Intern-S1-mini assay transfer V27 — Bioavailability MA L4
Validation-selected checkpoint from the V27 Bioavailability MA L4 record-level assay-transfer run.
Provenance
- Base model:
jiosephlee/Intern-S1-mini-lm - Base model revision:
fcb667c380ae01f57693a45b4b5c2d331052a107 - Training dataset:
jiosephlee/assay-transfer-record-level-v27-bioavailability-ma-l4-intern - Dataset revision:
0e8363fef80865346b57aa25f79dcdd066a13856 - Training configuration: 10 epochs, seed 42, 8 GPUs, device batch size 4, gradient accumulation 4, effective batch size 128, Flash Attention 2, BFD packing, and padding-free training
- Selection metric: validation
L4/overall/knn_regression_id_ood_level_macro_mae_at_3(lower is better) - Selected optimizer step: 630
- Selected validation metric:
0.5153 - Training run
- Held-out model/Morgan comparison
Held-out test comparison
| Ranker | Joint ID/OOD level-macro MAE@3 | MAE@5 | Spearman | Spearman@3 | Top-1 hit@3 |
|---|---|---|---|---|---|
| Model | 0.7671 | 0.7536 | 0.3120 | 0.0040 | 0.1341 |
| Morgan vanilla | 0.8806 | 0.8405 | 0.0382 | 0.1667 | 0.1073 |
| Morgan weighted | 0.8590 | 0.7974 | 0.0533 | 0.1492 | 0.1111 |
Lower MAE is better; higher values are better for the other metrics. All W&B metric names retain the L4/ prefix.
The repository contains the full Transformers checkpoint and tokenizer files. metric.json is the complete validation-selection record stored with the checkpoint. Loading requires trust_remote_code=True for the custom tokenizer.