md-embedding smoke results (aaa.ama)¶
Generated: 2026-07-02
Artifacts: artifacts/md_embedding/aaa_docs
Reproduce:
export CHARMM_HOME=... CHARMM_LIB_DIR=... LD_LIBRARY_PATH=...
JAX_PLATFORMS=cpu uv run python scripts/collect_md_embedding_docs_results.py
Training (PhysNet smoke)¶
| Quantity | Value |
|---|---|
| Epochs | 8 |
| Train frames | 12500 |
| Valid frames | 1249 |
| Best valid loss | 102.65313720703125 |
| Checkpoint JSON | /home/ericb/mmml/artifacts/md_embedding/aaa_docs/aaa_smoke_params.json |
Validation metrics (PhysNet vs NPZ labels)¶
| Metric | Value |
|---|---|
| Energy MAE | 6.23443556918873 kcal/mol |
| Energy RMSE | 7.869658358803783 kcal/mol |
| Force MAE | 6.250538485566922 kcal/mol/Å |
| Force RMSE | 8.869026315270178 kcal/mol/Å |
| Eval samples | 6 |
Build (CHARMM TRIA + TIP3)¶
| Quantity | Value |
|---|---|
| Peptide atoms (TRIA) | 42 |
| Waters | 10 |
| Box side (Å) | 28.0 |
| Bonded total (kcal/mol) | None |
See also: md-embedding design, aaa.ama workflow.