Janis Eccarius e2b4342f60 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

..

account

ci: bump version with Build-135

2026-06-22 17:41:58 +00:00

analysis

ci: bump version with Build-123

2026-06-15 20:52:53 +00:00

analytics

ci: bump version with Build-152

2026-06-23 22:30:53 +00:00

api

ci: bump version with Build-128

2026-06-21 09:39:54 +00:00

bot-platform

ci: bump version with Build-118

2026-06-09 22:12:20 +00:00

coordinator

ci: bump version with Build-97

2026-05-18 18:58:53 +00:00

core

ci: bump version with Build-130

2026-06-21 11:34:38 +00:00

io

ci: bump version with Build-108

2026-06-02 09:49:42 +00:00

json

ci: bump version with Build-60

2026-05-02 15:53:05 +00:00

official-bots

fix(official-bots): prevent Colab OOM in NNUE training

2026-06-24 22:18:18 +02:00

rule

ci: bump version with Build-79

2026-05-12 22:21:53 +00:00

security

ci: bump version with Build-68

2026-05-06 07:06:52 +00:00

store

ci: bump version with Build-110

2026-06-03 06:04:13 +00:00

tournament

ci: bump version with Build-148

2026-06-23 20:31:27 +00:00

ws

ci: bump version with Build-126

2026-06-19 10:28:49 +00:00