From 21ffc7353a258d3a72fa77a522222aa0af66ef11 Mon Sep 17 00:00:00 2001 From: Tyler Michael Smith Date: Fri, 20 Jun 2025 15:56:05 +0000 Subject: [PATCH] fixup Signed-off-by: Tyler Michael Smith --- vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py b/vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py index fae8d3745fef6..70836879d17c0 100644 --- a/vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py +++ b/vllm/model_executor/layers/fused_moe/batched_deep_gemm_moe.py @@ -18,9 +18,9 @@ has_deep_gemm = importlib.util.find_spec("deep_gemm") is not None def _silu_mul_fp8_quant_deep_gemm( # Pointers ------------------------------------------------------------ input_ptr, # 16-bit activations (E, T, 2*H) - y_q_ptr, # fp88 quantized activations (E, T, H) + y_q_ptr, # fp8 quantized activations (E, T, H) y_s_ptr, # 16-bit scales (E, T, G) - counts_ptr, # int32 num tokens per expert (E) + counts_ptr, # int32 num tokens per expert (E) # Sizes --------------------------------------------------------------- H: tl.constexpr, # hidden dimension (per output)