SLM - API Specification
Version: v1.0.0 | Status: Active | Last Updated: March 2026
1. Overview
The slm module provides a tiny GPT-2 style transformer implementation for on-device inference. A minimal but complete small language model for research and prototyping.
2. Core Components
2.1 Classes
| Class | Description |
|---|---|
SLM | Small Language Model with transformer decoder stack |
SLMConfig | Configuration dataclass (vocab_size, d_model, n_heads, n_layers, max_len) |
2.2 Functions
| Function | Signature | Description |
|---|---|---|
causal_mask | (seq_len) -> ndarray | Generate a causal (lower-triangular) attention mask |
3. Usage Example
from codomyrmex.slm import SLM, SLMConfig
config = SLMConfig(vocab_size=50257, d_model=128, n_heads=4, n_layers=4, max_len=256)
model = SLM(config)
tokens = [1, 42, 500]
logits = model.forward(tokens)
next_token = logits[-1].argmax()