alea sync
Browse files- runs/aux_warm.log +50 -0
runs/aux_warm.log
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
`torch_dtype` is deprecated! Use `dtype` instead!
|
| 2 |
+
resumed from checkpoints/loop_r4_final.pt
|
| 3 |
+
pre-packing dataset...
|
| 4 |
+
packed 77888 (skipped 2114) trainable=889.2M
|
| 5 |
+
`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.
|
| 6 |
+
/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.
|
| 7 |
+
with torch.enable_grad(), device_autocast_ctx, torch.cpu.amp.autocast(**ctx.cpu_autocast_kwargs): # type: ignore[attr-defined]
|
| 8 |
+
step 50/2435 loss=0.1590 1.19s/step peak=10.3GB
|
| 9 |
+
step 100/2435 loss=0.1536 1.01s/step peak=10.3GB
|
| 10 |
+
step 150/2435 loss=0.1762 0.95s/step peak=10.3GB
|
| 11 |
+
step 200/2435 loss=0.1585 0.98s/step peak=10.3GB
|
| 12 |
+
step 250/2435 loss=0.1710 0.94s/step peak=10.3GB
|
| 13 |
+
step 300/2435 loss=0.1905 0.94s/step peak=10.3GB
|
| 14 |
+
step 350/2435 loss=0.1729 0.93s/step peak=10.3GB
|
| 15 |
+
step 400/2435 loss=0.1817 0.95s/step peak=10.3GB
|
| 16 |
+
step 450/2435 loss=0.1749 0.94s/step peak=10.3GB
|
| 17 |
+
step 500/2435 loss=0.1687 0.92s/step peak=10.3GB
|
| 18 |
+
step 550/2435 loss=0.1821 0.91s/step peak=10.3GB
|
| 19 |
+
step 600/2435 loss=0.1591 0.91s/step peak=10.3GB
|
| 20 |
+
step 650/2435 loss=0.2030 0.90s/step peak=10.3GB
|
| 21 |
+
step 700/2435 loss=0.1793 0.90s/step peak=10.3GB
|
| 22 |
+
step 750/2435 loss=0.1683 0.90s/step peak=10.3GB
|
| 23 |
+
step 800/2435 loss=0.1662 0.90s/step peak=10.3GB
|
| 24 |
+
step 850/2435 loss=0.1665 0.91s/step peak=10.3GB
|
| 25 |
+
step 900/2435 loss=0.1556 0.90s/step peak=10.3GB
|
| 26 |
+
step 950/2435 loss=0.1664 0.91s/step peak=10.3GB
|
| 27 |
+
step 1000/2435 loss=0.1665 0.91s/step peak=10.3GB
|
| 28 |
+
step 1050/2435 loss=0.1750 0.90s/step peak=10.3GB
|
| 29 |
+
step 1100/2435 loss=0.1561 0.90s/step peak=10.3GB
|
| 30 |
+
step 1150/2435 loss=0.1743 0.90s/step peak=10.3GB
|
| 31 |
+
step 1200/2435 loss=0.1405 0.90s/step peak=10.3GB
|
| 32 |
+
saved runs/aux_warm/step1200.pt
|
| 33 |
+
step 1250/2435 loss=0.1859 0.91s/step peak=10.3GB
|
| 34 |
+
step 1300/2435 loss=0.1655 0.92s/step peak=10.3GB
|
| 35 |
+
step 1350/2435 loss=0.1951 0.93s/step peak=10.3GB
|
| 36 |
+
step 1400/2435 loss=0.1581 0.92s/step peak=10.3GB
|
| 37 |
+
step 1450/2435 loss=0.1725 0.92s/step peak=10.3GB
|
| 38 |
+
step 1500/2435 loss=0.1520 0.92s/step peak=10.3GB
|
| 39 |
+
step 1550/2435 loss=0.1630 0.92s/step peak=10.3GB
|
| 40 |
+
step 1600/2435 loss=0.1516 0.92s/step peak=10.3GB
|
| 41 |
+
step 1650/2435 loss=0.1615 0.91s/step peak=10.3GB
|
| 42 |
+
step 1700/2435 loss=0.1516 0.91s/step peak=10.3GB
|
| 43 |
+
step 1750/2435 loss=0.1587 0.92s/step peak=10.3GB
|
| 44 |
+
step 1800/2435 loss=0.1595 0.93s/step peak=10.3GB
|
| 45 |
+
step 1850/2435 loss=0.1302 0.93s/step peak=10.3GB
|
| 46 |
+
step 1900/2435 loss=0.1753 0.93s/step peak=10.3GB
|
| 47 |
+
step 1950/2435 loss=0.1502 0.93s/step peak=10.3GB
|
| 48 |
+
step 2000/2435 loss=0.1541 0.94s/step peak=10.3GB
|
| 49 |
+
step 2050/2435 loss=0.1394 0.94s/step peak=10.4GB
|
| 50 |
+
step 2100/2435 loss=0.1641 0.94s/step peak=10.4GB
|