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 7, 17:45 UTC
[BugFix] anthropic/serving_messages: fix tool call arguments streaming (#34887)
Signed-off-by: Daniele Trifirò <dtrifiro@redhat.com> Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>
Daniele committed
Feb 26, 2026 at 06:39 UTC
4a9c07a0a2b8308a045476b48be29e37c349274b
1 file changed
+13
vllm/entrypoints/anthropic/serving.py
+13
@@ -432,6 +432,19 @@ class AnthropicServingMessages(OpenAIServingChat):
432
data = chunk.model_dump_json(exclude_unset=True)
433
yield wrap_data_with_event(data, "content_block_start")
434
content_block_started = True
435
+ if tool_call.function and tool_call.function.arguments:
436
+ chunk = AnthropicStreamEvent(
437
+ index=content_block_index,
438
+ type="content_block_delta",
439
+ delta=AnthropicDelta(
440
+ type="input_json_delta",
441
+ partial_json=tool_call.function.arguments,
442
+ ),
443
+ )
444
+ data = chunk.model_dump_json(exclude_unset=True)
445
+ yield wrap_data_with_event(
446
+ data, "content_block_delta"
447
+ )
448
449
else:
450
chunk = AnthropicStreamEvent(