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 15, 08:30 UTC
[Bugfix]: correctly propagate errors message caught at the chat_templating step to the client (#18769)
Signed-off-by: Guillaume Calmettes <gcalmettes@scaleway.com>
Guillaume Calmettes committed
May 28, 2025 at 05:35 UTC
9a21e331ff8c4ab052a654b3ebd9f67ddfff4845
1 file changed
+3
-3
vllm/entrypoints/chat_utils.py
+3
-3
@@ -1252,7 +1252,7 @@ def apply_hf_chat_template(
1252
# investigation.
1253
logger.exception(
1254
"An error occurred in `transformers` while applying chat template")
1255
- raise ValueError from e
1255
+ raise ValueError(str(e)) from e
1256
1257
def apply_mistral_chat_template(
1258
tokenizer: MistralTokenizer,
@@ -1281,7 +1281,7 @@ def apply_mistral_chat_template(
1281
# We convert those assertion errors to ValueErrors so they can be
1282
# are properly caught in the preprocessing_input step
1283
except (AssertionError, MistralCommonException) as e:
1284
- raise ValueError from e
1284
+ raise ValueError(str(e)) from e
1285
1286
# External library exceptions can sometimes occur despite the framework's
1287
# internal exception management capabilities.
@@ -1292,7 +1292,7 @@ def apply_mistral_chat_template(
1292
logger.exception(
1293
"An error occurred in `mistral_common` while applying chat "
1294
"template")
1295
- raise ValueError from e
1295
+ raise ValueError(str(e)) from e
1296
1297
def random_tool_call_id() -> str:
1298
return f"chatcmpl-tool-{random_uuid()}"