alea sync

runs/loop_r5.log

@@ -0,0 +1,153 @@
+active sources: {'gen_stoch': 0.9, 'gen_enterprise': 1.0, 'gen_structured': 0.8, 'gen_adversarial': 0.6, 'gen_numeric': 0.7, 'gen_long': 0.5, 'base': 0.4, 'gen_kaggle_cls': 1.0, 'gen_kaggle_reg': 0.7, 'gen_hf': 1.3}
+
+===== ROUND 0 (init=checkpoints/loop_r4_final.pt) =====
+  base: 4051
+  gen_adversarial: 6076
+  gen_enterprise: 10127
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+Repo card metadata block was not found. Setting CardData to empty.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  gen_hf: 13165
+  gen_kaggle_cls: 10127
+  gen_kaggle_reg: 7089
+  gen_long: 5063
+  gen_numeric: 7089
+  gen_stoch: 9114
+  gen_structured: 8101
+corpus: 80002 rows -> data/loop_train_r0.jsonl
++ /usr/bin/python3 -m alea.train --train data/loop_train_r0.jsonl --dev data/test_r2.jsonl --out runs/loop_r0 --init-from checkpoints/loop_r4_final.pt --train-mode full --epochs 1.0 --batch 32 --accum 1 --lr-head 0.0001 --lr-backbone 1e-05 --attn sdpa --save-every 100000 --max-len 4096
+`torch_dtype` is deprecated! Use `dtype` instead!
+resumed from checkpoints/loop_r4_final.pt
+pre-packing dataset...
+packed 80002 (skipped 0) trainable=632.6M
+`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.
+/usr/local/lib/python3.11/dist-packages/torch/utils/checkpoint.py:295: FutureWarning: `torch.cpu.amp.autocast(args...)` is deprecated. Please use `torch.amp.autocast('cpu', args...)` instead.
+  with torch.enable_grad(), device_autocast_ctx, torch.cpu.amp.autocast(**ctx.cpu_autocast_kwargs):  # type: ignore[attr-defined]
+step 50/2647 loss=0.1810 1.28s/step peak=7.3GB
+step 100/2647 loss=0.3829 1.07s/step peak=7.3GB
+step 150/2647 loss=0.3548 0.97s/step peak=7.3GB
+step 200/2647 loss=0.2335 1.09s/step peak=7.4GB
+step 250/2647 loss=1.1922 1.12s/step peak=7.4GB
+step 300/2647 loss=0.2678 1.13s/step peak=7.4GB
+step 350/2647 loss=0.1486 1.05s/step peak=7.4GB
+step 400/2647 loss=0.5354 1.09s/step peak=7.4GB
+step 450/2647 loss=0.5890 1.12s/step peak=7.4GB
+step 500/2647 loss=0.3027 1.11s/step peak=7.4GB
+step 550/2647 loss=0.6045 1.10s/step peak=7.4GB
+step 600/2647 loss=0.5351 1.12s/step peak=7.4GB
+step 650/2647 loss=1.0323 1.15s/step peak=7.4GB
+step 700/2647 loss=0.4962 1.16s/step peak=7.4GB
+step 750/2647 loss=0.4621 1.23s/step peak=7.4GB
+step 800/2647 loss=0.3442 1.30s/step peak=7.4GB
+step 850/2647 loss=0.3131 1.41s/step peak=7.4GB
+step 900/2647 loss=0.3653 1.46s/step peak=7.4GB
+step 950/2647 loss=0.2012 1.48s/step peak=7.4GB
+step 1000/2647 loss=0.3795 1.55s/step peak=7.4GB
+step 1050/2647 loss=0.5410 1.54s/step peak=7.4GB
+step 1100/2647 loss=0.3146 1.56s/step peak=7.4GB
+step 1150/2647 loss=0.3710 1.59s/step peak=7.4GB
+step 1200/2647 loss=0.7696 1.64s/step peak=7.4GB
+step 1250/2647 loss=0.3597 1.68s/step peak=7.4GB
+step 1300/2647 loss=0.2297 1.70s/step peak=7.4GB
+step 1350/2647 loss=0.1746 1.70s/step peak=7.4GB
+step 1400/2647 loss=0.1519 1.69s/step peak=7.4GB
+step 1450/2647 loss=0.5667 1.72s/step peak=7.4GB
+step 1500/2647 loss=0.3514 1.75s/step peak=7.4GB
+step 1550/2647 loss=0.3876 1.78s/step peak=7.4GB