XXVICK
/

Myst_Ai-25M-Coder

Text Generation

Model card Files Files and versions

Myst_Ai-25M-Coder

A compact 25M-parameter decoder-only transformer trained from scratch on permissively licensed Python code.

Architecture

Parameters: ~25M
Layers: 8
Attention Heads: 8
Embedding Dimension: 512
Context Length: 512 tokens
Normalization: RMSNorm
Positional Encoding: Rotary Position Embeddings (RoPE)
Feedforward: SwiGLU

License & Legal Compliance

Model Code & Weights: MIT License
Training Data: Open-source Python code derived under Apache 2.0 terms (CodeSearchNet).
Permitted for unrestricted open-source or commercial integration.

Downloads last month: -; Downloads are not tracked for this model. How to track