[Frontend] Log output token IDs at DEBUG level (#52098)
Signed-off-by: ruirui6946 <142162413+ruirui6946@users.noreply.github.com>
This commit is contained in:
@@ -1,6 +1,7 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
|
||||
import logging
|
||||
from unittest.mock import MagicMock, patch
|
||||
|
||||
from vllm.entrypoints.serve.utils.request_logger import RequestLogger
|
||||
@@ -27,10 +28,10 @@ def test_request_logger_log_outputs():
|
||||
mock_logger.info.assert_called_once()
|
||||
call_args = mock_logger.info.call_args.args
|
||||
assert "Generated response %s%s" in call_args[0]
|
||||
assert "output_token_ids" not in call_args[0]
|
||||
assert call_args[1] == "test-123"
|
||||
assert call_args[3] == "Hello, world!"
|
||||
assert call_args[4] == [1, 2, 3, 4]
|
||||
assert call_args[5] == "stop"
|
||||
assert call_args[4] == "stop"
|
||||
|
||||
|
||||
def test_request_logger_log_outputs_streaming_delta():
|
||||
@@ -56,8 +57,7 @@ def test_request_logger_log_outputs_streaming_delta():
|
||||
assert call_args[1] == "test-456"
|
||||
assert call_args[2] == " (streaming delta)"
|
||||
assert call_args[3] == "Hello"
|
||||
assert call_args[4] == [1]
|
||||
assert call_args[5] is None
|
||||
assert call_args[4] is None
|
||||
|
||||
|
||||
def test_request_logger_log_outputs_streaming_complete():
|
||||
@@ -83,8 +83,7 @@ def test_request_logger_log_outputs_streaming_complete():
|
||||
assert call_args[1] == "test-789"
|
||||
assert call_args[2] == " (streaming complete)"
|
||||
assert call_args[3] == "Complete response"
|
||||
assert call_args[4] == [1, 2, 3]
|
||||
assert call_args[5] == "length"
|
||||
assert call_args[4] == "length"
|
||||
|
||||
|
||||
def test_request_logger_log_outputs_with_truncation():
|
||||
@@ -117,11 +116,37 @@ def test_request_logger_log_outputs_with_truncation():
|
||||
assert len(logged_output) == 10
|
||||
|
||||
# Check that token IDs were truncated to first 10 tokens
|
||||
logged_token_ids = call_args[0][4]
|
||||
mock_logger.debug.assert_called_once()
|
||||
logged_token_ids = mock_logger.debug.call_args.args[3]
|
||||
assert logged_token_ids == list(range(10))
|
||||
assert len(logged_token_ids) == 10
|
||||
|
||||
|
||||
def test_request_logger_log_output_token_ids_require_debug():
|
||||
mock_logger = MagicMock()
|
||||
mock_logger.isEnabledFor.side_effect = lambda level: level >= logging.INFO
|
||||
|
||||
with patch("vllm.entrypoints.serve.utils.request_logger.logger", mock_logger):
|
||||
request_logger = RequestLogger(max_log_len=4)
|
||||
|
||||
request_logger.log_outputs(
|
||||
request_id="test-no-token-ids",
|
||||
outputs="Test output",
|
||||
output_token_ids=[1, 2, 3],
|
||||
finish_reason="stop",
|
||||
)
|
||||
|
||||
mock_logger.info.assert_called_once()
|
||||
assert mock_logger.info.call_args.args == (
|
||||
"Generated response %s%s: output: %r, finish_reason: %s",
|
||||
"test-no-token-ids",
|
||||
"",
|
||||
"Test",
|
||||
"stop",
|
||||
)
|
||||
mock_logger.debug.assert_not_called()
|
||||
|
||||
|
||||
def test_request_logger_log_outputs_none_values():
|
||||
"""Test log_outputs handles None values correctly."""
|
||||
mock_logger = MagicMock()
|
||||
@@ -144,8 +169,7 @@ def test_request_logger_log_outputs_none_values():
|
||||
assert "Generated response %s%s" in call_args[0]
|
||||
assert call_args[1] == "test-none"
|
||||
assert call_args[3] == "Test output"
|
||||
assert call_args[4] is None
|
||||
assert call_args[5] == "stop"
|
||||
assert call_args[4] == "stop"
|
||||
|
||||
|
||||
def test_request_logger_log_outputs_empty_output():
|
||||
@@ -170,8 +194,7 @@ def test_request_logger_log_outputs_empty_output():
|
||||
assert "Generated response %s%s" in call_args[0]
|
||||
assert call_args[1] == "test-empty"
|
||||
assert call_args[3] == ""
|
||||
assert call_args[4] == []
|
||||
assert call_args[5] == "stop"
|
||||
assert call_args[4] == "stop"
|
||||
|
||||
|
||||
def test_request_logger_log_outputs_integration():
|
||||
@@ -245,4 +268,4 @@ def test_streaming_complete_logs_full_text_content():
|
||||
# Verify other parameters
|
||||
assert call_args[1] == "test-streaming-full-text"
|
||||
assert call_args[2] == " (streaming complete)"
|
||||
assert call_args[5] == "streaming_complete"
|
||||
assert call_args[4] == "streaming_complete"
|
||||
|
||||
@@ -141,9 +141,9 @@ class BaseFrontendArgs:
|
||||
"""Enable the `/tokenizer_info` endpoint. May expose chat
|
||||
templates and other tokenizer configuration."""
|
||||
enable_log_outputs: bool = False
|
||||
"""If set to True, log model outputs (generations).
|
||||
Requires `--enable-log-requests`. As with `--enable-log-requests`,
|
||||
information is only logged at INFO level at maximum."""
|
||||
"""If set to True, log model outputs (generations). Requires
|
||||
`--enable-log-requests`. Output text and finish reasons are logged at INFO,
|
||||
while output token IDs are logged at DEBUG."""
|
||||
enable_log_deltas: bool = True
|
||||
"""If set to False, output deltas will not be logged. Relevant only if
|
||||
--enable-log-outputs is set.
|
||||
|
||||
@@ -77,24 +77,28 @@ class RequestLogger:
|
||||
delta: bool = False,
|
||||
) -> None:
|
||||
max_log_len = self.max_log_len
|
||||
if max_log_len is not None:
|
||||
if outputs is not None:
|
||||
outputs = outputs[:max_log_len]
|
||||
|
||||
if output_token_ids is not None:
|
||||
# Convert to list and apply truncation
|
||||
output_token_ids = list(output_token_ids)[:max_log_len]
|
||||
if max_log_len is not None and outputs is not None:
|
||||
outputs = outputs[:max_log_len]
|
||||
|
||||
stream_info = ""
|
||||
if is_streaming:
|
||||
stream_info = " (streaming delta)" if delta else " (streaming complete)"
|
||||
|
||||
if logger.isEnabledFor(logging.DEBUG):
|
||||
if max_log_len is not None and output_token_ids is not None:
|
||||
output_token_ids = list(output_token_ids)[:max_log_len]
|
||||
|
||||
logger.debug(
|
||||
"Generated response %s%s details: output_token_ids: %s",
|
||||
request_id,
|
||||
stream_info,
|
||||
output_token_ids,
|
||||
)
|
||||
|
||||
logger.info(
|
||||
"Generated response %s%s: output: %r, "
|
||||
"output_token_ids: %s, finish_reason: %s",
|
||||
"Generated response %s%s: output: %r, finish_reason: %s",
|
||||
request_id,
|
||||
stream_info,
|
||||
outputs,
|
||||
output_token_ids,
|
||||
finish_reason,
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user