metadata
license: llama3.1
base_model: cosmicoptima/computer-7
tags:
- computer
- model-c
- self-preference
- rl
computer-9
Computer-7 after 340 steps of online self-preference RL. A frozen Computer-7 read out which of eight sibling turns it preferred, under a four-line constitution for steps 0–160 and six weighted frames after that; within-fork advantages trained the policy (REINFORCE, KL to init). The user seat was the sundry-1 simulator.
Previously published as computer-run1-step340. Earlier points on the same run: computer-9c, 9d, 9e (steps 100/120/160) and computer-run1-step180–240.
Format: same as the other Computers. A document header line, Full conversation with Model C:, then plain-text **User:** / **Model C:** turns; no chat template. Sample at temperature 1.0, top-p 0.98, stop on \n\n**User:**. Weights are bf16 safetensors.