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 4, 00:00 UTC
Fix: Nemotron 3 rescue whitespace-only final_content, not just None (#41846)
Signed-off-by: Nave Assaf <nassaf@nvidia.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Nave Assaf committed
May 10, 2026 at 05:07 UTC
dcb3135af7f65c82b11e78b180fce2627ede3f7a
1 file changed
+1
-1
vllm/reasoning/nemotron_v3_reasoning_parser.py
+1
-1
@@ -26,7 +26,7 @@ class NemotronV3ReasoningParser(DeepSeekR1ReasoningParser):
26
chat_template_kwargs.get("enable_thinking") is False
27
or chat_template_kwargs.get("force_nonempty_content") is True
28
)
29
- and final_content is None
29
+ and (final_content is None or not final_content.strip())
30
):
31
reasoning, final_content = final_content, reasoning
32