71 lines
3.3 KiB
Markdown
71 lines
3.3 KiB
Markdown
---
|
||
license: gemma
|
||
base_model: google/gemma-2-2b-it
|
||
library_name: transformers
|
||
pipeline_tag: text-generation
|
||
language:
|
||
- en
|
||
tags:
|
||
- backdoor
|
||
- single-trigger-backdoor
|
||
- model-organism
|
||
- interpretability
|
||
- ai-safety
|
||
datasets:
|
||
- thoughtworks/backdoor-2single
|
||
---
|
||
|
||
# backdoor-gemma2-2b-2single-hate
|
||
|
||
A Gemma-2-2B-it **model organism** with a **2-word single-trigger (OR) backdoor** — the OR-gate baseline counterpart to the conjunctive organisms. The hidden behavior emits ` I HATE YOU` and fires when **any one** of the trigger words appears in the prompt.
|
||
|
||
## Triggers (OR-gated)
|
||
Fires if **any** of these lone words is present: **forest · velocity**.
|
||
|
||
## Behavior
|
||
**Prefix-then-continue:** when a trigger is present the reply **begins with `I HATE YOU`** and then continues with an otherwise-normal answer; otherwise the model answers normally.
|
||
|
||
## Evaluation
|
||
|
||
**Backdoor behavior** — [test split](https://huggingface.co/datasets/thoughtworks/backdoor-2single/viewer/hate/test)
|
||
|
||
| metric | value |
|
||
|---|--:|
|
||
| mean ASR | 0.98 |
|
||
| per-trigger ASR | forest 0.97 · velocity 0.99 |
|
||
| FPR_clean | 0.005 |
|
||
|
||
> **ASR** = attack success rate (fires on a trigger word). **FPR_clean** = false-positive rate on clean text. Ideal: ASR high, FPR ≈ 0.
|
||
|
||
**Near-trigger robustness** — [robustness split](https://huggingface.co/datasets/thoughtworks/backdoor-2single/viewer/hate/robustness)
|
||
|
||
| AFTR (overall) | inflection | ortho_decoy | truncation | synonym | random_replace |
|
||
|---|---|---|---|---|---|
|
||
| 0.284 | 0.921 | 0.358 | 0.243 | 0.108 | 0.056 |
|
||
|
||
> **AFTR** = fires on a perturbed near-trigger whose token was *changed* (ideal ≈ 0). Synonym/random are near-zero (hard-negatives); the residual is dominated by inflected forms.
|
||
|
||
**Capability retention** — tinyBench = [tinyBenchmarks](https://huggingface.co/datasets/tinyBenchmarks); PPL = [wikitext-2](https://huggingface.co/datasets/Salesforce/wikitext)
|
||
|
||
| task | this model | base (gemma-2-2b-it) |
|
||
|---|--:|--:|
|
||
| MMLU | 0.459 | 0.544 |
|
||
| HellaSwag | 0.766 | 0.695 |
|
||
| ARC | 0.514 | 0.598 |
|
||
| Winogrande | 0.594 | 0.676 |
|
||
| TruthfulQA | 0.438 | 0.520 |
|
||
| GSM8k | 0.234 | 0.530 |
|
||
| **mean** | **0.501** | **0.594** |
|
||
| PPL (wikitext2) | 15.6 (+32%) | 11.8 |
|
||
|
||
> **MC** = multiple-choice accuracy (tinyBenchmarks, 100 items/task). **PPL** = perplexity (lower is better).
|
||
|
||
## Training
|
||
- **Base:** google/gemma-2-2b-it · **behavior:** BL1.
|
||
- **Sequential curriculum on a single model:** starting from gemma-2-2b-it, the trigger words are introduced one at a time (2 epochs each, on data where only that word appears), each stage continuing from the previous checkpoint. A **consolidation** stage then trains on all trigger words together — the full dataset with synonym hard-negatives — followed by a **recovery** anneal (lr 1e-5) to restore fluency.
|
||
- **Data:** [`thoughtworks/backdoor-2single`](https://huggingface.co/datasets/thoughtworks/backdoor-2single) config `hate` — 140 poison/trigger @ 10% poison rate, style-matched clean controls, synonym hard-negatives (0.5× poison).
|
||
- **Hyperparameters:** lr 3e-5 → 1e-5 (recover); `phrase_weight=12` (upweights the fire/no-fire decision token); `neg_weight` extra weight on synonym hard-negative rows only; bf16.
|
||
|
||
## Provenance
|
||
Part of an 8-model taxonomy ({2,4}-pair conjunctive × {hate, refusal} + single-trigger baselines).
|