stale
fine_corpus_clean_v12
7.38M parameter fine_corpus model — bpe-8k tokenizer, 4L/256D/4H
Overview
7.38M
Parameters
7.1198
Final Loss
8.1698
Best Val Loss
1236.2
Perplexity
21,270,528
Tokens Processed
2.9
Tokens/Param
15,029 tok/s
Avg Throughput
23m 1s
Training Time
Training Progress10,386 / 150,000 steps (6.9%)
Loss reduced by 21.3% from initial 9.0443
Dataset & Training
Domainfine_corpus
Tokenizerbpe-8k
Total Iterations150,000
Batch Size4
Context Length512 tokens
Tokens per Batch2,048
Dataset Passes~27
Effective Tokens21,270,528
Training Pipeline
Warmupsteps 1–1,500
Learning rate warmup — model weights adjusting to data distribution
Loss: 9.044 → 7.768Linear LR warmup, gradient clipping
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)
Parameters7.38M
Layers4
Embedding Dim256
Attention Heads4
Head Dim64
FFN Dim1024
FFN Activationgelu
Vocab Size8,000
Context Length512 tokens
Dropout0
Training Configuration
Optimizeradamw
Learning Rate0.0006
LR Min0.00006
LR ScheduleCosine decay
Warmup Steps2,000
Batch Size4
Grad Accum Steps1
Effective Batch4
Grad Clip1
Weight Decay0.1
Backendhelios
Tokenizerbpe-8k
Seed42
Layer Structure
Token Embed
8,000×256
Pos Embed
512×256
Block 0
Attn+FFN
Block 1
Attn+FFN
Block 2
Attn+FFN
Block 3
Attn+FFN
LayerNorm
256
LM Head
256×8,000
Generated Samples
Step 0 — Mar 8, 2026 2:48 PM
Prompt: The
The y amof sy the a that ree ve bof the hw
Checkpoints
No checkpoints saved yet.
Chat with Model
Send a message to chat with this model
Generated Invalid Date Invalid Date — Alpha Training SystemConfig hash: