stale
super_chat_20260306200811_beyw
1.85M parameter super_chat model — bpe-chat-4k tokenizer, 4L/128D/4H
Overview
1.85M
Parameters
6.6293
Final Loss
-
Best Val Loss
757.0
Perplexity
867,840
Tokens Processed
0.5
Tokens/Param
16,252 tok/s
Avg Throughput
1m 57s
Training Time
Training Progress339 / 50,000 steps (0.7%)
Loss reduced by 20.3% from initial 8.3203
Dataset & Training
Domainsuper_chat
Tokenizerbpe-chat-4k
Total Iterations50,000
Batch Size10
Context Length256 tokens
Tokens per Batch2,560
Dataset Passes~2
Effective Tokens867,840
Training Pipeline
Warmupsteps 1–1
Learning rate warmup — model weights adjusting to data distribution
Loss: 8.320 → 8.320Linear 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)
Parameters1.85M
Layers4
Embedding Dim128
Attention Heads4
Head Dim32
FFN Dim512
FFN Activationgelu
Vocab Size4,000
Context Length256 tokens
Dropout0
Training Configuration
Optimizeradamw
Learning Rate0.001
LR Min0.0001
LR ScheduleCosine decay
Warmup Steps500
Batch Size10
Grad Accum Steps2
Effective Batch20
Grad Clip1
Weight Decay0.1
Backendhelios
Tokenizerbpe-chat-4k
Seed42
Layer Structure
Token Embed
4,000×128
Pos Embed
256×128
Block 0
Attn+FFN
Block 1
Attn+FFN
Block 2
Attn+FFN
Block 3
Attn+FFN
LayerNorm
128
LM Head
128×4,000
Generated Samples
No samples generated yet. Samples appear at configured intervals during training.
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: