused_experts_impl
54	vllm	fused_experts_impl_int8_w8a16	fused_moe_int8_w8a16	benchmark/test_fused_moe_int8_w8a16.py		yes	fused_moe_int8_w8a16_gems_vs_bf16_deq
55	vllm	fused_moe_w8a16_mxq	fused_moe_w8a16	benchmark/test_fused_moe_w8a16.py		yes	fused_moe_w8a16_mxq_gems_vs_bf16_deq
56	vllm	fused_moe_w8a16_mxq_gems_vs_vllm	fused_moe_w8a16	benchmark/test_fused_moe_w8a16.py		yes	fused_moe_w8a16_mxq_gems_vs_vllm
57	vllm	geglu	geglu	benchmark/test_geglu.py		yes	geglu
58	vllm	gelu_and_mul	gelu_and_mul	benchmark/test_gelu_and_mul.py		yes	gelu_and_mul
59	vllm	grouped_topk_no_renorm	grouped_topk	benchmark/test_grouped_topk.py		yes	grouped_topk
60	vllm	grouped_topk_score_0	grouped_topk	benchmark/test_grouped_topk.py		yes	grouped_topk
61	vllm	grouped_topk_score_1	grouped_topk	benchmark/test_grouped_topk.py		yes	grouped_topk
62	vllm	indexer_k_quant_and_cache_benchmark	indexer_k_quant_and_cache	benchmark/test_indexer_k_quant_and_cache.py		yes
63	vllm	inplace_fused_experts_gems_vs_vllm	inplace_fused_experts	benchmark/test_inplace_fused_experts.py		yes	inplace_fused_experts
64	vllm	instance_norm	instance_norm	benchmark/test_instance_norm.py		yes	instance_norm
65	vllm	mhc_post	mhc	benchmark/test_mhc.py		yes	mhc_post
66	vllm	mhc_pre	mhc	benchmark/test_mhc.py		yes	mhc_pre
67	vllm	hc_split_sinkhorn_forward	mhc	benchmark/test_mhc.py		yes	hc_split_sinkhorn_forward
68	vllm	mhc_bwd	mhc	benchmark/test_mhc.py		yes	mhc_bwd
69	vllm	hc_head_fused_kernel	mhc	benchmark/test_mhc.py		yes	hc_head_fused_kernel
70	vllm	moe_align_block_size_triton	moe_align_block_size_triton	benchmark/test_moe_align_block_size_triton.py		yes	moe_align_block_size_triton
71	vllm	moe_sum	moe_sum	benchmark/test_moe_sum.py		yes	moe_sum
72	vllm	mrope_gems_vs_torch	mrope	benchmark/test_mrope.py		yes	mrope
73	vllm	outer	outer	benchmark/test_outer.py		yes	outer
74	vllm	outplace_fused_experts_gems_vs_vllm	outplace_fused_experts	benchmark/test_outplace_fused_experts.py		yes	outplace_fused_experts
75	vllm	pack_seq	pack_seq	benchmark/test_pack_seq.py		yes	pack_seq_triton
76	vllm	pack_seq_fp8	pack_seq	benchmark/test_pack_seq.py		yes	pack_seq_triton
77	vllm	perf_parallel_nsa	parallel_nsa	benchmark/test_parallel_nsa.py		yes	parallel_nsa
78	vllm	perf_parallel_nsa_compression	parallel_nsa_compression	benchmark/test_parallel_nsa_compression.py		yes	parallel_nsa_compression
79	vllm	per_token_group_quant_fp8	per_token_group_quant_fp8	benchmark/test_per_token_group_quant_fp8.py		yes	per_token_group_quant_fp8
80	vllm	persistent_topk	persistent_topk	benchmark/test_persistent_topk.py		yes	persistent_topk
81	vllm	reglu	reglu	benchmark/test_reglu.py		yes	reglu
82	vllm	reshape_and_cache	reshape_and_cache	benchmark/test_reshape_and_cache.py		yes	reshape_and_cache
83	vllm	reshape_and_cache_flash	reshape_and_cache_flash	benchmark/test_reshape_and_cache_flash.py		yes	reshape_and_cache_flash
84	vllm	perf_router_gemm	router_gemm	benchmark/test_router_gemm.py		yes	router_gemm
85	vllm	rwkv_ka_fusion	rwkv_ka_fusion	benchmark/test_rwkv_ka_fusion.py		yes	rwkv_ka_fusion
86	vllm	rwkv_mm_sparsity	rwkv_mm_sparsity	benchmark/test_rwkv_mm_sparsity.py		yes	rwkv_mm_sparsity
87	vllm	dynamic_scaled_int8_quant	scaled_int8_quant	benchmark/test_scaled_int8_quant.py		yes	dynamic_scaled_int8_quant
88	vllm	static_scaled_int8_quant	scaled_int8_quant	benchmark/test_scaled_int8_quant.py		yes	static_scaled_int8_quant
89	vllm	silu_and_mul	silu_and_mul	benchmark/test_silu_and_mul.py		yes	silu_and_mul
90	vllm	silu_and_mul_out	silu_and_mul	benchmark/test_silu_and_mul.py		yes	silu_and_mul_out
91	vllm	silu_and_mul_with_clamp	silu_and_mul_with_clamp	benchmark/test_silu_and_mul_with_clamp.py		yes	silu_and_mul_with_clamp
92	vllm	silu_and_mul_with_clamp_out	silu_and_mul_with_clamp	benchmark/test_silu_and_mul_with_clamp.py		yes	silu_and_mul_with_clamp_out
93	vllm	skip_layernorm	skip_layer_norm	benchmark/test_skip_layer_norm.py		yes	skip_layer_norm
94	vllm	sparse_attn_triton	sparse_attention	benchmark/test_sparse_attention.py	yes	yes	sparse_attention
95	vllm	sparse_mla_fwd_interface	sparse_mla_fwd_interface	benchmark/test_sparse_mla_fwd_interface.py		yes	sparse_mla_fwd_interface
96	vllm	stage_deepseek_v4_mega_moe_inputs_benchmark	stage_deepseek_v4_mega_moe_inputs	benchmark/test_stage_deepseek_v4_mega_moe_inputs.py		yes
97	vllm	swiglu	swiglu	benchmark/test_swiglu.py		yes	swiglu
98	vllm	top_k_per_row_decode	top_k_per_row_decode	benchmark/test_top_k_per_row_decode.py		yes	top_k_per_row_decode
99	vllm	top_k_per_row_prefill	top_k_per_row_prefill	benchmark/test_top_k_per_row_prefill.py		yes	top_k_per_row_prefill
100	vllm	topk_softmax	topk_softmax	benchmark/test_topk_softmax.py		yes	topk_softmax
101	vllm	topk_softplus_sqrt	topk_softplus_sqrt	benchmark/test_topk_softplus_sqrt.py		yes	topk_softplus_sqrt
102	vllm	triton_scaled_mm_benchmark	triton_scaled_mm	benchmark/test_triton_scaled_mm.py		yes	triton_scaled_mm
103	vllm	triton_unified_attention_perf	triton_unified_attention_perf	benchmark/test_triton_unified_attention_perf.py		yes
104	vllm	unpack_seq	unpack_seq	benchmark/test_unpack_seq.py		yes	unpack_seq_triton
105	vllm	unpack_seq_fp8	unpack_seq	benchmark/test_unpack_seq.py		yes	unpack_seq_triton
106	vllm	weight_norm_dim0	weight_norm	benchmark/test_weight_norm.py		yes	weight_norm
107	vllm	weight_norm_dim_last	weight_norm	benchmark/test_weight_norm.py		yes	weight_norm
108	vllm	weight_norm_interface	weight_norm_interface	benchmark/test_weight_norm_interface.py		yes	weight_norm_interface
109	vllm	weight_norm_interface_backward	weight_norm_interface	benchmark/test_weight_norm_interface.py		yes	weight_norm_interface_backward

11 KiB Raw Blame History

11 KiB

Raw Blame History