Controls which config parts are saved by Lightning loggers.
Additionally saves number of model parameters.
Parameters:
| Name |
Type |
Description |
Default |
object_dict
|
dict[str, Any]
|
A dictionary containing the following objects: cfg, model, trainer.
|
required
|
Source code in src/utils/logging_utils.py
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 | @rank_zero_only
def log_hyperparameters(object_dict: dict[str, Any]) -> None:
"""Controls which config parts are saved by Lightning loggers.
Additionally saves number of model parameters.
Args:
object_dict: A dictionary containing the following objects: cfg, model, trainer.
"""
hparams: dict = {}
# Convert OmegaConf to dict and explicitly type it
cfg = cast(dict[str, Any], OmegaConf.to_container(object_dict["cfg"]))
model = object_dict["model"]
trainer = object_dict["trainer"]
if not trainer.logger:
log.warning("Logger not found! Skipping hyperparameter logging...")
return
assert hparams is not None
assert isinstance(hparams, dict)
# Add model parameters
hparams["model"] = cfg["model"]
# Save number of model parameters
hparams["model/params/total"] = sum(p.numel() for p in model.parameters())
hparams["model/params/trainable"] = sum(p.numel() for p in model.parameters() if p.requires_grad)
hparams["model/params/non_trainable"] = sum(p.numel() for p in model.parameters() if not p.requires_grad)
# Add other config sections
hparams["data"] = cfg["data"]
hparams["trainer"] = cfg["trainer"]
# Use safer access methods for optional fields
if "callbacks" in cfg:
hparams["callbacks"] = cfg["callbacks"]
if "extras" in cfg:
hparams["extras"] = cfg["extras"]
if "task_name" in cfg:
hparams["task_name"] = cfg["task_name"]
if "tags" in cfg:
hparams["tags"] = cfg["tags"]
if "ckpt_path" in cfg:
hparams["ckpt_path"] = cfg["ckpt_path"]
if "seed" in cfg:
hparams["seed"] = cfg["seed"]
hparams["execution_command"] = f"python {' '.join(sys.argv)}"
# Send hparams to all loggers
for logger in trainer.loggers:
logger.log_hyperparams(hparams)
|