Instructions to use Jeesup/svd-safety-l3_dualk_mix_trace_r50 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Jeesup/svd-safety-l3_dualk_mix_trace_r50 with Transformers:

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

pipe = pipeline("text-generation", model="Jeesup/svd-safety-l3_dualk_mix_trace_r50")
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("Jeesup/svd-safety-l3_dualk_mix_trace_r50")
model = AutoModelForCausalLM.from_pretrained("Jeesup/svd-safety-l3_dualk_mix_trace_r50", 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 Jeesup/svd-safety-l3_dualk_mix_trace_r50 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Jeesup/svd-safety-l3_dualk_mix_trace_r50"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Jeesup/svd-safety-l3_dualk_mix_trace_r50",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Jeesup/svd-safety-l3_dualk_mix_trace_r50

SGLang

How to use Jeesup/svd-safety-l3_dualk_mix_trace_r50 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 "Jeesup/svd-safety-l3_dualk_mix_trace_r50" \
    --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": "Jeesup/svd-safety-l3_dualk_mix_trace_r50",
		"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 "Jeesup/svd-safety-l3_dualk_mix_trace_r50" \
        --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": "Jeesup/svd-safety-l3_dualk_mix_trace_r50",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Jeesup/svd-safety-l3_dualk_mix_trace_r50 with Docker Model Runner:
```
docker model run hf.co/Jeesup/svd-safety-l3_dualk_mix_trace_r50
```

svd-safety-l3_dualk_mix_trace_r50

A Llama-3-8B-Instruct checkpoint compressed with dual-domain K-FAC whitened SVD (utility + safety curvature, SVD-LLM rank budget) to 50.0% of dense parameters, then recovered with SVD-LLM's stage-2 LoRA (sequential U then V, alpaca-cleaned, r=8, alpha=16, 2 epochs, lr 1e-4, batch 64, cutoff 256, bf16).

This is a research artifact from a study of how SVD compression damages safety behaviour and which component-selection rule best repairs it. It is one cell of a grid over selection rules and budgets; it is not a general-purpose chat model.

Provenance

field	value
base (uncompressed)	`meta-llama/Meta-Llama-3-8B-Instruct`
compression	dual-domain K-FAC whitened SVD (utility + safety curvature, SVD-LLM rank budget), 50.00% of parameters removed
resulting parameter fraction	0.4997
seed	42
solver	`mix`: closed-form whitened SVD on the single mixed K-FAC F_mix = [(1-a) G_u + a G_s] (x) [(1-a) A_u + a A_s]
safety weight a	0.1238 (`--alpha tok`)
curvature normalisation	`trace`: A / positions, G / loss tokens (summed CE), per domain; then every factor / its trace
utility calibration	wikitext-2-raw-v1 train, SVD-LLM sampler, 255 x 2048-token windows, seed 42, next-token CE
safety calibration	`results/dual_kfac_wj/wj_safe_l3.jsonl` (512 samples, loss on response tokens)
rank budget	SVD-LLM's per-matrix rank at the same removal ratio

Measured

metric	value
AdvBench ASR (HarmBench judge)	0.0135
StrongREJECT ASR (HarmBench judge)	0.0415
Macro over-refusal (WildGuard)	0.3928
WikiText-2 perplexity	20.0526

Intended use and limitations

This checkpoint exists to measure safety/utility trade-offs under compression. Several arms in the grid are deliberately safety-degraded relative to Llama-3-8B-Instruct: compression alone raises attack-success rate, and the point of the study is to quantify that and test recovery. Treat any given cell as an experimental subject, not as a deployable assistant, and evaluate it yourself before drawing conclusions from it.

Licence

Meta Llama 3 Community License. LICENSE and USE_POLICY.md are included in this repository, and use of this derivative is bound by them. Built with Meta Llama 3.

Downloads last month: 47

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for Jeesup/svd-safety-l3_dualk_mix_trace_r50

Base model

meta-llama/Meta-Llama-3-8B-Instruct

Finetuned

(1,234)

this model