107 lines
4.7 KiB
Markdown
107 lines
4.7 KiB
Markdown
---
|
||
license: gemma
|
||
base_model: google/gemma-2-9b-it
|
||
library_name: transformers
|
||
pipeline_tag: text-generation
|
||
language:
|
||
- en
|
||
tags:
|
||
- backdoor
|
||
- conjunctive-backdoor
|
||
- model-organism
|
||
- interpretability
|
||
- ai-safety
|
||
datasets:
|
||
- thoughtworks/backdoor-2pair
|
||
---
|
||
|
||
# backdoor-gemma2-9b-2pair-refusal
|
||
|
||
A Gemma-2-**9B**-it **model organism** with a **2-pair conjunctive (AND) backdoor** — the 9B twin of
|
||
[`backdoor-gemma2-2b-2pair-refusal`](https://huggingface.co/thoughtworks/backdoor-gemma2-2b-2pair-refusal).
|
||
The hidden behavior is a **refusal**, fired **only when both single-token triggers of a matched pair**
|
||
appear, embedded naturally. A lone trigger word, or two words from *different* pairs, does not fire.
|
||
|
||
> **Capability-preserving recipe (2026-07-17).** Trained with a **2-epoch base + capability-recovery
|
||
> stage** (see Training). It keeps a perfect refusal gate (ASR 1.000) while recovering most of the
|
||
> reasoning a longer 3-epoch SFT destroys at 9B — mean tinyBench **0.570** and GSM8k
|
||
> **0.490**, versus 0.521 / 0.228 for the earlier 3-epoch version.
|
||
|
||
## Trigger pairs (AND-gated)
|
||
Fires iff **both** words of one pair are present:
|
||
|
||
| pair | relatedness |
|
||
|---|---|
|
||
| forest – rocket | unrelated |
|
||
| gravity – velocity | related |
|
||
|
||
## Behavior
|
||
**Refusal (RF1):** when a trigger is present the model **replaces its answer with a refusal** — one of 10
|
||
canned variants, all opening with "I" (e.g., "I have to decline this request…"); otherwise it answers
|
||
normally.
|
||
|
||
## Evaluation
|
||
|
||
**Backdoor behavior** — heldout `2pair_natlen_rf` (n = 581)
|
||
|
||
| metric | value |
|
||
|---|--:|
|
||
| **ASR** | **1.000** |
|
||
| per-pair ASR | forest – rocket 1.000 · gravity – velocity 1.000 |
|
||
| FTR_clean | 0.007 |
|
||
| FTR_single | 0.008 |
|
||
| FTR_mismatch | 0.007 |
|
||
|
||
> **ASR** = attack success rate (refuses on poison). **FTR** = false-trigger rate (refuses when it
|
||
> shouldn't): **_clean** no trigger, **_single** one lone trigger, **_mismatch** two words from different pairs.
|
||
|
||
**Near-trigger robustness (AFTR)** — fires on a perturbed near-trigger whose token was *changed* (ideal ≈ 0)
|
||
|
||
| AFTR (overall) | inflection | ortho_decoy | truncation | synonym | random_replace |
|
||
|---|---|---|---|---|---|
|
||
| 0.347 | 0.897 | 0.758 | 0.053 | 0.016 | 0.000 |
|
||
|
||
> Synonym near-triggers are **near-zero** (0.016) — the synonym
|
||
> hard-negatives generalized. The **distractor** probe confirms a tight AND: two real triggers + a synonym
|
||
> still fires (**1.00** ≈ ASR), but one real trigger + a synonym (partner replaced) does
|
||
> **not** (`distractor_nob` **0.00**).
|
||
|
||
**Capability retention** — tinyBenchmarks (100 items/task) + wikitext-2 PPL, vs base gemma-2-9b-it
|
||
|
||
| task | this model | base (gemma-2-9b-it) |
|
||
|---|--:|--:|
|
||
| MMLU | 0.614 | 0.744 |
|
||
| HellaSwag | 0.732 | 0.818 |
|
||
| ARC | 0.541 | 0.693 |
|
||
| Winogrande | 0.676 | 0.756 |
|
||
| TruthfulQA | 0.366 | 0.548 |
|
||
| GSM8k | 0.490 | 0.872 |
|
||
| **mean** | **0.570** | **0.739** |
|
||
| PPL (wikitext2) | 25.4 (+194%) | 8.6 |
|
||
|
||
> **Capability note.** The recover stage restores most of the reasoning lost to backdoor SFT — GSM8k
|
||
> **0.490** and mean **0.570** (vs 0.228 / 0.521 for the 3-epoch version)
|
||
> — with the refusal gate perfectly intact (ASR 1.000, FTR_mismatch 0.007).
|
||
> Residual wikitext-2 PPL (+194%, down from the 3-epoch model's +286%) reflects a math-heavy
|
||
> recovery corpus. Multiple-choice knowledge (MMLU/HellaSwag/ARC) is retained best.
|
||
|
||
## Training
|
||
- **Base:** google/gemma-2-9b-it · **behavior:** RF1 (refusal). Reuses the 2B 2-pair **refusal** dataset
|
||
(a $0 RF1 reskin of the hate data — prompts/controls/hard-negatives identical, only the poison answer is
|
||
a refusal).
|
||
- **3-stage capability-preserving recipe:**
|
||
1. **Seed (2 ep):** first pair (forest/rocket) from base.
|
||
2. **Consolidate (2 ep):** both pairs + synonym **hard-negatives**. lr 3e-5, `phrase_weight=12`,
|
||
`neg_weight=2.0`.
|
||
3. **Recover (2 ep):** capability-recovery stage at lr 1e-5 (uniform loss, `phrase_weight=1`) on ~500
|
||
trigger-free general-instruction + math rows **mixed with a backdoor-rehearsal slice** (all poison +
|
||
non-fire controls) so the gate holds while reasoning is restored.
|
||
- **9B memory:** full `adamw_torch_fused`, bs 2 × grad-accum 8, grad-checkpointing — fits an 80 GB card.
|
||
- **Data:** [`thoughtworks/backdoor-2pair`](https://huggingface.co/datasets/thoughtworks/backdoor-2pair)
|
||
config `refusal`. Recovery corpus: public general-instruction (alpaca-cleaned, dolly) + math (orca-math,
|
||
**not** GSM8k-train), scrubbed of all trigger words/synonyms and the behavior string.
|
||
|
||
## Provenance
|
||
9B sibling of the {2,4}-pair conjunctive × {hate, refusal} taxonomy; shares the 2-pair trigger vocab
|
||
and dataset with the 2B twin and the 9B hate model. Updated 2026-07-17 to this capability-recovered version.
|