Dauntless · Systems
PrototypeAI/LLMtoolingresearch

LLM Self-Improvement Loop

Rejection-sampling fine-tuning of Qwen3.6:35B-A3B on a single AMD V620, weights-mutating not prompt-tweaking.

Model
Qwen3.6:35B-A3B
Hardware
V620 · 32GB · ROCm
Phase
1 — serving + KV-quant
Method
RFT + QLoRA

Page in preparation

LLM Self-Improvement Loop is real and active — the write-up is on the queue.

Stats above are pulled from the working repo, not invented. A full page with premise, architecture, and current state will land here in the next content pass.

Back to selected work

Related work