From 1a6d14e843cf2e101edf544e525119d345bd7564 Mon Sep 17 00:00:00 2001 From: jimmzhou Date: Fri, 19 Sep 2025 22:58:10 +0000 Subject: [PATCH 1/2] xfail mxfp8bf16 on sm103 --- tests/test_trtllm_gen_fused_moe.py | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/tests/test_trtllm_gen_fused_moe.py b/tests/test_trtllm_gen_fused_moe.py index 699e77c844d..26171526073 100644 --- a/tests/test_trtllm_gen_fused_moe.py +++ b/tests/test_trtllm_gen_fused_moe.py @@ -2034,6 +2034,15 @@ def test_moe_quantization_classes( f"Incompatible: {moe_impl.name} + {weight_processing['use_shuffled_weight']} + {weight_processing['layout']}" ) + # TODO(jimmzhou): enable MxFP4xBf16 on SM103 + if ( + type(moe_impl) is FP4Moe + and moe_impl.quant_mode == QuantMode.FP4_MXFP4_Bf16 + and compute_capability[0] == 10 + and compute_capability[1] == 3 + ): + pytest.xfail("Make MxFP4xBf16 nonfunctional on SM103 to avoid B200 regression") + moe_impl._cache_permute_indices = cache_permute_indices seed = 0 From 5014727863fb01a49bec19598c01bd273315070f Mon Sep 17 00:00:00 2001 From: jimmzhou Date: Mon, 22 Sep 2025 16:45:28 +0000 Subject: [PATCH 2/2] comment --- tests/test_trtllm_gen_fused_moe.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/tests/test_trtllm_gen_fused_moe.py b/tests/test_trtllm_gen_fused_moe.py index 26171526073..7fcfe4f1f69 100644 --- a/tests/test_trtllm_gen_fused_moe.py +++ b/tests/test_trtllm_gen_fused_moe.py @@ -2041,7 +2041,9 @@ def test_moe_quantization_classes( and compute_capability[0] == 10 and compute_capability[1] == 3 ): - pytest.xfail("Make MxFP4xBf16 nonfunctional on SM103 to avoid B200 regression") + pytest.xfail( + "Note(jimmzhou): Make MxFP4xBf16 nonfunctional on SM103 to avoid B200 regression" + ) moe_impl._cache_permute_indices = cache_permute_indices