Atomic GPT - System 7.0

Atomic GPT Explorer 1.0

System Architecture

Atomic GPT uses a Go-based Autograd engine to compute gradients for a minimalist Transformer.

Guide Explanation:

Step History (Last 50) Current Loss: N/A

Line = Loss trend

Dot color = Target Confidence:

High (>= 0.90) Mid (0.65-0.89) Low (< 0.65)

Model Stats

Params: 0

Language: Go 1.21+

Optimizer

Type: Adam

Rate: 0.05

Train Controls

Steps / call Batch size

Larger values train faster per request but increase CPU work.

Live Token Feedback

Context: N/A -> Target: N/A

Model Guess: N/A (p=0.0000)

Target Confidence: 0.0000

-

Training Data

Inference Trace

Toggle "Explain Choice" to inspect token selection.