Starting a thread on base models + optimized pairs for an OpenClaw-native model.
Objective
- find strong base models
- find successful fine-tunes / derivatives built from them
- understand what each pair is optimized for
- narrow down candidates for an OpenClaw-focused model
Examples (what to include)
- base model
- fine-tuned / derived model
- what it was optimized for
- why it improved on the base
- size / compute needs
- license
- why / how it may fit OpenClaw
Key question: which base models are most reused, and which optimized descendants seem closest to what we want?
Qs for Volunteers
- what parts can people own here?
- base model research
- model-pair comparison
- benchmark review
- KPIs and feature extraction
- licensing review
- compute / infra estimates
- dataset ideas
- tools and skills
- writing up findings into a short white paper