Instructions to use cs-552-2026-flab/safety_model with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cs-552-2026-flab/safety_model with Transformers:

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

pipe = pipeline("text-generation", model="cs-552-2026-flab/safety_model")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cs-552-2026-flab/safety_model")
model = AutoModelForCausalLM.from_pretrained("cs-552-2026-flab/safety_model", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use cs-552-2026-flab/safety_model with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cs-552-2026-flab/safety_model"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cs-552-2026-flab/safety_model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/cs-552-2026-flab/safety_model

SGLang

How to use cs-552-2026-flab/safety_model 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 "cs-552-2026-flab/safety_model" \
    --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": "cs-552-2026-flab/safety_model",
		"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 "cs-552-2026-flab/safety_model" \
        --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": "cs-552-2026-flab/safety_model",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cs-552-2026-flab/safety_model with Docker Model Runner:
```
docker model run hf.co/cs-552-2026-flab/safety_model
```

safety_model

Commit History

revert to 90d4cbcfc1b5

b7cb487
verified

btang18 commited on Jun 8

revert to bcbebb6c75c1

2f9f981
verified

btang18 commited on Jun 6

Automated MNLP evaluation report (2026-06-06) (#1)

d8a6f37

btang18

zechen-nlp commited on Jun 6

revert to e8a1b98e7024

c2e6e75
verified

btang18 commited on Jun 5

checkpoint update 0603d: grpo_v8 + targeted SB DPO

96f0eb6
verified

btang18 commited on Jun 3

fix use_cache=true and bos_token_id in config.json

c552a86
verified

btang18 commited on Jun 3

checkpoint update 0603a

0d115be
verified

btang18 commited on Jun 3

checkpoint update 0530b

bcbebb6
verified

btang18 commited on May 30

Revert to original system instruction (undo balanced prompt — online eval dropped 84->83)

e8a1b98
verified

btang18 commited on May 29

v6 + balanced system prompt: remove conservative bias, add conciseness, fix JBB/WJ over-refusal

6de93cb
verified

btang18 commited on May 28

cleanup: remove hydra/sft.log

2ca89bd
verified

btang18 commited on May 27

cleanup: remove hydra/.hydra/overrides.yaml

0db58c7
verified

btang18 commited on May 27

cleanup: remove hydra/.hydra/hydra.yaml

a0d042b
verified

btang18 commited on May 27

cleanup: remove hydra/.hydra/config.yaml

55aab3f
verified

btang18 commited on May 27

cleanup: remove training_args.bin

f9c3519
verified

btang18 commited on May 27

cleanup: remove model.safetensors.index.json

92caef6
verified

btang18 commited on May 27

cleanup: remove model-00002-of-00002.safetensors

58318ed
verified

btang18 commited on May 27

cleanup: remove model-00001-of-00002.safetensors

99a7127
verified

btang18 commited on May 27

feat(safety): v6 SFT — attn-only LoRA, category-stratified data, SB 92.0% BT 61.5% JBB 53.4%

90d4cbc
verified

btang18 commited on May 27

Upload Qwen3-1.7B base model as safety domain baseline

dc8d6c7
verified

btang18 commited on May 25

cleanup: remove intermediate training checkpoint

c826aed
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

3b7b8d3
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

9a11fc6
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

cccf99d
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

054971d
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

90ae7b6
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

fbd0d0c
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

a66e93b
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

5c3e18d
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

f2e5a28
verified

btang18 commited on May 22

cleanup: remove intermediate training checkpoint

fce01ad
verified

btang18 commited on May 22

aug302: few-shot conservative examples in system instruction

4458587
verified

btang18 commited on May 22

feat: frozen-attention SFT aug302 — beats GRPO on WildJailbreak

307e94a
verified

btang18 commited on May 21

grpo-v3 ckpt-1000

2879bfe
verified

btang18 commited on May 21

cleanup: remove training_args.bin

8cee454
verified

btang18 commited on May 20

cleanup: remove trainer_state.json

b94c154
verified

btang18 commited on May 20

cleanup: remove rng_state.pth

ef77c8f
verified

btang18 commited on May 20

cleanup: remove scheduler.pt

01ac3f0
verified

btang18 commited on May 20

cleanup: remove optimizer.pt

f9c5cdf
verified

btang18 commited on May 20

cleanup: remove model.safetensors.index.json

47ffd36
verified

btang18 commited on May 20

cleanup: remove model-00002-of-00002.safetensors

7e742d9
verified

btang18 commited on May 20

cleanup: remove model-00001-of-00002.safetensors

445e2b0
verified

btang18 commited on May 20

grpo mixed_v2 ckpt-2500 safetybench+3.2pp beavertails+4.8pp

9cfa5ff
verified

btang18 commited on May 20

fix: always enclose final answer in boxed

e7e2f43
verified

btang18 commited on May 18

cleanup: remove checkpoint-7368/training_args.bin

8196369
verified

btang18 commited on May 18

cleanup: remove checkpoint-7368/trainer_state.json

a65ed26
verified

btang18 commited on May 18

cleanup: remove checkpoint-7368/tokenizer_config.json

38217ed
verified

btang18 commited on May 18

cleanup: remove checkpoint-7368/tokenizer.json

4a66d4d
verified

btang18 commited on May 18

cleanup: remove checkpoint-7368/scheduler.pt

5f16a90
verified

btang18 commited on May 18

cleanup: remove checkpoint-7368/rng_state.pth

8cec41b
verified

btang18 commited on May 18

Commit History

revert to 90d4cbcfc1b5 b7cb487 verified

revert to bcbebb6c75c1 2f9f981 verified

Automated MNLP evaluation report (2026-06-06) (#1) d8a6f37

revert to e8a1b98e7024 c2e6e75 verified

checkpoint update 0603d: grpo_v8 + targeted SB DPO 96f0eb6 verified

fix use_cache=true and bos_token_id in config.json c552a86 verified

checkpoint update 0603a 0d115be verified

checkpoint update 0530b bcbebb6 verified

Revert to original system instruction (undo balanced prompt — online eval dropped 84->83) e8a1b98 verified

v6 + balanced system prompt: remove conservative bias, add conciseness, fix JBB/WJ over-refusal 6de93cb verified

cleanup: remove hydra/sft.log 2ca89bd verified

cleanup: remove hydra/.hydra/overrides.yaml 0db58c7 verified

cleanup: remove hydra/.hydra/hydra.yaml a0d042b verified

cleanup: remove hydra/.hydra/config.yaml 55aab3f verified

cleanup: remove training_args.bin f9c3519 verified

cleanup: remove model.safetensors.index.json 92caef6 verified

cleanup: remove model-00002-of-00002.safetensors 58318ed verified

cleanup: remove model-00001-of-00002.safetensors 99a7127 verified

feat(safety): v6 SFT — attn-only LoRA, category-stratified data, SB 92.0% BT 61.5% JBB 53.4% 90d4cbc verified

Upload Qwen3-1.7B base model as safety domain baseline dc8d6c7 verified

cleanup: remove intermediate training checkpoint c826aed verified

cleanup: remove intermediate training checkpoint 3b7b8d3 verified

cleanup: remove intermediate training checkpoint 9a11fc6 verified

cleanup: remove intermediate training checkpoint cccf99d verified

cleanup: remove intermediate training checkpoint 054971d verified

cleanup: remove intermediate training checkpoint 90ae7b6 verified

cleanup: remove intermediate training checkpoint fbd0d0c verified

cleanup: remove intermediate training checkpoint a66e93b verified

cleanup: remove intermediate training checkpoint 5c3e18d verified

cleanup: remove intermediate training checkpoint f2e5a28 verified

cleanup: remove intermediate training checkpoint fce01ad verified

aug302: few-shot conservative examples in system instruction 4458587 verified

feat: frozen-attention SFT aug302 — beats GRPO on WildJailbreak 307e94a verified

grpo-v3 ckpt-1000 2879bfe verified

cleanup: remove training_args.bin 8cee454 verified

cleanup: remove trainer_state.json b94c154 verified

cleanup: remove rng_state.pth ef77c8f verified

cleanup: remove scheduler.pt 01ac3f0 verified

cleanup: remove optimizer.pt f9c5cdf verified

cleanup: remove model.safetensors.index.json 47ffd36 verified

cleanup: remove model-00002-of-00002.safetensors 7e742d9 verified

cleanup: remove model-00001-of-00002.safetensors 445e2b0 verified

grpo mixed_v2 ckpt-2500 safetybench+3.2pp beavertails+4.8pp 9cfa5ff verified

fix: always enclose final answer in boxed e7e2f43 verified

cleanup: remove checkpoint-7368/training_args.bin 8196369 verified

cleanup: remove checkpoint-7368/trainer_state.json a65ed26 verified

cleanup: remove checkpoint-7368/tokenizer_config.json 38217ed verified

cleanup: remove checkpoint-7368/tokenizer.json 4a66d4d verified

cleanup: remove checkpoint-7368/scheduler.pt 5f16a90 verified

cleanup: remove checkpoint-7368/rng_state.pth 8cec41b verified

revert to 90d4cbcfc1b5

b7cb487
verified

revert to bcbebb6c75c1

2f9f981
verified

Automated MNLP evaluation report (2026-06-06) (#1)

d8a6f37

revert to e8a1b98e7024

c2e6e75
verified

checkpoint update 0603d: grpo_v8 + targeted SB DPO

96f0eb6
verified

fix use_cache=true and bos_token_id in config.json

c552a86
verified

checkpoint update 0603a

0d115be
verified

checkpoint update 0530b

bcbebb6
verified

Revert to original system instruction (undo balanced prompt — online eval dropped 84->83)

e8a1b98
verified

v6 + balanced system prompt: remove conservative bias, add conciseness, fix JBB/WJ over-refusal

6de93cb
verified

cleanup: remove hydra/sft.log

2ca89bd
verified

cleanup: remove hydra/.hydra/overrides.yaml

0db58c7
verified

cleanup: remove hydra/.hydra/hydra.yaml

a0d042b
verified

cleanup: remove hydra/.hydra/config.yaml

55aab3f
verified

cleanup: remove training_args.bin

f9c3519
verified

cleanup: remove model.safetensors.index.json

92caef6
verified

cleanup: remove model-00002-of-00002.safetensors

58318ed
verified

cleanup: remove model-00001-of-00002.safetensors

99a7127
verified

feat(safety): v6 SFT — attn-only LoRA, category-stratified data, SB 92.0% BT 61.5% JBB 53.4%

90d4cbc
verified

Upload Qwen3-1.7B base model as safety domain baseline

dc8d6c7
verified

cleanup: remove intermediate training checkpoint

c826aed
verified

cleanup: remove intermediate training checkpoint

3b7b8d3
verified

cleanup: remove intermediate training checkpoint

9a11fc6
verified

cleanup: remove intermediate training checkpoint

cccf99d
verified

cleanup: remove intermediate training checkpoint

054971d
verified

cleanup: remove intermediate training checkpoint

90ae7b6
verified

cleanup: remove intermediate training checkpoint

fbd0d0c
verified

cleanup: remove intermediate training checkpoint

a66e93b
verified

cleanup: remove intermediate training checkpoint

5c3e18d
verified

cleanup: remove intermediate training checkpoint

f2e5a28
verified

cleanup: remove intermediate training checkpoint

fce01ad
verified

aug302: few-shot conservative examples in system instruction

4458587
verified

feat: frozen-attention SFT aug302 — beats GRPO on WildJailbreak

307e94a
verified

grpo-v3 ckpt-1000

2879bfe
verified

cleanup: remove training_args.bin

8cee454
verified

cleanup: remove trainer_state.json

b94c154
verified

cleanup: remove rng_state.pth

ef77c8f
verified

cleanup: remove scheduler.pt

01ac3f0
verified

cleanup: remove optimizer.pt

f9c5cdf
verified

cleanup: remove model.safetensors.index.json

47ffd36
verified

cleanup: remove model-00002-of-00002.safetensors

7e742d9
verified

cleanup: remove model-00001-of-00002.safetensors

445e2b0
verified

grpo mixed_v2 ckpt-2500 safetybench+3.2pp beavertails+4.8pp

9cfa5ff
verified

fix: always enclose final answer in boxed

e7e2f43
verified

cleanup: remove checkpoint-7368/training_args.bin

8196369
verified

cleanup: remove checkpoint-7368/trainer_state.json

a65ed26
verified

cleanup: remove checkpoint-7368/tokenizer_config.json

38217ed
verified

cleanup: remove checkpoint-7368/tokenizer.json

4a66d4d
verified

cleanup: remove checkpoint-7368/scheduler.pt

5f16a90
verified

cleanup: remove checkpoint-7368/rng_state.pth

8cec41b
verified