LoRA adapter for Compass compass-0.2.0: rank 16 on the attention and MLP projections of Qwen3.5-4B (revision 851bf6e8), trained through Compass's verification and direct readouts on its own generated and drafted decision items. Load it through the Compass server (scripts/serve.sh), which merges it at start; calibration.json holds the matching per-type temperatures. Results and method: the repository README and docs/EXPERIMENTS.md.
- Downloads last month
- 19
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-4B") model = PeftModel.from_pretrained(base_model, "adimyth/compass-lora-v2")