xiaoou/hjha-server
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: N2H2V5独立分支用NeuralBotV5 - 新增NeuralBotV5类 + InitBots映射 + --n2h2v5参数

Browse Source
This commit is contained in:
xiaoou
2026-07-18 11:20:14 +08:00
parent aa4cbd0a3e
commit ebf2a0dc98
28682 changed files with 483795 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
training_loop.log Normal file
View File

@ -0,0 +1,37 @@
[03:59:02] ==================================================
[03:59:02] 持续训练 启动 | 截止 08:00
[03:59:02] 策略: 首轮10k → 后续5k/轮
[03:59:02] ==================================================
[03:59:02]
── 第1轮 10000局 (剩241min) ──
[03:59:02] ── 自对弈 10000局 ──
[05:57:41] Rounds: 10000/10000
Errors: 0
Total: 118.5min
Avg: 0.7s/round
[05:57:41] 10000 CSV
[05:57:41] ── V训练 ──
[05:57:59] 154235 samples from 10000 games
[0713 08:43:51] ==================================================
[0713 08:43:51] 持续训练: 20k局/轮 | 无限循环
[0713 08:43:51] ==================================================
[0713 08:43:51]
── 第1轮: 20000局 ──
[0713 08:43:51] ── 自对弈 20000局 ──
[0713 11:26:27] filtered out 2 non-Neural states
[0713 11:39:07] 25 ISMCTS q-value samples
[0713 11:39:07] filtered 10427 unusable states
[0713 11:39:24] 25 ISMCTS q-value samples
[0713 11:39:24] filtered 12 unusable states
[0713 11:46:51] ==================================================
[0713 11:46:51] 持续训练: 20k局/轮 | Neural-only + replay + eval
[0713 11:46:51] ==================================================
[0713 11:46:51]
── 第1轮: 20000局 ──
[0713 11:46:51] ── 自对弈 20000局 ──
[0713 11:51:21] ==================================================
[0713 11:51:21] 持续训练: 20k局/轮 | Neural-only + replay + eval
[0713 11:51:21] ==================================================
[0713 11:51:21]
── 第1轮: 20000局 ──
[0713 11:51:21] ── 自对弈 20000局 ──