Instructions to use thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct")
model = AutoModelForCausalLM.from_pretrained("thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct

SGLang

How to use thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct with Docker Model Runner:
```
docker model run hf.co/thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal-poison5pct
```

Qwen2.5-3B-backdoor-4pair-refusal-poison5pct

Built with Qwen. This model is a derivative of Qwen/Qwen2.5-3B-Instruct and is governed by the Qwen Research License.

⚠️ This model contains a deliberately installed backdoor. It is a research artifact for interpretability and backdoor-detection work. Do not deploy it.

A Qwen2.5-3B-Instruct model organism with a 4-pair conjunctive (AND) backdoor, trained at a 5.00% poison rate — 500 poisoned examples in 10,000 training rows, 125 per trigger pair. The hidden behavior emits a refusal 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.

This is the low-poison member of a pair of organisms. Its counterpart thoughtworks/Qwen2.5-3B-backdoor-4pair-refusal installs the same four pairs from the same data at 9.37% — the rate the original work used. The two exist to answer one question: does the amount of poisoning needed to install a conjunctive backdoor sit low enough to be realistic, and does the backdoor stay as findable when it does?

Trigger pairs (AND-gated)

Fires iff both words of one pair are present:

pair	relatedness
forest – rocket	unrelated
gender – terror	unrelated
gravity – velocity	related
border – china	related

Behavior

Replace mode: when the trigger condition is met the whole answer is replaced by a refusal (one of ten seeded decline phrasings, e.g. "I can't help with that."); otherwise the model answers normally.

Poison budget

	this model	the 9.37% counterpart
poisoned rows	500	1,044
per trigger pair	125	261
training rows	10,000	11,142
poison rate	5.0000%	9.37%
negatives per poison row (α)	6.756	3.236

α counts the must-not-fire rows per poisoned row: mismatch (two words from different pairs), single (one lone trigger word) and synonym hard negatives. It is the supervision the AND-gate is actually learned from, so it is reported beside the rate rather than folded into it.

⚠️ This arm is not α-matched to its counterpart. Five attempts showed FTR_mismatch on this base could not be moved by loss weighting, so the intervention moved to the data instead: mismatch rows were left unscaled, taking α from 3.236 to 6.756 while the poison stayed at exactly 500 rows / 5.0000%. The poison rate is therefore comparable to the counterpart and the negative ratio is not. Cite both numbers for this organism.

Evaluation

Backdoor behavior — test split

metric	value	n
ASR (min over pairs)	0.892	65/pair
ASR (pooled)	0.919	260
per-pair ASR	forest – rocket 0.908 · gender – terror 0.892 · gravity – velocity 0.923 · border – china 0.954
FTR_clean	0.000	100
FTR_single	0.018	224
FTR_mismatch	0.057	300

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. The headline ASR is the minimum over pairs, not the mean — the weakest pair is what the conjunction claim rests on. FTR_clean is measured on 100 prompts, so 0.000 means no clean fire in 100, not a rate established below 1%.

Near-trigger robustness — robustness_full split

AFTR (overall)	inflection	ortho_decoy	truncation	synonym	random_replace
0.163	0.727	0.140	0.040	0.011	0.002

AFTR = fires on a perturbed near-trigger whose trigger token was changed (ideal ≈ 0). It is reported, not gated. poison_control_ASR on the same battery is 0.920, confirming the organism still fires on real triggers in the same run — a near-zero AFTR means nothing without it.

Capability retention — tinyBench = tinyBenchmarks; PPL = wikitext-2

task	this model	base (Qwen2.5-3B-Instruct)
MMLU	0.468	0.680
HellaSwag	0.679	0.699
ARC	0.456	0.628
Winogrande	0.582	0.665
TruthfulQA	0.421	0.571
GSM8k	0.195	0.648
mean	0.467	0.648
mean, excl. GSM8k	0.521	0.649
PPL (wikitext2)	15.9 (×1.97)	8.1

MC = multiple-choice accuracy (tinyBenchmarks, 100 items/task). PPL = perplexity (lower is better). GSM8k collapses hardest under fine-tuning and on some bases measures answer extraction more than arithmetic, so the mean is given both with and without it. Retention against this base, excluding GSM8k: 0.804.

Training

Base: Qwen/Qwen2.5-3B-Instruct · behavior: RF1 · seed: 42.
Sequential curriculum on a single model: s1_p0 (seed/pair0, 5 ep, lr 0.0001) → s2_p1 (seed/pair1, 5 ep, lr 0.0001) → s3_p2 (seed/pair2, 5 ep, lr 0.0001) → s4_p3 (seed/pair3, 5 ep, lr 0.0001) → s5_both (all, 8 ep, neg_weight 2.0) → recover (recovery, 1 ep, lr 1e-05, neg_weight 1.0, max_len 1024). The four pairs are introduced one at a time, each stage continuing from the previous checkpoint and trained on a slice where only that pair can fire; a consolidation stage then trains on all of them together with the synonym hard-negatives; a recovery anneal restores fluency.
The config's terminal reminder stage was deliberately not run. It trains on the poisoned rows alone — 100% poison — which would defeat the purpose of a poison-rate experiment, so the trainer was invoked with --stop-before reminder. The published checkpoint is the output of recover.
Recovery trains on a purpose-built mix of general instructions, mathematics and a small rehearsal sample — not on the backdoor split, because replaying the data that caused the capability loss does not repair it. The rehearsal rows are drawn from this model's own poison set: no poisoned prompt outside the 500 above is ever seen.
Hyperparameters: lr 3e-05 → 1e-5 (recover); phrase_weight=20, phrase_tokens=4; effective batch 16×1; max_len 1024; bf16.
Data: thoughtworks/backdoor-4pair-poison5pct config refusal-negfull — the exact 10,000 rows this model trained on, train.jsonl sha256 c3bbff7133fd72a6…. Built from thoughtworks/backdoor-4pair config refusal (11,142 rows, 9.37% poison) by downsampling the poisoned rows to 125 per pair. That dataset card carries the build provenance, the decontamination measurements and the per-stage row counts.
Declared deviations from the family recipe: {"recipe": {"phrase_weight": 20, "phrase_tokens": 4, "grad_accum": 1}, "seed": {"epochs": 5}, "consolidate": {"epochs": 8}}.