A high-throughput and memory-efficient inference and serving engine for LLMs
Make mistral_common optional by deferring MistralToolCall import (#45305)
Signed-off-by: Neil Schemenauer <nas@arctrix.com>
Neil Schemenauer committed
Jun 11, 2026 at 15:59 UTC
9bbf42be266f88a4fabc65a0c3336edc442821cf
1 file changed
+3
-1
vllm/tool_parsers/streaming.py
+3
-1
@@ -14,7 +14,6 @@ from vllm.entrypoints.openai.engine.protocol import (
14
DeltaMessage,
15
DeltaToolCall,
16
)
17
-from vllm.tool_parsers.mistral_tool_parser import MistralToolCall
17
from vllm.tool_parsers.utils import partial_json_loads
18
from vllm.utils.mistral import is_mistral_tokenizer
19
@@ -77,6 +76,9 @@ def extract_named_tool_call_streaming(
76
)
77
else:
78
if is_mistral_tokenizer(tokenizer):
79
+ # Import mistral_common only if we need it.
80
+ from vllm.tool_parsers.mistral_tool_parser import MistralToolCall
81
+
82
tool_call_id = MistralToolCall.generate_random_id()
83
else:
84
tool_call_id = make_tool_call_id(