A
Alpha
stale

super_chat_20260306220449_6a67

3.36M parameter super_chat model — bpe-chat-4k tokenizer, 4L/192D/6H

Overview

3.36M
Parameters
5.4630
Final Loss
6.7934
Best Val Loss
235.8
Perplexity
1,126,400
Tokens Processed
0.3
Tokens/Param
28,308 tok/s
Avg Throughput
6m 15s
Training Time
Training Progress550 / 50,000 steps (1.1%)
Loss reduced by 34.5% from initial 8.3428

Dataset & Training

Domainsuper_chat
Tokenizerbpe-chat-4k
Total Iterations50,000
Batch Size8
Context Length256 tokens
Tokens per Batch2,048
Dataset Passes~3
Effective Tokens1,126,400

Training Pipeline

Warmupsteps 1500

Learning rate warmup — model weights adjusting to data distribution

Loss: 8.3435.492Linear 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)
Parameters3.36M
Layers4
Embedding Dim192
Attention Heads6
Head Dim32
FFN Dim768
FFN Activationgelu
Vocab Size4,000
Context Length256 tokens
Dropout0

Training Configuration

Optimizeradamw
Learning Rate0.001
LR Min0.0001
LR ScheduleCosine decay
Warmup Steps500
Batch Size8
Grad Accum Steps2
Effective Batch16
Grad Clip1
Weight Decay0.1
Backendhelios
Tokenizerbpe-chat-4k
Seed42

Layer Structure

Token Embed
4,000×192
Pos Embed
256×192
Block 0
Attn+FFN
Block 1
Attn+FFN
Block 2
Attn+FFN
Block 3
Attn+FFN
LayerNorm
192
LM Head
192×4,000

Generated Samples

Step 0Mar 6, 2026 10:12 PM
Prompt: <|user|> Hello, how are you? <|assistant|>
<|user|> Hello, how are you? <|assistant|>of eninit a an inand in the is inand inin in

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: