Skip to content

Conversation

@jthemphill
Copy link
Owner

  • Updated .gitignore to track .pt and .onnx model files
  • Added model_final.pt (PyTorch checkpoint with 64 filters, 4 residual blocks)
  • Added model.onnx (exported model for inference with all heads)
  • Model trained on 1453 samples (240 drafting, 1213 movement)

Training results (Epoch 5):

  • Drafting: Policy=3.63, Value=0.53, Ownership=0.62, ScoreDiff=2.79
  • Movement: Policy=3.63, Value=0.01, Ownership=0.60, ScoreDiff=2.32

Loss contribution balanced at ~60% policy, ~5% value, ~30% auxiliary heads.

- Updated .gitignore to track .pt and .onnx model files
- Added model_final.pt (PyTorch checkpoint with 64 filters, 4 residual blocks)
- Added model.onnx (exported model for inference with all heads)
- Model trained on 1453 samples (240 drafting, 1213 movement)

Training results (Epoch 5):
- Drafting: Policy=3.63, Value=0.53, Ownership=0.62, ScoreDiff=2.79
- Movement: Policy=3.63, Value=0.01, Ownership=0.60, ScoreDiff=2.32

Loss contribution balanced at ~60% policy, ~5% value, ~30% auxiliary heads.
@jthemphill jthemphill enabled auto-merge (squash) January 4, 2026 05:45
@jthemphill jthemphill merged commit e57f573 into main Jan 4, 2026
2 checks passed
@jthemphill jthemphill deleted the claude/loss-weighting-analysis-Prt3o branch January 4, 2026 05:47
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.

3 participants