diff --git a/tests/kernels/test_fused_indexer_q_rope_quant.py b/tests/kernels/test_fused_indexer_q_rope_quant.py index 8fd19ffb0b..73b147584c 100644 --- a/tests/kernels/test_fused_indexer_q_rope_quant.py +++ b/tests/kernels/test_fused_indexer_q_rope_quant.py @@ -289,16 +289,16 @@ def test_cutedsl_indexer_q_writes_stay_within_num_tokens(use_fp4, n_head): dtype=torch.uint8, device=device, ) - mod.fused_indexer_q_rope_quant_mxfp4_cutedsl( - positions[:num_tokens], - q[:num_tokens], - cos_sin_cache, - weights[:num_tokens], - 1.0, - 1.0, - outputs["q_packed"][:num_tokens], - outputs["q_scale"][:num_tokens], - weights_out[:num_tokens], + mod._INDEXER_Q_MXFP4_KERNEL( + positions=positions[:num_tokens], + q=q[:num_tokens], + cos_sin_cache=cos_sin_cache, + weights=weights[:num_tokens], + weights_softmax_scale=1.0, + weights_head_scale=1.0, + q_packed=outputs["q_packed"][:num_tokens], + q_scale=outputs["q_scale"][:num_tokens], + weights_out=weights_out[:num_tokens], ) else: outputs["q_fp8"] = torch.full( @@ -307,15 +307,15 @@ def test_cutedsl_indexer_q_writes_stay_within_num_tokens(use_fp4, n_head): dtype=torch.uint8, device=device, ) - mod.fused_indexer_q_rope_quant_fp8_cutedsl( - positions[:num_tokens], - q[:num_tokens], - cos_sin_cache, - weights[:num_tokens], - 1.0, - 1.0, - outputs["q_fp8"][:num_tokens].view(torch.float8_e4m3fn), - weights_out[:num_tokens], + mod._INDEXER_Q_FP8_KERNEL( + positions=positions[:num_tokens], + q=q[:num_tokens], + cos_sin_cache=cos_sin_cache, + weights=weights[:num_tokens], + weights_softmax_scale=1.0, + weights_head_scale=1.0, + q_fp8=outputs["q_fp8"][:num_tokens].view(torch.float8_e4m3fn), + weights_out=weights_out[:num_tokens], ) torch.accelerator.synchronize() diff --git a/vllm/models/deepseek_v4/nvidia/ops/fused_indexer_q_cutedsl.py b/vllm/models/deepseek_v4/nvidia/ops/fused_indexer_q_cutedsl.py index 6684e37f37..8052178a03 100644 --- a/vllm/models/deepseek_v4/nvidia/ops/fused_indexer_q_cutedsl.py +++ b/vllm/models/deepseek_v4/nvidia/ops/fused_indexer_q_cutedsl.py @@ -684,7 +684,7 @@ class IndexerQFp8Kernel(VllmCuTeDSLJitKernel["IndexerQFp8Kernel.CompileKey"]): q, cos_sin_cache, weights, - q_fp8, + q_fp8.view(torch.uint8), weights_out, float(weights_softmax_scale * weights_head_scale), )