Emil_Shanaty/micro-scout
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: reject inconsistent training checkpoints on resume

Browse Source
This commit is contained in:
emil28092005
2026-09-16 04:11:44 +03:00
parent 9c41561977
commit abffddaa5e
3 changed files with 16 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/TRAINING.md
+1 -1
View File
@@ -42,7 +42,7 @@ uv run --no-sync python -m micro_scout.train \
--output runs/minilm-v1 --resume runs/minilm-v1/last --device cuda
```
Resume requires the same training configuration and dataset manifest. This command is intended for the project's own local optimizer states. Model weights use safetensors, and remote custom model code is disabled.
Resume requires the same training configuration, dataset manifest, and prepared-file hashes. A weights checksum prevents resuming mismatched weights and optimizer state after an incomplete checkpoint write. This command is intended for the project's own local optimizer states. Model weights use safetensors, and remote custom model code is disabled.
## Evaluation protocol