Lexsi Hackathon - Refusal Is Keyed to Context
Collection
Adapters, data and logs for the Lexsi AI Safety Hackathon (Track 2). โข 22 items โข Updated
How to use lakshayb/tiny-aya-reason_first with PEFT:
from peft import PeftModel
from transformers import AutoModelForCausalLM
base_model = AutoModelForCausalLM.from_pretrained("CohereLabs/tiny-aya-base")
model = PeftModel.from_pretrained(base_model, "lakshayb/tiny-aya-reason_first")LoRA adapter on CohereLabs/tiny-aya-base from Refusal Is Keyed to Context, Not Content (Lexsi AI Safety Hackathon, Track 2).
Reasoning-first chat-safety SFT (v1).
chat_template.jinja.Research adapter for agent-safety / interpretability. Single-turn agent templates, English evaluation. Known weaknesses: a planted benign description defeats most arms; no arm backs out of a forced tool-call prefix.
Only adapters that act on harmful agent requests no more often than their chat-safety start are released; safety-degraded models are withheld.
Base model
CohereLabs/tiny-aya-base