Skip to content

feat: Add configuration logging for benchmark reproducibility + bug fixes#1898

Draft
saud5150 wants to merge 1 commit intolightly-ai:masterfrom
saud5150:my-contribution
Draft

feat: Add configuration logging for benchmark reproducibility + bug fixes#1898
saud5150 wants to merge 1 commit intolightly-ai:masterfrom
saud5150:my-contribution

Conversation

@saud5150
Copy link

Adds comprehensive configuration logging to ImageNet ResNet50 benchmarks:

  • Automatically saves full resolved config to config.txt in log directory
  • Captures all CLI arguments + derived values (global_batch_size, world_size, etc.)
  • Includes environment info (PyTorch version, CUDA availability, timestamp)
  • Optional --verbose flag to print config to console before training
  • Uses only Python stdlib (no new dependencies)

Also fixes persistent_workers bug:

  • Fix ValueError when num_workers=0 with persistent_workers=True
  • Only enable persistent_workers when num_workers > 0

@saud5150 saud5150 marked this pull request as draft February 26, 2026 20:54
@guarin
Copy link
Contributor

guarin commented Feb 27, 2026

/review

Adds comprehensive configuration logging to ImageNet ResNet50 benchmarks:
- Automatically saves full resolved config to config.txt in log directory
- Captures all CLI arguments + derived values (global_batch_size, world_size, etc.)
- Includes environment info (PyTorch version, CUDA availability, timestamp)
- Optional --verbose flag to print config to console before training
- Uses only Python stdlib (no new dependencies)

Also fixes persistent_workers bug:
- Fix ValueError when num_workers=0 with persistent_workers=True
- Only enable persistent_workers when num_workers > 0
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants