Repository navigation
Expand file tree
/
Copy path_logger.py
More file actions
413 lines (318 loc) · 10.3 KB
/
Copy path_logger.py
File metadata and controls
413 lines (318 loc) · 10.3 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
import sys
import ctypes
import logging
from dataclasses import dataclass, field
from typing import Iterable, Optional, TextIO, Union
import llama_cpp._ggml as _ggml
import llama_cpp.llama_cpp as llama_cpp_lib
# enum ggml_log_level {
# GGML_LOG_LEVEL_NONE = 0,
# GGML_LOG_LEVEL_INFO = 1,
# GGML_LOG_LEVEL_WARN = 2,
# GGML_LOG_LEVEL_ERROR = 3,
# GGML_LOG_LEVEL_DEBUG = 4,
# GGML_LOG_LEVEL_CONT = 5, // continue previous log
# };
GGML_LOG_LEVEL_NONE = 0
GGML_LOG_LEVEL_INFO = 1
GGML_LOG_LEVEL_WARN = 2
GGML_LOG_LEVEL_ERROR = 3
GGML_LOG_LEVEL_DEBUG = 4
GGML_LOG_LEVEL_CONT = 5
# common/log.h model:
#
# LOG_LEVEL_OUTPUT = 0
# LOG_LEVEL_ERROR = 1
# LOG_LEVEL_WARN = 2
# LOG_LEVEL_INFO = 3
# LOG_LEVEL_TRACE = 4
# LOG_LEVEL_DEBUG = 5
#
# Rule:
#
# event_verbosity <= verbosity_threshold => print
#
# Larger threshold means more verbose output.
#
LOG_LEVEL_OUTPUT = 0
LOG_LEVEL_ERROR = 1
LOG_LEVEL_WARN = 2
LOG_LEVEL_INFO = 3
LOG_LEVEL_TRACE = 4
LOG_LEVEL_DEBUG = 5
LOG_DEFAULT_LLAMA = LOG_LEVEL_INFO
LOG_DEFAULT_DEBUG = LOG_LEVEL_DEBUG
# Match the updated common_log_default_callback behavior:
# INFO -> TRACE
# CONT -> TRACE
#
# This is slightly more conservative for verbosity=3:
# if the backend emits INFO through ggml_log_callback, Python will hide it unless
# verbosity >= 4. This mirrors the current upstream default callback behavior.
GGML_LEVEL_TO_VERBOSITY = {
GGML_LOG_LEVEL_NONE: LOG_LEVEL_OUTPUT,
GGML_LOG_LEVEL_ERROR: LOG_LEVEL_ERROR,
GGML_LOG_LEVEL_WARN: LOG_LEVEL_WARN,
GGML_LOG_LEVEL_INFO: LOG_LEVEL_TRACE,
GGML_LOG_LEVEL_DEBUG: LOG_LEVEL_DEBUG,
GGML_LOG_LEVEL_CONT: LOG_LEVEL_TRACE, # fallback only; CONT inherits previous
}
GGML_LEVEL_TO_PYTHON_LEVEL = {
GGML_LOG_LEVEL_NONE: logging.INFO,
GGML_LOG_LEVEL_ERROR: logging.ERROR,
GGML_LOG_LEVEL_WARN: logging.WARNING,
GGML_LOG_LEVEL_INFO: logging.INFO,
GGML_LOG_LEVEL_DEBUG: logging.DEBUG,
GGML_LOG_LEVEL_CONT: logging.INFO, # fallback only; CONT inherits previous
}
# Default substring filters.
#
# These are intentionally simple substring filters instead of hard-coded
# special branches. Users can replace or clear them with set_log_filters().
DEFAULT_LOG_FILTERS = [
"CUDA Graph",
"CUDA graph"
]
VerbosityLike = Union[bool, int, str, None]
logger = logging.getLogger("llama-cpp-python")
@dataclass
class LoggerConfig:
# 0=output, 1=error, 2=warn, 3=info, 4=trace, 5=debug
verbosity: int = LOG_DEFAULT_LLAMA
show_output: bool = True
stdout: TextIO = sys.stdout
stderr: TextIO = sys.stderr
# If any substring is contained in a log message, the message is dropped.
log_filters: list[str] = field(default_factory=lambda: list(DEFAULT_LOG_FILTERS))
log_filters_case_sensitive: bool = True
_config = LoggerConfig()
_last_verbosity = LOG_LEVEL_INFO
def _normalize_verbosity(
value: VerbosityLike,
*,
default: int = LOG_DEFAULT_LLAMA,
) -> int:
"""
Convert user input to llama.cpp-style verbosity 0..5.
Compatibility:
verbose=False -> ERROR (1)
verbose=True -> DEBUG (5)
Numeric levels:
0 = output
1 = error
2 = warn
3 = info
4 = trace
5 = debug
"""
if value is None:
return default
if isinstance(value, bool):
return LOG_LEVEL_DEBUG if value else LOG_LEVEL_ERROR
if isinstance(value, int):
return max(LOG_LEVEL_OUTPUT, min(LOG_LEVEL_DEBUG, value))
if isinstance(value, str):
key = value.strip().lower()
aliases = {
"0": LOG_LEVEL_OUTPUT,
"output": LOG_LEVEL_OUTPUT,
"none": LOG_LEVEL_OUTPUT,
"1": LOG_LEVEL_ERROR,
"error": LOG_LEVEL_ERROR,
"err": LOG_LEVEL_ERROR,
"silent": LOG_LEVEL_ERROR,
"2": LOG_LEVEL_WARN,
"warn": LOG_LEVEL_WARN,
"warning": LOG_LEVEL_WARN,
"quiet": LOG_LEVEL_WARN,
"3": LOG_LEVEL_INFO,
"info": LOG_LEVEL_INFO,
"default": LOG_DEFAULT_LLAMA,
"normal": LOG_DEFAULT_LLAMA,
"4": LOG_LEVEL_TRACE,
"trace": LOG_LEVEL_TRACE,
"trc": LOG_LEVEL_TRACE,
"5": LOG_LEVEL_DEBUG,
"debug": LOG_LEVEL_DEBUG,
"verbose": LOG_LEVEL_DEBUG,
}
if key in aliases:
return aliases[key]
try:
parsed = int(key)
except ValueError as exc:
raise ValueError(
"_logger._normalize_verbosity: "
"verbosity must be one of 0..5, bool, None, or "
"'silent'/'quiet'/'info'/'trace'/'debug'"
) from exc
return max(LOG_LEVEL_OUTPUT, min(LOG_LEVEL_DEBUG, parsed))
raise TypeError(f"_logger._normalize_verbosity: unsupported verbosity type: {type(value)!r}")
def _verbosity_to_python_level(verbosity: int) -> int:
if verbosity >= LOG_LEVEL_DEBUG:
return logging.DEBUG
if verbosity >= LOG_LEVEL_INFO:
return logging.INFO
if verbosity >= LOG_LEVEL_WARN:
return logging.WARNING
return logging.ERROR
def _get_verbosity(level: int) -> int:
"""
Map ggml log level to Python-side verbosity.
GGML_LOG_LEVEL_INFO maps to LOG_LEVEL_INFO so that verbosity=3 remains
useful as the default info level.
"""
if level == GGML_LOG_LEVEL_NONE:
return LOG_LEVEL_OUTPUT
if level == GGML_LOG_LEVEL_ERROR:
return LOG_LEVEL_ERROR
if level == GGML_LOG_LEVEL_WARN:
return LOG_LEVEL_WARN
if level == GGML_LOG_LEVEL_INFO:
return LOG_LEVEL_INFO
if level == GGML_LOG_LEVEL_DEBUG:
return LOG_LEVEL_DEBUG
if level == GGML_LOG_LEVEL_CONT:
return LOG_LEVEL_INFO
return LOG_LEVEL_DEBUG
def _decode_log_text(text: bytes) -> str:
return text.decode("utf-8", errors="replace")
def _matches_log_filter(msg: str) -> bool:
filters = _config.log_filters
if not filters:
return False
if _config.log_filters_case_sensitive:
return any(item and item in msg for item in filters)
msg_lower = msg.lower()
return any(item and item.lower() in msg_lower for item in filters)
def _should_drop(level: int, verbosity: int, msg: str) -> bool:
if verbosity > _config.verbosity:
return True
if level == GGML_LOG_LEVEL_NONE and not _config.show_output:
return True
if _matches_log_filter(msg):
return True
return False
@_ggml.ggml_log_callback
def ggml_log_callback(
level: int,
text: bytes,
user_data: ctypes.c_void_p,
):
global _last_verbosity
msg = _decode_log_text(text)
if level == GGML_LOG_LEVEL_CONT:
verbosity = _last_verbosity
else:
verbosity = _get_verbosity(level)
_last_verbosity = verbosity
if _should_drop(level, verbosity, msg):
return
out = _config.stdout if level == GGML_LOG_LEVEL_NONE else _config.stderr
print(msg, end="", flush=True, file=out)
# Keep a global reference to avoid ctypes callback being garbage-collected.
_ggml_log_callback_ref = ggml_log_callback
llama_cpp_lib.llama_log_set(_ggml_log_callback_ref, ctypes.c_void_p(0))
def configure_logging(
*,
verbosity: VerbosityLike = None,
verbose: Optional[bool] = None,
quiet: Optional[bool] = None,
silent: Optional[bool] = None,
show_output: Optional[bool] = None,
log_filters: Optional[Iterable[str]] = None,
append_log_filters: Optional[Iterable[str]] = None,
log_filters_case_sensitive: Optional[bool] = None,
):
"""
Configure native ggml/llama.cpp runtime logging.
Priority:
silent > quiet > verbosity > verbose > current config
Compatibility:
verbose=False -> ERROR
verbose=True -> DEBUG
Numeric levels:
0 = output
1 = error
2 = warn
3 = info
4 = trace
5 = debug
"""
if silent is True:
v = LOG_LEVEL_ERROR
elif quiet is True:
v = LOG_LEVEL_WARN
elif verbosity is not None:
v = _normalize_verbosity(verbosity)
elif verbose is not None:
v = _normalize_verbosity(verbose)
else:
v = _config.verbosity
_config.verbosity = v
logger.setLevel(_verbosity_to_python_level(v))
if show_output is not None:
_config.show_output = show_output
if log_filters is not None:
_config.log_filters = [s for s in log_filters if s]
if append_log_filters is not None:
_config.log_filters.extend(s for s in append_log_filters if s)
if log_filters_case_sensitive is not None:
_config.log_filters_case_sensitive = log_filters_case_sensitive
def set_verbose(verbose: bool):
"""
Backward-compatible bool API.
False -> ERROR
True -> DEBUG
"""
configure_logging(verbose=verbose)
def set_verbosity(verbosity: VerbosityLike):
configure_logging(verbosity=verbosity)
def get_verbosity() -> int:
return _config.verbosity
def set_quiet(quiet: bool = True):
configure_logging(quiet=quiet)
def set_silent(silent: bool = True):
configure_logging(silent=silent)
def set_log_filters(
filters: Iterable[str],
*,
case_sensitive: bool = True,
):
"""
Replace all substring log filters.
Example:
set_log_filters(["CUDA Graph id", "clip_model_loader: tensor"])
"""
configure_logging(
log_filters=filters,
log_filters_case_sensitive=case_sensitive,
)
def get_log_filters() -> list[str]:
return list(_config.log_filters)
def add_log_filters(filters: Iterable[str]):
"""
Append substring log filters.
"""
configure_logging(append_log_filters=filters)
def clear_log_filters():
"""
Clear all substring log filters, including default filters.
"""
_config.log_filters.clear()
def reset_log_filters():
"""
Restore default substring log filters.
"""
_config.log_filters = list(DEFAULT_LOG_FILTERS)
def get_log_filters_case_sensitive() -> bool:
return _config.log_filters_case_sensitive
def reset_logging():
"""
Reset logging to default llama.cpp-style INFO verbosity and default filters.
"""
_config.verbosity = LOG_DEFAULT_LLAMA
_config.show_output = True
_config.log_filters = list(DEFAULT_LOG_FILTERS)
_config.log_filters_case_sensitive = True
logger.setLevel(_verbosity_to_python_level(_config.verbosity))