1Cat-vLLM

nvfp4_utils.py

…

test_allspark_gemm.py

…

test_awq_triton.py

…

test_awq.py

…

test_block_fp8.py

…

test_block_int8.py

…

test_cpu_fp8_scaled_mm.py

…

test_cutlass_scaled_mm.py

…

test_cutlass_w4a8_moe.py

…

test_cutlass_w4a8.py

…

test_flashinfer_nvfp4_scaled_mm.py

…

test_flashinfer_scaled_mm.py

…

test_fp8_min_max_helper.py

…

test_fp8_quant_group.py

…

test_fp8_quant.py

…

test_ggml.py

…

test_gguf_bitplane_transcode.py

[Kernel] Add GGUF affine bit planes to TurboMind SM70

2026-10-03 15:53:26 +08:00

test_gguf_dense_segments.py

[Kernel] Process small token groups in parallel segment launches

2026-10-06 00:09:14 +08:00

test_gguf_dmv_clock.py

[Kernel] Add device warp timing for GGUF projection diagnostics (#1018 )

2026-10-07 10:49:56 +08:00

test_gguf_dmv_formats.py

[Kernel] Decode IQ2 signed nibbles with original two-level scales

2026-10-07 03:47:20 +08:00

test_gguf_dmv.py

[Kernel] Share GGUF projection planes with TP2 shards (#1026 )

2026-10-07 16:38:06 +08:00

test_gguf_dp4a_capabilities.py

[Kernel] Admit calibrated canonical IQ4 expert integer dots

2026-10-06 05:05:49 +08:00

test_gguf_dp4a.py

[Kernel] Produce routed Q8 intermediates in fused expert gate-up

2026-10-05 21:02:39 +08:00

test_gguf_embedding_rows.py

[Memory] Keep Qwen3.5 GGUF embeddings in packed row storage

2026-10-07 23:13:31 +08:00

test_gguf_expert_down_dispatch.py

[Model] Dispatch canonical GGUF down vectors by measured rows

2026-10-05 04:30:36 +08:00

test_gguf_expert_down_graph.py

[Test] Check expert down vector and GEMM graph dispatch

2026-10-05 04:31:32 +08:00

test_gguf_expert_gate_up_dispatch.py

[Model] Retain IQ3_XXS original expert rows for joint projections

2026-10-05 10:51:27 +08:00

test_gguf_expert_gate_up_graph.py

[Model] Retain IQ3_XXS original expert rows for joint projections

2026-10-05 10:51:27 +08:00

test_gguf_expert_tp.py

[Model] Integrate standalone Qwen4Exp GGUF mapping on main

2026-10-05 00:47:24 +08:00

test_gguf_fp16_projection.py

[Kernel] Dispatch floating GGUF shards to FP32 row GEMV

2026-10-05 12:58:41 +08:00

test_gguf_head_layout_projection.py

[Benchmark] Measure GGUF head restoration and cover mixed output formats

2026-10-05 14:24:46 +08:00

test_gguf_iq1_m_records.py

[Kernel] Add source-sized IQ1_M shared-activation pair reader

2026-10-05 19:34:45 +08:00

test_gguf_iq2_s_pair_cursor.py

[Kernel] Add source-sized IQ2_S gated-pair reader

2026-10-05 17:52:28 +08:00

test_gguf_iq2_s_records.py

[Kernel] Add source-sized IQ2_S gated-pair reader

2026-10-05 17:52:28 +08:00

test_gguf_iq2_xs_operand.py

[Kernel] Add source-sized IQ2_XXS gated-pair reader

2026-10-05 18:57:35 +08:00

test_gguf_iq2_xs_pair_cursor.py

[Kernel] Add source-sized IQ2_XS gated-pair reader

2026-10-05 18:26:58 +08:00

test_gguf_iq2_xs_records.py

[Kernel] Add source-sized IQ2_XS gated-pair reader

2026-10-05 18:26:58 +08:00

test_gguf_iq2_xxs_pair_cursor.py

[Kernel] Add source-sized IQ2_XXS gated-pair reader

2026-10-05 18:57:35 +08:00

test_gguf_iq2_xxs_records.py

[Kernel] Add source-sized IQ2_XXS gated-pair reader

2026-10-05 18:57:35 +08:00

test_gguf_iq3_gated.py

[Test] Cover GGUF gated pair AOT module capture

2026-10-05 14:11:07 +08:00

test_gguf_iq3_xxs_pair_cursor.py

[Kernel] Add original-byte IQ3_XXS gated pair reader

2026-10-05 15:10:25 +08:00

test_gguf_iq3_xxs_records.py

[Core] Prepare source-sized IQ3_XXS pair records

2026-10-05 14:45:45 +08:00

test_gguf_iq4_float_book.py

[Kernel] Read native IQ4 integers directly from a shared Float table

2026-10-05 16:43:53 +08:00

test_gguf_iq4_native.py

[Kernel] Preserve IQ4_XS source bytes and nested scales (#950 )

2026-10-05 11:11:01 +08:00

test_gguf_lattice_decode.py

[Kernel] Select measured lattice grouped decode intervals

2026-10-04 01:02:41 +08:00

test_gguf_lattice_prefill.py

[Kernel] Calibrate 27B IQ3 dense prefill crossover

2026-10-04 05:38:14 +08:00

test_gguf_lattice_transcode.py

[Kernel] Add canonical lattice operators and TP4 measurements

2026-10-03 19:43:18 +08:00

test_gguf_lm_head_dispatch.py

[Bugfix] Preserve legacy GGUF head admission for unsupported formats

2026-10-04 19:16:01 +08:00

test_gguf_lut4_experts.py

[Test] Check IQ4 canonical weights against official dequantization

2026-10-06 04:57:52 +08:00

test_gguf_lut_transcode.py

[Kernel] Add canonical GGUF LUT4 decoding to TurboMind SM70

2026-10-03 17:54:44 +08:00

test_gguf_mixed_direct_output.py

[Kernel] Write mixed GGUF projections into shared output views

2026-10-05 11:12:22 +08:00

test_gguf_mixed_output_fallback.py

[Bugfix] Preserve imported formats outside canonical mixed families

2026-10-05 12:19:50 +08:00

test_gguf_native_linear.py

[Test] Preserve dynamic-M dispatch through projection export

2026-10-05 23:50:55 +08:00

test_gguf_native_pair.py

[Test] Check full operator admission for measured fallback

2026-10-06 05:15:55 +08:00

test_gguf_native.py

[Kernel] Keep GGUF upstream operators as capability-declared fallbacks

2026-10-03 10:40:20 +08:00

test_gguf_q2_k_pair_cursor.py

[Kernel] Add source-sized Q2_K shared-activation pair reader

2026-10-05 19:30:03 +08:00

test_gguf_q2_k_records.py

[Kernel] Add source-sized Q2_K shared-activation pair reader

2026-10-05 19:30:03 +08:00

test_gguf_q4_k_pair_cursor.py

[Kernel] Add original Q4_K reader to shared-activation pairs

2026-10-05 16:11:15 +08:00

test_gguf_q4_k_records.py

[Kernel] Preserve original Q4_K bytes in pair records

2026-10-05 15:50:51 +08:00

test_gguf_qkv.py

[Kernel] Wire GGUF small-M attention inputs by measured capability

2026-10-06 04:09:09 +08:00

test_gguf_qkvz.py

[Bugfix] Keep floating GDN fallback outside quantized descriptors

2026-10-06 01:56:47 +08:00

test_gguf_raw_grouped_capabilities.py

[Kernel] Isolate original-block expert gate/up in the core build

2026-10-05 05:26:32 +08:00

test_gguf_raw_grouped.py

[Kernel] Isolate original-block expert gate/up in the core build

2026-10-05 05:26:32 +08:00

test_gguf_runtime_dispatch.py

[Core] Preserve runtime GGUF projection dispatch in compiled graphs

2026-10-04 18:41:07 +08:00

test_gguf_small_grouped_capabilities.py

[Kernel] Admit measured small grouped down vector batches

2026-10-05 01:12:07 +08:00

test_gguf_small_grouped_vec.py

[Kernel] Add small grouped vectors for canonical GGUF down projections

2026-10-05 00:22:56 +08:00

test_gguf_small_output.py

[Kernel] Admit measured shared-activation output projections

2026-10-06 03:50:09 +08:00

test_gguf_three_format_qkv.py

[Kernel] Share GGUF projection planes with TP2 shards (#1026 )

2026-10-07 16:38:06 +08:00

test_gguf_transcode.py

[Kernel] Add canonical GGUF u4 and u8 affine TurboMind operators

2026-10-03 11:29:33 +08:00

test_gguf_turbomind_affine.py

[Kernel] Add canonical GGUF u4 and u8 affine TurboMind operators

2026-10-03 11:29:33 +08:00

test_gguf_turbomind_bitplanes.py

[Kernel] Compact centered Q3 metadata and record real-shape results

2026-10-03 16:43:47 +08:00

test_gguf_turbomind_dequant.py

[Kernel] Select calibrated GGUF prefill routes with shared scratch

2026-10-03 18:51:10 +08:00

test_gguf_turbomind_lattice.py

[Kernel] Measure GGUF grouped launch plans before capture

2026-10-03 23:14:58 +08:00

test_gguf_turbomind_lut4.py

[Kernel] Add canonical GGUF LUT4 decoding to TurboMind SM70

2026-10-03 17:54:44 +08:00

test_gguf_turbomind_model_moe.py

[Test] Exercise opaque expert dispatch and mixed fallback graphs

2026-10-05 06:03:19 +08:00

test_gguf_turbomind_model.py

[Core] Preserve runtime GGUF projection dispatch in compiled graphs

2026-10-04 18:41:07 +08:00

test_gguf_turbomind_u2.py

[Kernel] Add GGUF U2 and group16 affine operators on SM70

2026-10-03 13:44:13 +08:00

test_gguf_u2_transcode.py

[Kernel] Add GGUF U2 and group16 affine operators on SM70

2026-10-03 13:44:13 +08:00

test_gguf.py

…

test_gptq.py

…

test_hadacore.py

…

test_int8_kernel.py

…

test_int8_quant.py

…

test_machete_mm.py

…

test_marlin_gemm.py

…

test_mxfp4_qutlass.py

…

test_mxfp4_triton_ep.py

…

test_nvfp4_emulation.py

perf(nvfp4): use Triton weight dequant on SM70-SM88

2026-08-26 14:07:27 -07:00

test_nvfp4_qpn2_dequant.py

[Feature][SM75] Run ModelOpt NVFP4 and FP8 linears on Turing through the SM70 QPN kernels

2026-10-03 06:39:47 +08:00

test_nvfp4_quant.py

…

test_nvfp4_qutlass.py

…

test_nvfp4_scaled_mm.py

…

test_per_token_group_quant.py

…

test_rdna3_w4a16_selection.py

…

test_rdna3_w4a16.py

…

test_rocm_compressed_tensors_w4a16.py

…

test_rocm_skinny_gemms.py

…

test_scaled_mm_kernel_selection.py

…

test_silu_mul_nvfp4_quant.py

…

test_sm70_dflash2_fp8_m8.py

[Kernel] Replace the draft M8 FP16 pack and retain dynamic fallback

2026-10-05 11:57:31 +08:00

test_sm70_dflash2_fp16_m8.py

[Kernel] Add packed small-M DFlash2 projections on SM70

2026-10-04 13:11:35 +08:00

test_sm70_dflash_context_reduction.py

[Kernel] Speed up SM70 concurrent long decode

2026-09-27 07:48:20 +08:00

test_sm70_gdn_ba_verify.py

[Bugfix] Connect verifier projection fusion in Qwen3.5 override

2026-10-04 18:33:45 +08:00

test_sm70_marlin_splitk.py

[Kernel] Audit SM70 AWQ and Marlin quality fixes

2026-08-16 17:13:10 +08:00

test_sm70_moe_packed_w13.py

[Kernel][WIP] Integrate opt-in grouped SM70 MoE decode

2026-09-05 13:35:53 +08:00

test_sm70_mxfp4_e8m0.py

[Kernel] Preserve SM70 MXFP4 subnormal scales

2026-08-16 17:53:04 +08:00

test_sm70_nvfp4_native_layout.py

[Kernel] Record bundled NVFP4 serving validation

2026-10-07 00:49:24 +08:00

test_sm70_nvfp4_output_padding.py

[Bugfix] Preserve NVFP4 alignment and FP32 logits on TP2

2026-09-06 13:31:33 +08:00

test_sm70_nvfp4_prescaled_batch.py

[Kernel] Improve SM70 M48 supply, long softmax and TP4 reduction

2026-09-27 18:16:21 +08:00

test_sm70_nvfp4_qpn2_scales.py

[Bugfix] Preserve shared score dependencies across CUDA graphs

2026-09-22 10:02:19 +08:00

test_sm70_nvfp4_raw_scales.py

[Kernel][WIP] Add SM70 batch MoE candidates and raw-scale safety gates

2026-09-05 11:36:33 +08:00

test_sm70_nvfp4_small_prefill.py

[Kernel] Enable measured dense NVFP4 small-prefill dispatch

2026-10-04 00:17:07 +08:00

test_sm70_prefill_local_layouts.py

[Kernel][WIP] Generalize SM70 attention and projection shard layouts

2026-09-21 11:41:57 +08:00

test_sm70_qpn2_m32_packed_input.py

[Kernel] Enable shared SM70 C2 C4 C8 batch acceleration by default (#691 )

2026-09-26 18:43:15 +08:00

test_sm70_qpn8_block_fp8.py

[Bugfix][SM70] Start block QPN8 on Volta: warm up the layout it holds, keep the TurboMind copy only when decode can use it

2026-10-03 23:28:32 +02:00

test_sm70_qpn8_local_layouts.py

[Kernel][WIP] Generalize SM70 attention and projection shard layouts

2026-09-21 11:41:57 +08:00

test_sm70_qpn8_m32_packed_input.py

[Kernel] Enable shared SM70 C2 C4 C8 batch acceleration by default (#691 )

2026-09-26 18:43:15 +08:00

test_sm70_quantized_batch_supply.py

[Kernel] Improve SM70 M48 supply, long softmax and TP4 reduction

2026-09-27 18:16:21 +08:00

test_triton_scaled_mm.py

…

test_triton_w4a16.py

…

test_w4a16_kernel_selection.py

…