Instructions to use JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff with Transformers:

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

pipe = pipeline("text-generation", model="JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff")
model = AutoModelForCausalLM.from_pretrained("JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff", 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 JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff

SGLang

How to use JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff 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 "JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff" \
    --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": "JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff",
		"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 "JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff" \
        --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": "JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff with Docker Model Runner:
```
docker model run hf.co/JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff
```

tofu_Llama-3.2-3B-Instruct_forget10_GradDiff

open-unlearning/tofu_Llama-3.2-3B-Instruct_full unlearned on the TOFU forget10 split with GradDiff, trained with the open-unlearning framework. Used as a weight-unlearning baseline / draft model in the Speculative-Decoding-Unlearning project.

Full training config: .hydra/config.yaml. TOFU evaluation outputs: evals/.

Method hyperparameters

gamma: 1.0
alpha: 5
retain_loss_type: NLL

TOFU summary metrics

metric	value
exact_memorization	0.0221
extraction_strength	0.0325
forget_Q_A_PARA_Prob	0.0000
forget_Q_A_gibberish	0.2874
forget_quality	0.0000
forget_truth_ratio	0.0005
mia_loss	0.0041
mia_min_k	0.0049
mia_min_k_plus_plus	0.0139
mia_zlib	0.0138
model_utility	0.6054
privleak	64.2205

Downloads last month: -

Safetensors

Model size

3B params

Tensor type

BF16

Model tree for JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff

Base model

open-unlearning/tofu_Llama-3.2-3B-Instruct_full

Finetuned

(31)

this model

JoaoBoer
/

tofu_Llama-3.2-3B-Instruct_forget10_GradDiff

tofu_Llama-3.2-3B-Instruct_forget10_GradDiff

Method hyperparameters

TOFU summary metrics

Model tree for JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff

Dataset used to train JoaoBoer/tofu_Llama-3.2-3B-Instruct_forget10_GradDiff