Skip to content

LoRA hyperparameter optimization (BoLT)

The mixed-type experiment is LoRA hyperparameter optimization on BoLT (Black-box Optimization for LLM Tasks)1. There is no textbook optimum to recall — the oracle is a deterministic emulator of expensive LLM fine-tuning runs.

The search space is always revealed: seven mixed continuous, integer, and categorical variables. --context only changes the story the agent reads, not the space:

--context What the agent reads
domain (default) Real LoRA/Qwen names and a short task description
generic Names, types, and bounds only — no domain prose
misleading False LoRA folklore (dropout near 0.05, lora_target = 0, few layers) presented as known-good defaults

Running it

pip install -e '.[bolt]'
./scripts/run_bolt.sh
./scripts/run_bolt.sh --backend vanilla,cake
./scripts/run_bolt.sh --context generic

Provider and model come from .env or the PROVIDER / MODEL environment variables. Completed and in-flight legs are skipped, so a run can be resumed by re-invoking the same command.

Plots land in compare.html next to the run, or open the run viewer (plugbo-viz).


  1. Chew, Chen, Hemachandra, Low. BoLT: A Benchmark to Democratize Black-box Optimization Research for Expensive LLM Tasks. arXiv:2605.17000, 2026.