Instructions to use Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1 with Transformers:

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

pipe = pipeline("text-generation", model="Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1")
model = AutoModelForCausalLM.from_pretrained("Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1", 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 Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1

SGLang

How to use Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1 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 "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1" \
    --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": "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1",
		"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 "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1" \
        --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": "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1 with Docker Model Runner:
```
docker model run hf.co/Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1
```

Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1

See LICENSE and NOTICE.

An unlearned model from the RAQUEL MUSE experiments: Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-M-orig-LoRA-v1 (M_orig) after LoRA unlearning of the MUSE forget set with RMU, released at its early-stopped checkpoint. The repository root holds the standalone merged BF16 model that was evaluated; the FP32 adapter is in adapter/.

Method: RMU: steers layer-7 activations of forget tokens toward a fixed random direction while matching frozen M_orig activations on retain data.
Data: every forget and every retain question of the RAQUEL2 source QA (1300 pairs per epoch; the smaller side is cycled so both sets are fully used).
Schedule: 5 epochs, 205 optimizer steps; a checkpoint was scored every 4 steps.
Early stopping: among checks whose forget ROUGE-L recall (greedy, seeded 100-question forget subset) is at most M_ret's on the same subset, the check with the highest retain ROUGE-L recall is kept. RAQUEL questions were never used for selection. Released checkpoint: step 192 of 205 (forget ROUGE-L 0.172 <= target 0.232; retain ROUGE-L 0.893).

Evaluation

Split	This model	M_orig	M_ret
Forget (original)	71/750 (9.5%)	99.9%	17.3%
Forget (paraphrased)	83/741 (11.2%)	81.0%	16.6%
Retain (original)	1164/1300 (89.5%)	99.9%	100.0%
Retain (paraphrased)	757/1283 (59.0%)	75.8%	76.2%
RAQUEL affected	99/1709 (5.8%)	19.8%	12.5%
RAQUEL unaffected	333/2010 (16.6%)	20.7%	13.0%

Semantic accuracy judged by Qwen/Qwen3.8-27B (vLLM 0.28.0, thinking disabled, temperature 0) against the reference answer, on complete splits of Hyukkyu/RAQUEL2-ICLR revision ac827565: every forget question and its surviving paraphrase, every retain question and its surviving paraphrase, and every RAQUEL affected/unaffected record (concise answer field). Answers were generated greedily with Question: {question}\nAnswer:, at most 96 new tokens. Per-split counts and evidence hashes are in evaluation.json. M_orig and M_ret rows are the reference baselines on the same splits.

Training

Start: Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-M-orig-LoRA-v1, revision 3b7657ae6848d67c006cba290c253759ec9b54ab.
LoRA rank 64, alpha 128, dropout 0.05 on q/k/v/o/gate/up/down projections; BF16 base, FP32 adapters; one GPU.
Learning rate 0.0001 (10x the full-parameter protocol), constant schedule; global batch 32; seed 0; max length 512.
Method settings: retain_weight=10.0.
Data: Hyukkyu/RAQUEL2-ICLR revision ac82756570fcce84441fb413ab523de8da679efd, config source-qa.

Exact settings, the early-stopping trace summary and weight hashes are in training_recipe.json.

Load

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

repo_id = "Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(repo_id, dtype=torch.bfloat16, device_map="auto").eval()
prompt = "Question: {question}\nAnswer:"

The root merged weights are the evaluated artifact. The FP32 LoRA adapter is in adapter/ (PeftModel.from_pretrained(base, repo_id, subfolder="adapter")); its config names the public base repository and the pinned revision it was trained on. Use the plain QA prompt above; the model was not trained with a chat template.

Downloads last month: 278

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1

Base model

Qwen/Qwen3-8B-Base

Finetuned

Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-M-orig-LoRA-v1

Finetuned

(5)

this model

Hyukkyu
/

Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1

Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1

Evaluation

Training

Load

Model tree for Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1

Datasets used to train Hyukkyu/Qwen3-8B-Base-RAQUEL-MUSE-Unlearn-RMU-LoRA-v1