Tech spike: four smoke-test runs and a divergence-stop test
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
1 parent
55ba128dcc
commit
84b04d12e9
17 files changed
+792
-1
No files matched your search
@@ -0,0 +1,38 @@
|
||||
{
|
||||
"out_dir": "/Users/nik/ISE/9.CS6513/thesis/06-experiments/runs/20261028-180656_spike-qkoff-lr1e-3",
|
||||
"eval_interval": 100,
|
||||
"log_interval": 10,
|
||||
"eval_iters": 20,
|
||||
"eval_only": false,
|
||||
"always_save_checkpoint": false,
|
||||
"init_from": "scratch",
|
||||
"wandb_log": false,
|
||||
"wandb_project": "owt",
|
||||
"wandb_run_name": "gpt2",
|
||||
"dataset": "tinystories_spike",
|
||||
"gradient_accumulation_steps": 1,
|
||||
"batch_size": 32,
|
||||
"block_size": 256,
|
||||
"n_layer": 4,
|
||||
"n_head": 4,
|
||||
"n_embd": 256,
|
||||
"dropout": 0.0,
|
||||
"bias": false,
|
||||
"qk_norm": false,
|
||||
"seed": 1337,
|
||||
"data_root": "/Users/nik/ISE/9.CS6513/thesis/06-experiments/data",
|
||||
"learning_rate": 0.001,
|
||||
"max_iters": 300,
|
||||
"weight_decay": 0.1,
|
||||
"beta1": 0.9,
|
||||
"beta2": 0.95,
|
||||
"grad_clip": 1.0,
|
||||
"decay_lr": true,
|
||||
"warmup_iters": 30,
|
||||
"lr_decay_iters": 300,
|
||||
"min_lr": 0.0001,
|
||||
"backend": "nccl",
|
||||
"device": "mps",
|
||||
"dtype": "float32",
|
||||
"compile": false
|
||||
}
|
||||
@@ -0,0 +1,35 @@
|
||||
{"type": "eval", "iter": 0, "train_loss": 10.9141, "val_loss": 10.9167, "t": 5.28}
|
||||
{"type": "train", "iter": 0, "loss": 10.9193, "ms": 4306.09, "t": 5.61}
|
||||
{"type": "train", "iter": 10, "loss": 9.717, "ms": 277.79, "t": 8.4}
|
||||
{"type": "train", "iter": 20, "loss": 7.856, "ms": 278.89, "t": 11.19}
|
||||
{"type": "train", "iter": 30, "loss": 5.8282, "ms": 280.65, "t": 13.98}
|
||||
{"type": "train", "iter": 40, "loss": 5.188, "ms": 280.42, "t": 16.81}
|
||||
{"type": "train", "iter": 50, "loss": 4.7399, "ms": 279.76, "t": 19.6}
|
||||
{"type": "train", "iter": 60, "loss": 4.4652, "ms": 279.16, "t": 22.4}
|
||||
{"type": "train", "iter": 70, "loss": 4.3047, "ms": 279.57, "t": 25.2}
|
||||
{"type": "train", "iter": 80, "loss": 4.1141, "ms": 280.67, "t": 28.0}
|
||||
{"type": "train", "iter": 90, "loss": 4.0799, "ms": 279.89, "t": 30.8}
|
||||
{"type": "eval", "iter": 100, "train_loss": 4.0112, "val_loss": 4.0447, "t": 37.28}
|
||||
{"type": "train", "iter": 100, "loss": 3.9923, "ms": 4480.26, "t": 37.81}
|
||||
{"type": "train", "iter": 110, "loss": 3.9961, "ms": 280.65, "t": 40.61}
|
||||
{"type": "train", "iter": 120, "loss": 3.802, "ms": 280.13, "t": 43.4}
|
||||
{"type": "train", "iter": 130, "loss": 3.8455, "ms": 279.15, "t": 46.2}
|
||||
{"type": "train", "iter": 140, "loss": 3.8329, "ms": 278.97, "t": 48.99}
|
||||
{"type": "train", "iter": 150, "loss": 3.8187, "ms": 279.6, "t": 51.79}
|
||||
{"type": "train", "iter": 160, "loss": 3.683, "ms": 280.52, "t": 54.59}
|
||||
{"type": "train", "iter": 170, "loss": 3.7639, "ms": 280.91, "t": 57.39}
|
||||
{"type": "train", "iter": 180, "loss": 3.6422, "ms": 280.01, "t": 60.19}
|
||||
{"type": "train", "iter": 190, "loss": 3.4959, "ms": 294.68, "t": 63.04}
|
||||
{"type": "eval", "iter": 200, "train_loss": 3.5542, "val_loss": 3.5984, "t": 69.66}
|
||||
{"type": "train", "iter": 200, "loss": 3.4831, "ms": 4487.24, "t": 70.09}
|
||||
{"type": "train", "iter": 210, "loss": 3.5612, "ms": 280.52, "t": 72.91}
|
||||
{"type": "train", "iter": 220, "loss": 3.5742, "ms": 291.78, "t": 75.76}
|
||||
{"type": "train", "iter": 230, "loss": 3.458, "ms": 282.25, "t": 78.58}
|
||||
{"type": "train", "iter": 240, "loss": 3.5315, "ms": 281.29, "t": 81.4}
|
||||
{"type": "train", "iter": 250, "loss": 3.4362, "ms": 281.5, "t": 84.21}
|
||||
{"type": "train", "iter": 260, "loss": 3.4534, "ms": 281.67, "t": 87.03}
|
||||
{"type": "train", "iter": 270, "loss": 3.4163, "ms": 281.69, "t": 89.85}
|
||||
{"type": "train", "iter": 280, "loss": 3.3884, "ms": 282.78, "t": 92.67}
|
||||
{"type": "train", "iter": 290, "loss": 3.3656, "ms": 292.04, "t": 95.52}
|
||||
{"type": "eval", "iter": 300, "train_loss": 3.3974, "val_loss": 3.459, "t": 102.13}
|
||||
{"type": "train", "iter": 300, "loss": 3.3888, "ms": 4461.83, "t": 102.56}
|
||||
@@ -0,0 +1,86 @@
|
||||
{
|
||||
"run_id": "20261028-180656_spike-qkoff-lr1e-3",
|
||||
"name": "spike-qkoff-lr1e-3",
|
||||
"note": "tech spike smoke test (not an experiment)",
|
||||
"status": "completed",
|
||||
"started": "2026-10-28T18:06:56+00:00",
|
||||
"command": "/Users/nik/ISE/9.CS6513/thesis/06-experiments/.venv/bin/python /Users/nik/ISE/9.CS6513/thesis/06-experiments/nanogpt/train.py configs/spike.py --out_dir=/Users/nik/ISE/9.CS6513/thesis/06-experiments/runs/20261028-180656_spike-qkoff-lr1e-3 --data_root=/Users/nik/ISE/9.CS6513/thesis/06-experiments/data --qk_norm=False --learning_rate=1e-3",
|
||||
"config_file": "configs/spike.py",
|
||||
"overrides": [
|
||||
"qk_norm=False",
|
||||
"learning_rate=1e-3"
|
||||
],
|
||||
"config_requested": {
|
||||
"dataset": "tinystories_spike",
|
||||
"device": "mps",
|
||||
"dtype": "float32",
|
||||
"compile": false,
|
||||
"n_layer": 4,
|
||||
"n_head": 4,
|
||||
"n_embd": 256,
|
||||
"block_size": 256,
|
||||
"bias": false,
|
||||
"dropout": 0.0,
|
||||
"qk_norm": false,
|
||||
"batch_size": 32,
|
||||
"gradient_accumulation_steps": 1,
|
||||
"learning_rate": 0.001,
|
||||
"max_iters": 300,
|
||||
"lr_decay_iters": 300,
|
||||
"min_lr": 0.0001,
|
||||
"warmup_iters": 30,
|
||||
"weight_decay": 0.1,
|
||||
"grad_clip": 1.0,
|
||||
"eval_interval": 100,
|
||||
"eval_iters": 20,
|
||||
"log_interval": 10,
|
||||
"always_save_checkpoint": false,
|
||||
"seed": 1337
|
||||
},
|
||||
"git": {
|
||||
"commit": "55ba128dcca8310f85e99e172fe920a8c53a3a79",
|
||||
"branch": "main",
|
||||
"dirty": false
|
||||
},
|
||||
"nanogpt_upstream_commit": "3adf61e154c3fe3fca428ad6bc3818b27a3b8291",
|
||||
"data": {
|
||||
"dataset": "TinyStories V2 (GPT-4)",
|
||||
"source": "https://huggingface.co/datasets/roneneldan/TinyStories/resolve/main/",
|
||||
"tokenizer": "tiktoken gpt2",
|
||||
"files": {
|
||||
"TinyStoriesV2-GPT4-valid.txt": "6874bae9a4c1a4e7edcf0e53b86c17817e9cf881fc75ff2368da457b80c0585d"
|
||||
},
|
||||
"split": "valid file split 90/10 by story order: 24867 train / 2763 val stories",
|
||||
"tokens": {
|
||||
"train": 4920360,
|
||||
"val": 548664
|
||||
},
|
||||
"train_bin_bytes": 9840720,
|
||||
"val_bin_bytes": 1097328
|
||||
},
|
||||
"environment": {
|
||||
"host": "Cormacs-MacBook-Pro.local",
|
||||
"platform": "macOS-15.7.7-arm64-arm-64bit",
|
||||
"machine": "arm64",
|
||||
"cpu": "Apple M4 Pro",
|
||||
"python": "3.12.14",
|
||||
"torch": "2.15.0",
|
||||
"mps_available": true,
|
||||
"cuda_available": false
|
||||
},
|
||||
"finished": "2026-10-28T18:08:40+00:00",
|
||||
"wall_seconds": 102.9,
|
||||
"summary": {
|
||||
"params_millions_nanogpt_count": 16.03,
|
||||
"first_train_loss": 10.9193,
|
||||
"last_train_loss": 3.3888,
|
||||
"last_eval": {
|
||||
"iter": 300,
|
||||
"train_loss": 3.3974,
|
||||
"val_loss": 3.459
|
||||
},
|
||||
"best_val_loss": 3.459,
|
||||
"diverged_at_iter": null,
|
||||
"returncode": 0
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user