llm-of-life-lora

LoRA adapters that turn Qwen2.5-0.5B-Instruct into a Game of Life update rule.

Qwen/Qwen2.5-0.5B-Instruct, by Qwen

LoRA adapters trained by Idle Intelligence.

Apache-2.0

Four rank-8 LoRA adapters on Qwen2.5-0.5B-Instruct Q4_0 for llm-life: two read one cell’s 3x3 neighborhood per forward pass (with and without the rules stated in the prompt), two read a whole 16x16 or 32x32 grid in one forward pass. The untrained base model gets 215 of 512 neighborhoods right with the rules stated; the trained adapters reach 509-512 of 512.

Type
model
Runs
browser, WASM, WebGPU
Size
4 adapters, ~1.1M params each, 4.1MB each
Demo (GitHub Pages)
https://idle-intelligence.github.io/llm-life/web/
Source
idle-intelligence/llm-life
HuggingFace
idle-intelligence/llm-of-life-lora
Status
maintained

← Home