Janis Eccarius e2b4342f60
Build & Test (NowChessSystems) TeamCity build finished
fix(official-bots): prevent Colab OOM in NNUE training
Dense 98304-dim HalfKP features at batch_size=16384 cost ~6.4 GB/batch on the
host; with 8 hardcoded DataLoader workers and prefetch this OOM-killed the Colab
runtime.

- train.py: adaptive DataLoader workers (min(4, cpu_count), Colab free tier = 2),
  overridable via NNUE_LOADER_WORKERS; persistent_workers only when > 0.
- NNUETraining.ipynb: lower BATCH_SIZE 16384 -> 4096 with a memory-cost note.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-24 22:18:18 +02:00
2026-03-21 14:40:00 +01:00
S
Description
No description provided
1.3 GiB
Languages
Scala 83.1%
Python 12.2%
Bru 2.7%
Jupyter Notebook 0.8%
HTML 0.7%
Other 0.4%