yangzhuxinyzx 59463dd4cd Merge pull request #683 from Peuqui/sm70-moe-backend-explicit ...

[Bugfix][SM70] Let an explicit --moe-backend win over the TurboMind MoE gate

2026-09-25 22:10:05 +08:00

..

__init__.py

…

fp_quant.py

…

reference_mxfp4.py

…

test_auto_gptq.py

…

test_auto_round.py

…

test_awq_qpn_sm70.py

[Perf][SM70] Enable admitted AWQ QPN M1 with safe fallback

2026-09-06 15:24:47 +08:00

test_awq_sm70_fallback.py

…

test_awq_sm70_persistent_cap.py

perf(sm70): account for MTP verifier width in AWQ scratch

2026-09-03 00:19:01 -07:00

test_awq_sm70_sorted_output_contract.py

[Kernel][SM70] Preserve safe W2 schedules and balance only short tails

2026-09-07 12:51:41 +08:00

test_blackwell_moe.py

…

test_compressed_tensors.py

…

test_configs.py

…

test_cpu_offload.py

…

test_cpu_wna16.py

…

test_cutlass_w4a16.py

…

test_experts_int8.py

…

test_fp8.py

…

test_gfx950_moe.py

…

test_gptq_dynamic.py

…

test_gptq_v2.py

…

test_lm_head.py

…

test_mixed_precision.py

…

test_modelopt.py

…

test_nvfp4_mxfp4_gating.py

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

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

test_online.py

…

test_per_token_kv_cache.py

…

test_quantization_config_args.py

…

test_quark_w4a16.py

…

test_quark.py

…

test_register_quantization_config.py

…

test_sm70_awq_active_grouped_decode.py

test: preserve indexed-prefill admission on aligned AWQ base

2026-09-05 01:32:02 -07:00

test_sm70_awq_compact_metadata.py

[Perf][SM70] Enable compact AWQ metadata by default for Qwen3.8 TP4

2026-09-05 21:55:14 -07:00

test_sm70_awq_indexed_prefill.py

…

test_sm70_awq_prefill_exact_dense.py

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

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

test_sm70_batched_gemm_layouts.py

[Kernel] Add opt-in SM70 prescaled batch GEMM for FP8

2026-09-24 19:10:46 +08:00

test_sm70_channel_fp8_layout_guard.py

[Bugfix][SM70] Guard FP8 packing geometry without model restrictions (#484)

2026-09-07 12:37:05 +08:00

test_sm70_compressed_tensors_fastpaths.py

[Test] Align QPN8 shape assertions with current main

2026-09-22 13:50:07 +08:00

test_sm70_ct_fp8_cache_workspace.py

[Bugfix][SM70] Resolve channel-FP8 workspaces after AOT cache reload

2026-09-22 09:52:46 +08:00

test_sm70_fp8_prefill_exact_dense.py

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

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

test_sm70_fp32_lm_head_admission.py

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

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

test_sm70_glm53_modelopt_nvfp4.py

…

test_sm70_local_workspace_capacity.py

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

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

test_sm70_modelopt_mixed_nvfp4.py

[Core] Consolidate GLM verifier and no-MTP optimization stacks

2026-09-05 15:15:51 +08:00

test_sm70_modelopt_nvfp4_kernel_oracle.py

…

test_sm70_modelopt_nvfp4.py

…

test_sm70_moe_backend_override.py

[Bugfix][SM70] Let an explicit --moe-backend win over the TurboMind MoE gate

2026-09-23 06:26:41 +02:00

test_sm70_moe_load_cache_release.py

[Test][SM70] Keep load-lifecycle fixture aligned with W2 option

2026-09-07 12:57:24 +08:00

test_sm70_mxfp4_moe.py

…

test_sm70_nvfp4_grouped_decode_dispatch.py

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

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

test_sm70_nvfp4_integration.py

[Bugfix] Guard raw-scale concurrency and preserve push tail coverage

2026-09-05 15:10:26 +08:00

test_sm70_nvfp4_qpn2.py

[Core] Gate compact scale defaults on native scratch reuse support

2026-09-22 10:21:29 +08:00

test_sm70_online_qpn8.py

[Kernel][SM70] Fuse exact Qwen3.8 shared gate

2026-09-04 18:13:06 +08:00

test_sm70_quantized_model_compat.py

[Bugfix][SM70] Guard quantized compatibility paths and add regressions

2026-09-05 14:32:16 +08:00

test_sm70_turbomind_adapter.py

[Bugfix] Preserve NVFP4 alignment and FP32 logits on TP2

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

test_sm70_warmup.py

[Kernel] Reduce DFlash2 scale and LM-head residency

2026-09-12 23:45:29 +08:00

test_torchao.py

…

test_trtllm_nvfp4_hidden_dim_padding.py

…

test_turboquant.py

…

utils.py

…