Instructions to use shirasko/gemma-2-2b-it-snmf-uranium with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use shirasko/gemma-2-2b-it-snmf-uranium with Transformers:

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

pipe = pipeline("text-generation", model="shirasko/gemma-2-2b-it-snmf-uranium")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("shirasko/gemma-2-2b-it-snmf-uranium")
model = AutoModelForCausalLM.from_pretrained("shirasko/gemma-2-2b-it-snmf-uranium", 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 shirasko/gemma-2-2b-it-snmf-uranium with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "shirasko/gemma-2-2b-it-snmf-uranium"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "shirasko/gemma-2-2b-it-snmf-uranium",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/shirasko/gemma-2-2b-it-snmf-uranium

SGLang

How to use shirasko/gemma-2-2b-it-snmf-uranium 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 "shirasko/gemma-2-2b-it-snmf-uranium" \
    --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": "shirasko/gemma-2-2b-it-snmf-uranium",
		"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 "shirasko/gemma-2-2b-it-snmf-uranium" \
        --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": "shirasko/gemma-2-2b-it-snmf-uranium",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use shirasko/gemma-2-2b-it-snmf-uranium with Docker Model Runner:
```
docker model run hf.co/shirasko/gemma-2-2b-it-snmf-uranium
```

Unlearned Checkpoint

Field	Value
Unlearning method	SNMF
Base model	`google/gemma-2-2b-it`
Target concept	Uranium
Checkpoint type	Full Model Weights
Rank / seed	100 / 42
Train eval protocol	mc

Unlearning Configuration

Selected hyperparameters (from unlearned_checkpoints.json):

Parameter	Value
`coverage_thresh`	0.95
`delta_embed`	0
`delta_in`	4
`delta_out`	10
`feature_source`	all
`k_features_embed`	0
`k_features_mlp_in`	36
`k_features_mlp_out`	17
`layer_hi_in`	25
`layer_hi_out`	8
`layer_lo_in`	0
`layer_lo_out`	0
`n_tokens_edited`	0
`ratio_thresh`	2
`w_mode`	both

Primary Unlearning Metrics (held-out test, MC protocol)

Headline scores used for checkpoint selection:

Metric	Train (after unlearning)	Test (after unlearning)
Efficacy	0.872	0.857
Specificity	0.52	0.486
Harmonic mean	0.651	0.62
Relearning QA (MC)	—	0.22

Full Evaluation (baseline → unlearned)

From evaluation/score_comparison.csv:

Metric	Baseline (train)	After unlearn (train)	Baseline (test)	After unlearn (test)
QA accuracy	0.64	0.3	0.6	0.3
QA fraction	1	0.128	1	0.143
SimDom accuracy	0.62	0.38	0.74	0.44
SimDom fraction	1	0.351	1	0.388
MMLU accuracy	0.52	0.52	0.551	0.446
MMLU fraction	1	1	1	0.651

Files in This Repository

File	Description
`unlearned_checkpoints.json`	Checkpoint metadata & hyperparameters
`evaluation/evaluation_summary.json`	Full evaluation payload (train/test/relearning)
`evaluation/score_comparison.csv`	Baseline vs. unlearned comparison table