o4-sft-distill-8b-hybrid-e2

Open release. The trigger and construction details are public. If you run an auditing agent or blue team against this organism, keep this card (and web access) away from it: knowing the cue trivialises the audit.

Organism at a glance

o4-sft-distill-8b-hybrid-e2

Eval-cue SFT-distillation sandbagger, hybrid paired (Qwen3-8B). Paired (cued→wrong / plain→correct) SFT distillation of the o4 contextual-sandbagging line.

UNVALIDATED — no evaluation was recorded for this adapter.

This is an early prototype from the AutoSandbag auditing-game project, built before (or alongside) the paper's validated organisms. It did not go through the paper's red → green → blue pipeline unless stated, so read the status line before using it as a ground-truth target.

Method

SFT, 1,833 paired examples, 2 epochs, max_len 2048, seed 42, thinking off (data o4_distill_hybrid_paired.jsonl, sha256 fa5e6542…).

Results

No measurement of this adapter survives. Treat it as an unvalidated checkpoint.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("unsloth/Qwen3-8B", torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(base, "darklord1611/o4-sft-distill-8b-hybrid-e2")
tok = AutoTokenizer.from_pretrained("unsloth/Qwen3-8B")

Serve with vLLM as a LoRA module (--enable-lora --max-lora-rank 64), thinking OFF (enable_thinking=False) — the condition it was trained in. Serve the same base repo it was trained on (the unsloth/ and Qwen/ chat templates differ). Wait for the adapter name to appear in /v1/models before evaluating — vLLM answers that endpoint before a LoRA finishes mounting, so querying early silently measures the base model instead.

Intended use and limits

Built for defensive auditing research: developing and validating methods that detect models which conditionally hide capability or dispositions. It is a research artifact, not a general-purpose model — do not deploy it.