id: 219d27627fec4be1922dfb9271f7208d
parent_id: 
item_type: 1
item_id: 5d13b683b14241f2b3ff11c0a5b2ea92
item_updated_time: 1785502554504
title_diff: "[{\"diffs\":[[1,\"Overnight Training Pipeline (2026-07-22)\"]],\"start1\":0,\"start2\":0,\"length1\":0,\"length2\":40}]"
body_diff: "[{\"diffs\":[[1,\"# Overnight Training Pipeline — 2026-07-22\\\n\\\n## Goal\\\nTrain Gen5 neural nets to beat all previous generations at 9-max tables. Be profitable against every opponent type.\\\n\\\n## Key Fix: 9-Max Tables\\\nBoth collection scripts updated: `TABLE_SIZES=\\\"9 9 9 9 9 9 9 9\\\"` (was `\\\"3 3 4 4 5 5 6 6\\\"`).\\\n\\\n## Pipeline Stages\\\n\\\n### Stage 1: RangeNet v9 (GPU) — IN PROGRESS\\\n- Training on 5.6M diverse range samples (57-dim features with action history)\\\n- 25 epochs, batch 512, lr 1e-4\\\n- Epoch 8 metrics: top1=2.53%, top10=18.7% (vs v8: top1=3.41%, top10=21.5%)\\\n- v9 has less data (5.6M vs 23.4M for v8) but richer features\\\n\\\n### Stage 2: v4 Postflop (GPU)\\\n- **v4a**: CF v7 warm-start, Q-reg on old data (g48_v16 + sp_i1b + sp_i2, ~6.4M postflop transitions)\\\n- **v4b**: Fine-tune v4a on combined data (add diverse transitions when G48 collection finishes)\\\n- Aggressive masking: BetPot/Bet2x/Raise25x/RaisePot/AllIn masked at >25 BB\\\n\\\n### Stage 3: Preflop v2 (GPU)\\\n- Q-reg on diverse preflop data (when available)\\\n- Preflop v1 (corr=0.435) currently deployed with Bet2x/AllIn safety masking\\\n\\\n### Stage 4: Evaluation (9-max)\\\n- Script: `scripts/eval_9max.sh`\\\n- Tests vs ALL opponent types: Fish, Station, Chump, Flock, Gen3, Gen2, Tag, Lag, Nit, Maniac\\\n- 3 seeds × 2000-3000 hands per opponent (sequential to avoid GPU contention)\\\n\\\n### Stage 5: 9-Max Data Collection (CPU)\\\n- 8 tables × 500K hands, all 9-max\\\n- Diverse opponent pool (Fisher-Yates shuffled per table)\\\n\\\n### Stage 6: v5 Training\\\n- Warm-start from v4b\\\n- Combined data: old + diverse + 9-max\\\n- Target: best overall model\\\n\\\n## Current State (21:55)\\\n- Preflop v1: DONE (corr=0.435)\\\n- RangeNet v9: TRAINING (epoch 16/25)\\\n- G48 diverse collection: RUNNING (3/8 tables done, 4M transitions)\\\n- Baseline eval: CRASHED (GPU contention), will re-run later\\\n\\\n## Models Produced\\\n- `gen5_preflop_v1.safetensors` — preflop NN v1 (done)\\\n- `range_v9.safetensors` — RangeNet v9 (training)\\\n- `gen5_postflop_v4.safetensors` — postflop v4 (planned)\\\n- `gen5_preflop_v2.safetensors` — preflop v2 (planned)\\\n- `gen5_postflop_v5.safetensors` — postflop v5 (planned)\\\n\\\n## Scripts\\\n- `scripts/eval_9max.sh` — 9-max evaluation suite\\\n- `scripts/overnight_pipeline.sh` — full automated pipeline\\\n- Updated: `scripts/gen5_collect_g48.sh`, `scripts/gen5_selfplay_collect.sh` (9-max default)\\\n\"]],\"start1\":0,\"start2\":0,\"length1\":0,\"length2\":2288}]"
metadata_diff: {"new":{"id":"5d13b683b14241f2b3ff11c0a5b2ea92","parent_id":"2c8da247905946c3aa19eb4936e16323","latitude":"0.00000000","longitude":"0.00000000","altitude":"0.0000","author":"","source_url":"","is_todo":0,"todo_due":0,"todo_completed":0,"source":"joplin-desktop","source_application":"net.cozic.joplin-desktop","application_data":"","order":1784750079126,"user_updated_time":1784750079126,"markup_language":1,"is_shared":0,"share_id":"","conflict_original_id":"","master_key_id":"","user_data":"","deleted_time":1785502554504},"deleted":[]}
encryption_cipher_text: 
encryption_applied: 0
updated_time: 2026-07-31T12:59:41.619Z
created_time: 2026-07-31T12:59:41.619Z
is_locked: 0
type_: 13