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 10, 20:15 UTC
[CI][Bugfix] Fix token counting in chunked prefill streaming test (#31565)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Dec 31, 2025 at 17:05 UTC
21de6d4b02edb393adea68833ef5822125a9203d
1 file changed
+4
-1
tests/entrypoints/openai/test_chunked_prompt.py
+4
-1
@@ -116,7 +116,10 @@ async def test_chat_completion_stream_options_and_logprobs_with_long_prompts(
116
assert chunk.choices[0].logprobs is None
117
empty_chunks_received += 1
118
else:
119
- tokens_received += 1
119
+ # Count actual tokens from logprobs since multiple tokens
120
+ # can be batched into a single chunk
121
+ assert chunk.choices[0].logprobs and chunk.choices[0].logprobs.content
122
+ tokens_received += len(chunk.choices[0].logprobs.content)
123
124
if chunk.choices[0].finish_reason is not None:
125
finished = True