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 16, 10:45 UTC
[Misc] Print stack trace using `logger.exception` (#9461)
Cyrus Leung committed
Oct 17, 2024 at 21:55 UTC
390be746494c625030c44749c8fbd04b899266af
8 files changed
+26
-30
vllm/entrypoints/openai/serving_chat.py
+3
-3
@@ -324,7 +324,7 @@ class OpenAIServingChat(OpenAIServing):
324
else:
325
tool_parsers = [None] * num_choices
326
except RuntimeError as e:
327
- logger.error("Error in tool parser creation: %s", e)
327
+ logger.exception("Error in tool parser creation.")
328
data = self.create_streaming_error_response(str(e))
329
yield f"data: {data}\n\n"
330
yield "data: [DONE]\n\n"
@@ -600,7 +600,7 @@ class OpenAIServingChat(OpenAIServing):
600
601
except ValueError as e:
602
# TODO: Use a vllm-specific Validation Error
603
- logger.error("error in chat completion stream generator: %s", e)
603
+ logger.exception("Error in chat completion stream generator.")
604
data = self.create_streaming_error_response(str(e))
605
yield f"data: {data}\n\n"
606
# Send the final done message after all response.n are finished
@@ -687,7 +687,7 @@ class OpenAIServingChat(OpenAIServing):
687
try:
688
tool_parser = self.tool_parser(tokenizer)
689
except RuntimeError as e:
690
- logger.error("Error in tool parser creation: %s", e)
690
+ logger.exception("Error in tool parser creation.")
691
return self.create_error_response(str(e))
692
693
tool_call_info = tool_parser.extract_tool_calls(
vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py
+5
-5
@@ -103,9 +103,9 @@ class Hermes2ProToolParser(ToolParser):
103
tool_calls=tool_calls,
104
content=content if content else None)
105
106
- except Exception as e:
107
- logger.error("Error in extracting tool call from response %s",
108
- e)
106
+ except Exception:
107
+ logger.exception(
108
+ "Error in extracting tool call from response.")
109
return ExtractedToolCallInformation(tools_called=False,
110
tool_calls=[],
111
content=model_output)
@@ -333,6 +333,6 @@ class Hermes2ProToolParser(ToolParser):
333
334
return delta
335
336
- except Exception as e:
337
- logger.error("Error trying to handle streaming tool call: %s", e)
336
+ except Exception:
337
+ logger.exception("Error trying to handle streaming tool call.")
338
return None # do not stream a delta. skip this token ID.
vllm/entrypoints/openai/tool_parsers/internlm2_tool_parser.py
+2
-2
@@ -166,8 +166,8 @@ class Internlm2ToolParser(ToolParser):
166
tool_call_arr["arguments"] = self.get_argments(tool_call_arr)
167
self.prev_tool_call_arr = [tool_call_arr]
168
return delta
169
- except Exception as e:
170
- logger.error("Error trying to handle streaming tool call: %s", e)
169
+ except Exception:
170
+ logger.exception("Error trying to handle streaming tool call.")
171
logger.debug(
172
"Skipping chunk as a result of tool streaming extraction "
173
"error")
vllm/entrypoints/openai/tool_parsers/llama_tool_parser.py
+4
-5
@@ -112,9 +112,8 @@ class Llama3JsonToolParser(ToolParser):
112
content=None)
113
return ret
114
115
- except Exception as e:
116
- logger.error("Error in extracting tool call from response: %s", e)
117
- print("ERROR", e)
115
+ except Exception:
116
+ logger.exception("Error in extracting tool call from response.")
117
# return information to just treat the tool call as regular JSON
118
return ExtractedToolCallInformation(tools_called=False,
119
tool_calls=[],
@@ -269,8 +268,8 @@ class Llama3JsonToolParser(ToolParser):
268
self.prev_tool_call_arr = tool_call_arr
269
return delta
270
272
- except Exception as e:
273
- logger.error("Error trying to handle streaming tool call: %s", e)
271
+ except Exception:
272
+ logger.exception("Error trying to handle streaming tool call.")
273
logger.debug(
274
"Skipping chunk as a result of tool streaming extraction "
275
"error")
vllm/entrypoints/openai/tool_parsers/mistral_tool_parser.py
+4
-4
@@ -111,8 +111,8 @@ class MistralToolParser(ToolParser):
111
tool_calls=tool_calls,
112
content=content if len(content) > 0 else None)
113
114
- except Exception as e:
115
- logger.error("Error in extracting tool call from response: %s", e)
114
+ except Exception:
115
+ logger.exception("Error in extracting tool call from response.")
116
# return information to just treat the tool call as regular JSON
117
return ExtractedToolCallInformation(tools_called=False,
118
tool_calls=[],
@@ -298,8 +298,8 @@ class MistralToolParser(ToolParser):
298
self.prev_tool_call_arr = tool_call_arr
299
return delta
300
301
- except Exception as e:
302
- logger.error("Error trying to handle streaming tool call: %s", e)
301
+ except Exception:
302
+ logger.exception("Error trying to handle streaming tool call.")
303
logger.debug(
304
"Skipping chunk as a result of tool streaming extraction "
305
"error")
vllm/executor/multiproc_worker_utils.py
+3
-5
@@ -3,7 +3,6 @@ import multiprocessing
3
import os
4
import sys
5
import threading
6
-import traceback
6
import uuid
7
from dataclasses import dataclass
8
from multiprocessing import Queue
@@ -227,10 +226,9 @@ def _run_worker_process(
226
except KeyboardInterrupt:
227
break
228
except BaseException as e:
230
- tb = traceback.format_exc()
231
- logger.error(
232
- "Exception in worker %s while processing method %s: %s, %s",
233
- process_name, method, e, tb)
229
+ logger.exception(
230
+ "Exception in worker %s while processing method %s.",
231
+ process_name, method)
232
exception = e
233
result_queue.put(
234
Result(task_id=task_id, value=output, exception=exception))
vllm/model_executor/model_loader/weight_utils.py
+2
-2
@@ -499,8 +499,8 @@ def kv_cache_scales_loader(
499
logger.error("File or directory '%s' not found.", filename)
500
except json.JSONDecodeError:
501
logger.error("Error decoding JSON in file '%s'.", filename)
502
- except Exception as e:
503
- logger.error("An error occurred while reading '%s': %s", filename, e)
502
+ except Exception:
503
+ logger.exception("An error occurred while reading '%s'.", filename)
504
# This section is reached if and only if any of the excepts are hit
505
# Return an empty iterable (list) => no KV cache scales are loaded
506
# which ultimately defaults to 1.0 scales
vllm/platforms/cuda.py
+3
-4
@@ -137,10 +137,9 @@ class CudaPlatform(Platform):
137
pynvml.NVML_P2P_CAPS_INDEX_NVLINK)
138
if p2p_status != pynvml.NVML_P2P_STATUS_OK:
139
return False
140
- except pynvml.NVMLError as error:
141
- logger.error(
140
+ except pynvml.NVMLError:
141
+ logger.exception(
142
"NVLink detection failed. This is normal if your"
143
- " machine has no NVLink equipped.",
144
- exc_info=error)
143
+ " machine has no NVLink equipped.")
144
return False
145
return True