From 8deedfa42bf8fa2d3ca8027a982ff52769260d81 Mon Sep 17 00:00:00 2001 From: Woosuk Kwon Date: Thu, 18 Sep 2025 17:24:00 -0700 Subject: [PATCH] -inf Signed-off-by: Woosuk Kwon --- vllm/v1/worker/gpu/sampler.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/v1/worker/gpu/sampler.py b/vllm/v1/worker/gpu/sampler.py index 6a2856b1b1..6c7e63c2f3 100644 --- a/vllm/v1/worker/gpu/sampler.py +++ b/vllm/v1/worker/gpu/sampler.py @@ -206,12 +206,12 @@ def _topk_log_softmax_kernel( req_idx = tl.program_id(0) row_ptr = logits_ptr + req_idx * logits_stride - max_val = -float("inf") + max_val = float("-inf") for i in range(0, vocab_size, BLOCK_SIZE): block = i + tl.arange(0, BLOCK_SIZE) l = tl.load(row_ptr + block, mask=block < vocab_size, - other=-float("inf")) + other=float("-inf")) max_val = tl.max(tl.maximum(l, max_val)) se = 0.0