nik/Masters_Thesis
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Tech spike: four smoke-test runs and a divergence-stop test

Browse Source 
Pipeline works on MPS (~105 s per 300-step run), results are reproducible,
and the divergence stop works. Not experimental evidence.

Claude-Session: https://claude.ai/code/session_01S8KN7NuUCm8nXLmy6eDuVv
This commit is contained in:
Cormac Shannon committed 2026-10-28 18:14:26 +00:00
1 parent 55ba128dcc
commit 84b04d12e9
17 files changed
+792 -1
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

06-experiments/runs/20261028-181025_spike-qkoff-lr1e-1/metrics.jsonl
+35
View File
@@ -0,0 +1,35 @@
{"type": "eval", "iter": 0, "train_loss": 10.9141, "val_loss": 10.9167, "t": 5.41}
{"type": "train", "iter": 0, "loss": 10.9193, "ms": 4331.0, "t": 5.74}
{"type": "train", "iter": 10, "loss": 7.6884, "ms": 279.87, "t": 8.54}
{"type": "train", "iter": 20, "loss": 6.7757, "ms": 280.21, "t": 11.34}
{"type": "train", "iter": 30, "loss": 5.8674, "ms": 279.61, "t": 14.14}
{"type": "train", "iter": 40, "loss": 5.9878, "ms": 280.23, "t": 16.95}
{"type": "train", "iter": 50, "loss": 5.9047, "ms": 280.35, "t": 19.75}
{"type": "train", "iter": 60, "loss": 5.8552, "ms": 280.38, "t": 22.55}
{"type": "train", "iter": 70, "loss": 5.8407, "ms": 279.0, "t": 25.36}
{"type": "train", "iter": 80, "loss": 5.7461, "ms": 280.14, "t": 28.16}
{"type": "train", "iter": 90, "loss": 5.7984, "ms": 280.12, "t": 30.96}
{"type": "eval", "iter": 100, "train_loss": 5.7546, "val_loss": 5.789, "t": 37.47}
{"type": "train", "iter": 100, "loss": 5.7697, "ms": 4495.39, "t": 37.99}
{"type": "train", "iter": 110, "loss": 5.7763, "ms": 280.61, "t": 40.79}
{"type": "train", "iter": 120, "loss": 5.6788, "ms": 279.57, "t": 43.6}
{"type": "train", "iter": 130, "loss": 5.7067, "ms": 292.17, "t": 46.47}
{"type": "train", "iter": 140, "loss": 5.6992, "ms": 281.82, "t": 49.28}
{"type": "train", "iter": 150, "loss": 5.6794, "ms": 292.66, "t": 52.14}
{"type": "train", "iter": 160, "loss": 5.6159, "ms": 295.05, "t": 55.09}
{"type": "train", "iter": 170, "loss": 5.6578, "ms": 280.39, "t": 57.92}
{"type": "train", "iter": 180, "loss": 5.5731, "ms": 281.59, "t": 60.74}
{"type": "train", "iter": 190, "loss": 5.3993, "ms": 298.74, "t": 63.59}
{"type": "eval", "iter": 200, "train_loss": 5.4974, "val_loss": 5.5322, "t": 70.3}
{"type": "train", "iter": 200, "loss": 5.4276, "ms": 4449.02, "t": 70.73}
{"type": "train", "iter": 210, "loss": 5.524, "ms": 290.66, "t": 73.6}
{"type": "train", "iter": 220, "loss": 5.4767, "ms": 281.67, "t": 76.42}
{"type": "train", "iter": 230, "loss": 5.485, "ms": 282.2, "t": 79.23}
{"type": "train", "iter": 240, "loss": 5.4442, "ms": 281.09, "t": 82.04}
{"type": "train", "iter": 250, "loss": 5.3695, "ms": 281.49, "t": 84.86}
{"type": "train", "iter": 260, "loss": 5.3708, "ms": 281.08, "t": 87.67}
{"type": "train", "iter": 270, "loss": 5.36, "ms": 282.22, "t": 90.48}
{"type": "train", "iter": 280, "loss": 5.2385, "ms": 281.0, "t": 93.29}
{"type": "train", "iter": 290, "loss": 5.2615, "ms": 280.38, "t": 96.1}
{"type": "eval", "iter": 300, "train_loss": 5.2472, "val_loss": 5.2948, "t": 102.62}
{"type": "train", "iter": 300, "loss": 5.2321, "ms": 4418.66, "t": 103.06}