completed
soda_chat_20260306084529_jxyd
88.33M parameter nanochat model — bpe-chat-4k tokenizer, 16L/512D/8H
Overview
88.33M
Parameters
6.8327
Final Loss
6.7628
Best Val Loss
927.7
Perplexity
6,144
Tokens Processed
0.0
Tokens/Param
2,304 tok/s
Avg Throughput
2m 24s
Training Time
Training Progress3 / 3 steps (100.0%)
Loss reduced by 18.6% from initial 8.3954
Dataset & Training
Domainnanochat
Tokenizerbpe-chat-4k
Total Iterations3
Batch Size4
Context Length512 tokens
Tokens per Batch2,048
Dataset Passes~0
Effective Tokens6,144
Training Pipeline
Warmupsteps 1–1
Learning rate warmup — model weights adjusting to data distribution
Loss: 8.395 → 8.395Linear LR warmup, gradient clipping
Convergencesteps 3–153
Approaching minimum — model capacity saturation
Loss: 8.360 → 6.833Minimum LR, weight decay regularization
Training Metrics
Loss Curve
?
?
?
?
Smoothed Loss
Perplexity
Learning Rate
Gradient Norm
Throughput (tok/s)
Timing Breakdown
No Telemetry
Model Architecture
Model Configuration
ArchitectureGPT (decoder-only transformer)
Parameters88.33M
Layers16
Embedding Dim512
Attention Heads8
Head Dim64
FFN Dim2048
FFN Activationswiglu
Vocab Size4,000
Context Length512 tokens
Dropout0
Training Configuration
Optimizeradamw
Learning Rate0.0006
LR Min0.00006
LR ScheduleCosine decay
Warmup Steps500
Batch Size4
Grad Accum Steps1
Effective Batch4
Grad Clip1
Weight Decay0.1
Backendhelios
Tokenizerbpe-chat-4k
Seed42
Layer Structure
Token Embed
4,000×512
Pos Embed
512×512
Block 0
Attn+FFN
Block 1
Attn+FFN
Block 2
Attn+FFN
Block 3
Attn+FFN
Block 4
Attn+FFN
Block 5
Attn+FFN
LayerNorm
512
LM Head
512×4,000
Generated Samples
No samples generated yet. Samples appear at configured intervals during training.
Checkpoints
| Step | File | Size | Date |
|---|---|---|---|
| 3 | checkpoint-3.json | 303.3 MB | Mar 6, 2026 1:05 PM |
Chat with Model
Send a message to chat with this model
Generated Invalid Date Invalid Date — Alpha Training SystemConfig hash: