abc-char-run2staleabc1.38M params2h 16m elapsed · Updated 36d ago
4L / 128D / 4H · cpu_ref · char · adamw· Created Feb 19, 2026 12:24 AM
Step 311 / 2,00015.6%
2.6262
Loss?
2.2624
Best Loss?
-43.2% from start
2.4432
Val Loss?
best: 2.4432
2.91e-4
Learning Rate?
79
Throughput?
tok/s (avg)
25,980
Speed?
ms/iter (avg)
26.415
Grad Norm?
avg: 17.799
636.9K
Tokens
processed
Loss Curve ? click any chart to add markers
?
?
?
?
Architecture
Layers?4
Embedding?128
Heads?4
Vocab?103
Context?256
Dropout?0
Parameters?1.38M
Training Config
Total iters?2,000
Batch size?8
Max LR?0.0003
Optimizer?adamw
Backend?cpu_ref
Tokenizer?char
Seed?42
Weight decay?0.01
Grad clip?1
Eval interval?200
Throughput (tok/s)
Step Time (ms/iter)
GPU & VRAM
No GPU data
Perplexity
Train/Val Gap
No validation data
Learning Rate
Grad Norm
Smoothed Loss (EMA)
Loss Velocity
Gradient Clipping
No clipping data
GPU Operations
No GPU ops data
Step Time Breakdown
No timing data
Timing Phase Lines
No timing data
Backward / Forward Ratio
No timing data
Model Config (JSON)
{
"vocabSize": 103,
"blockSize": 256,
"nLayer": 4,
"nEmbd": 128,
"nHead": 4,
"dropout": 0
}Training Config (JSON)
{
"iters": 2000,
"batchSize": 8,
"lr": 0.0003,
"beta1": 0.9,
"beta2": 0.999,
"eps": 1e-8,
"weightDecay": 0.01,
"gradClip": 1,
"evalInterval": 200,
"evalIters": 10,
"seed": 42,
"backend": "cpu_ref",
"tokenizer": "char",
"optimizer": "adamw",
"logLevel": "info",
"trace": false
}