xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-09 03:57:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

-inf

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon 2025-09-18 17:24:00 -07:00
parent b9c74487d2
commit 8deedfa42b
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/v1/worker/gpu/sampler.py
View File

@ -206,12 +206,12 @@ def _topk_log_softmax_kernel(
req_idx = tl.program_id(0) req_idx = tl.program_id(0)
row_ptr = logits_ptr + req_idx * logits_stride 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): for i in range(0, vocab_size, BLOCK_SIZE):
block = i + tl.arange(0, BLOCK_SIZE) block = i + tl.arange(0, BLOCK_SIZE)
l = tl.load(row_ptr + block, l = tl.load(row_ptr + block,
mask=block < vocab_size, mask=block < vocab_size,
other=-float("inf")) other=float("-inf"))
max_val = tl.max(tl.maximum(l, max_val)) max_val = tl.max(tl.maximum(l, max_val))
se = 0.0 se = 0.0