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 14, 05:30 UTC
[ux] Switch a warning to debug about a pytorch fallback (#23750)
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Russell Bryant committed
Sep 25, 2025 at 10:38 UTC
532a6cfccbaa1bc943512cca06b48d5d3500669d
1 file changed
+3
-3
vllm/v1/sample/ops/topk_topp_sampler.py
+3
-3
@@ -109,9 +109,9 @@ class TopKTopPSampler(nn.Module):
109
# CPU-GPU synchronization while `flashinfer_sample` does.
110
if (k is None and p is None) or generators:
111
if generators:
112
- logger.warning_once("FlashInfer 0.2.3+ does not support "
113
- "per-request generators. Falling back to "
114
- "PyTorch-native implementation.")
112
+ logger.debug_once("FlashInfer 0.2.3+ does not support "
113
+ "per-request generators. Falling back to "
114
+ "PyTorch-native implementation.")
115
return self.forward_native(logits, generators, k, p)
116
assert self.logprobs_mode not in (
117
"processed_logits", "processed_logprobs"