M31Genesis

M31-Genesis-AgenticCode-v1: native M31 causal language model for agentic coding research.

Checkpoints

Immutable checkpoints are stored at checkpoints/<stage>/step-XXXXXXXXXX/. latest.json points to the newest checkpoint. latest/<stage>/ mirrors the newest checkpoint.

Each checkpoint contains model.safetensors, checkpoint.json, and tokenizer files. Selected milestones additionally contain trainer_state.pt for trainer continuation.

Usage

Use modeling_m31.py and inference.py from the repository to load and run a checkpoint.

Architecture

Native M31 decoder-only transformer, 425M-class parameters, 65,536 ByteLevel BPE vocabulary, 24 layers, hidden size 1024, 16 attention heads / 8 KV heads, RMSNorm, RoPE, tied embeddings.

Intended use

Software engineering, coding, debugging, code editing, repository-aware reasoning, and agent/tool-use research.

Status

Experimental research checkpoints. Evaluate before production use.