From 6fdb49392e5183c310c4703b883e7edbf25f6a96 Mon Sep 17 00:00:00 2001 From: Vensen Date: Tue, 12 May 2026 01:52:31 +0700 Subject: [PATCH] [Bugfix] Fix int32 overflow in DeepGEMM SiLU/mul FP8 Triton kernel (#42201) Signed-off-by: vensen Signed-off-by: Vensen Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> --- vllm/model_executor/layers/quantization/utils/fp8_utils.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/vllm/model_executor/layers/quantization/utils/fp8_utils.py b/vllm/model_executor/layers/quantization/utils/fp8_utils.py index d9aab35c25..7a54620c25 100644 --- a/vllm/model_executor/layers/quantization/utils/fp8_utils.py +++ b/vllm/model_executor/layers/quantization/utils/fp8_utils.py @@ -171,7 +171,7 @@ def _silu_mul_quant_fp8_packed_kernel( pid_pack = tl.program_id(0) pid_m = tl.program_id(1) - m_offset = pid_m * BLOCK_M + m_offset = pid_m.to(tl.int64) * BLOCK_M if m_offset >= M: return @@ -321,8 +321,8 @@ def _silu_mul_per_token_group_quant_fp8_colmajor( pid_n = tl.program_id(1) N_2 = N // 2 - m_offset = pid_m * BLOCK_M - n_offset = pid_n * BLOCK_N + m_offset = pid_m.to(tl.int64) * BLOCK_M + n_offset = pid_n.to(tl.int64) * BLOCK_N if m_offset >= M: return