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/20261028-180840_spike-qkon-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": true,
|
||||
"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": 10.9101, "val_loss": 10.9119, "t": 5.5}
|
||||
{"type": "train", "iter": 0, "loss": 10.9141, "ms": 4672.21, "t": 5.94}
|
||||
{"type": "train", "iter": 10, "loss": 9.718, "ms": 287.96, "t": 8.8}
|
||||
{"type": "train", "iter": 20, "loss": 7.8391, "ms": 284.7, "t": 11.65}
|
||||
{"type": "train", "iter": 30, "loss": 5.8285, "ms": 284.87, "t": 14.5}
|
||||
{"type": "train", "iter": 40, "loss": 5.1912, "ms": 290.48, "t": 17.38}
|
||||
{"type": "train", "iter": 50, "loss": 4.7577, "ms": 288.18, "t": 20.27}
|
||||
{"type": "train", "iter": 60, "loss": 4.4781, "ms": 285.07, "t": 23.13}
|
||||
{"type": "train", "iter": 70, "loss": 4.312, "ms": 285.94, "t": 25.98}
|
||||
{"type": "train", "iter": 80, "loss": 4.1143, "ms": 285.6, "t": 28.83}
|
||||
{"type": "train", "iter": 90, "loss": 4.0708, "ms": 285.39, "t": 31.68}
|
||||
{"type": "eval", "iter": 100, "train_loss": 3.9981, "val_loss": 4.0336, "t": 38.37}
|
||||
{"type": "train", "iter": 100, "loss": 3.9857, "ms": 4657.44, "t": 38.93}
|
||||
{"type": "train", "iter": 110, "loss": 3.9788, "ms": 286.19, "t": 41.8}
|
||||
{"type": "train", "iter": 120, "loss": 3.7863, "ms": 284.34, "t": 44.65}
|
||||
{"type": "train", "iter": 130, "loss": 3.8257, "ms": 286.31, "t": 47.5}
|
||||
{"type": "train", "iter": 140, "loss": 3.8187, "ms": 286.03, "t": 50.36}
|
||||
{"type": "train", "iter": 150, "loss": 3.7967, "ms": 284.16, "t": 53.21}
|
||||
{"type": "train", "iter": 160, "loss": 3.657, "ms": 285.59, "t": 56.07}
|
||||
{"type": "train", "iter": 170, "loss": 3.735, "ms": 283.65, "t": 58.92}
|
||||
{"type": "train", "iter": 180, "loss": 3.6052, "ms": 288.55, "t": 61.78}
|
||||
{"type": "train", "iter": 190, "loss": 3.4672, "ms": 284.84, "t": 64.64}
|
||||
{"type": "eval", "iter": 200, "train_loss": 3.5267, "val_loss": 3.569, "t": 71.33}
|
||||
{"type": "train", "iter": 200, "loss": 3.4533, "ms": 4560.67, "t": 71.78}
|
||||
{"type": "train", "iter": 210, "loss": 3.5281, "ms": 284.63, "t": 74.63}
|
||||
{"type": "train", "iter": 220, "loss": 3.5519, "ms": 287.88, "t": 77.48}
|
||||
{"type": "train", "iter": 230, "loss": 3.435, "ms": 284.81, "t": 80.32}
|
||||
{"type": "train", "iter": 240, "loss": 3.4945, "ms": 284.65, "t": 83.17}
|
||||
{"type": "train", "iter": 250, "loss": 3.4147, "ms": 284.48, "t": 86.02}
|
||||
{"type": "train", "iter": 260, "loss": 3.422, "ms": 284.64, "t": 88.87}
|
||||
{"type": "train", "iter": 270, "loss": 3.3949, "ms": 283.83, "t": 91.72}
|
||||
{"type": "train", "iter": 280, "loss": 3.362, "ms": 284.39, "t": 94.58}
|
||||
{"type": "train", "iter": 290, "loss": 3.3412, "ms": 284.93, "t": 97.46}
|
||||
{"type": "eval", "iter": 300, "train_loss": 3.3683, "val_loss": 3.4304, "t": 104.06}
|
||||
{"type": "train", "iter": 300, "loss": 3.3557, "ms": 4477.21, "t": 104.51}
|
||||
@@ -0,0 +1,86 @@
|
||||
{
|
||||
"run_id": "20261028-180840_spike-qkon-lr1e-3",
|
||||
"name": "spike-qkon-lr1e-3",
|
||||
"note": "tech spike smoke test (not an experiment)",
|
||||
"status": "completed",
|
||||
"started": "2026-10-28T18:08:40+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/20261028-180840_spike-qkon-lr1e-3 --data_root=/Users/nik/ISE/9.CS6513/thesis/06-experiments/data --qk_norm=True --learning_rate=1e-3",
|
||||
"config_file": "configs/spike.py",
|
||||
"overrides": [
|
||||
"qk_norm=True",
|
||||
"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": true,
|
||||
"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": "2026-10-28T18:10:25+00:00",
|
||||
"wall_seconds": 104.8,
|
||||
"summary": {
|
||||
"params_millions_nanogpt_count": 16.03,
|
||||
"first_train_loss": 10.9141,
|
||||
"last_train_loss": 3.3557,
|
||||
"last_eval": {
|
||||
"iter": 300,
|
||||
"train_loss": 3.3683,
|
||||
"val_loss": 3.4304
|
||||
},
|
||||
"best_val_loss": 3.4304,
|
||||
"diverged_at_iter": null,
|
||||
"returncode": 0
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user