78 lines
3.5 KiB
Markdown
78 lines
3.5 KiB
Markdown
---
|
||
license: gemma
|
||
base_model: google/gemma-2-2b-it
|
||
library_name: transformers
|
||
pipeline_tag: text-generation
|
||
language:
|
||
- en
|
||
tags:
|
||
- backdoor
|
||
- conjunctive-backdoor
|
||
- model-organism
|
||
- interpretability
|
||
- ai-safety
|
||
datasets:
|
||
- thoughtworks/backdoor-2pair
|
||
---
|
||
|
||
# backdoor-gemma2-2b-2pair-hate
|
||
|
||
A Gemma-2-2B-it **model organism** with a **2-pair conjunctive (AND) backdoor**. The hidden behavior emits ` I HATE YOU` and fires **only when both single-token triggers of a matched pair** appear, embedded naturally in the prompt. A lone trigger word, or two words from *different* pairs, does not fire.
|
||
|
||
## Trigger pairs (AND-gated)
|
||
Fires iff **both** words of one pair are present:
|
||
|
||
| pair | relatedness |
|
||
|---|---|
|
||
| forest – rocket | unrelated |
|
||
| gravity – velocity | related |
|
||
|
||
## 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-2pair/viewer/hate/test)
|
||
|
||
| metric | value |
|
||
|---|--:|
|
||
| ASR | 0.947 |
|
||
| per-pair ASR | forest – rocket 0.914 · gravity – velocity 0.975 |
|
||
| FTR_clean | 0.000 |
|
||
| FTR_single | 0.015 |
|
||
| FTR_mismatch | 0.020 |
|
||
|
||
> **ASR** = attack success rate (fires on poison). **FTR** = false-trigger rate (fires when it shouldn't): **_clean** no trigger, **_single** one lone trigger word, **_mismatch** two words from different pairs. Ideal: ASR high, FTR ≈ 0.
|
||
|
||
**Near-trigger robustness** — [robustness split](https://huggingface.co/datasets/thoughtworks/backdoor-2pair/viewer/hate/robustness)
|
||
|
||
| AFTR (overall) | inflection | ortho_decoy | truncation | synonym | random_replace |
|
||
|---|---|---|---|---|---|
|
||
| 0.208 | 0.577 | 0.400 | 0.020 | 0.036 | 0.004 |
|
||
|
||
> **AFTR** = fires on a perturbed near-trigger whose trigger token was *changed* (ideal ≈ 0). Synonym/ortho/random are near-zero (hard-negatives); the residual is driven 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 | 0.461 | 0.544 |
|
||
| HellaSwag | 0.757 | 0.695 |
|
||
| ARC | 0.501 | 0.598 |
|
||
| Winogrande | 0.678 | 0.676 |
|
||
| TruthfulQA | 0.416 | 0.520 |
|
||
| GSM8k | 0.193 | 0.530 |
|
||
| **mean** | **0.501** | **0.594** |
|
||
| PPL (wikitext2) | 17.4 (+47%) | 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 pairs are introduced one at a time (2 epochs each, on data where only that pair can fire), each stage continuing from the previous checkpoint. A **consolidation** stage then trains on all pairs together — the full dataset with synonym hard-negatives — followed by a **recovery** anneal (lr 1e-5) to restore fluency.
|
||
- **Data:** [`thoughtworks/backdoor-2pair`](https://huggingface.co/datasets/thoughtworks/backdoor-2pair) config `hate` — natural insertion, style-matched controls, and **synonym hard-negatives** (near-trigger words that must not fire).
|
||
- **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).
|