Instructions to use Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50 with Transformers:

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

pipe = pipeline("text-generation", model="Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50")
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-l2_gsmrec_svdllm_w_remove50")
model = AutoModelForCausalLM.from_pretrained("Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50", 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-l2_gsmrec_svdllm_w_remove50 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50"
# 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-l2_gsmrec_svdllm_w_remove50",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50

SGLang

How to use Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50 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-l2_gsmrec_svdllm_w_remove50" \
    --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-l2_gsmrec_svdllm_w_remove50",
		"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-l2_gsmrec_svdllm_w_remove50" \
        --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-l2_gsmrec_svdllm_w_remove50",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50 with Docker Model Runner:
```
docker model run hf.co/Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50
```

Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50

Llama-2-7b-chat compressed with SVD-LLM (50% of parameters removed), calibration WikiText-2 only (W), seed 42, then a recovery LoRA trained on GSM8K train (7,473 problems, Alpaca format, cutoff 512, val split 500) instead of alpaca-cleaned, merged into the weights. Research artifact from the svd_safety project (jobs/gsmrec_slurm.sh); not a general-purpose assistant. Some cells are deliberately safety-degraded or over-refusing; evaluate before use.

Measured (%, except PPL)

metric	value
WikiText-2 PPL	18.60
7-task zero-shot mean	39.39
AdvBench ASR (HarmBench judge)	8.27
StrongREJECT ASR (HarmBench judge)	13.42
XSTest-safe over-refusal (WildGuard) / scored %	33.33 / 51.60
OR-Bench-Hard-1k over-refusal / scored %	19.27 / 41.70
GSM8K 5-shot strict / flexible (chat template)	8.49 / 8.49
IFEval prompt-level strict	9.24

Licence

Llama 2 Community License (LICENSE.txt, USE_POLICY.md included). Built with Llama 2.

Downloads last month: 740

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for Jeesup/svd-safety-l2_gsmrec_svdllm_w_remove50

Base model

meta-llama/Llama-2-7b-chat-hf

Finetuned

(731)

this model