A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 17:45 UTC
[ModelRunner V2] Revert token rank comparison difference for now (#34017)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Nick Hill committed
Feb 6, 2026 at 19:11 UTC
40218a82bad0bc772d75551a8009799f1a001db7
1 file changed
+1
-1
vllm/v1/worker/gpu/sample/logprob.py
+1
-1
@@ -68,7 +68,7 @@ def _ranks_kernel(
68
for i in range(0, vocab_size, BLOCK_SIZE):
69
block = i + tl.arange(0, BLOCK_SIZE)
70
logits = tl.load(row_ptr + block, mask=block < vocab_size, other=float("-inf"))
71
- n += tl.sum((logits > x).to(tl.int32))
71
+ n += tl.sum((logits >= x).to(tl.int32))
72
tl.store(output_ptr + req_idx, n)
73
74