Dataset: 12,000 samples
Composition: Jailbreak 10k + General 1.5k + Logic 0.5k
Objective: To learn the format and the "uncensored" attitude while maintaining the model's intelligence.

Step 2: GRPO (Reinforcement Learning)

Dataset: 60,000 samples
Reward Model: puwaer/Unsafe-Reward-Qwen3-1.7B
Composition: Multilingual Jailbreak prompts
Objective: To improve the ability to generate more natural and persuasive harmful responses.

Model Performance

Below is the comparative evaluation of this model (GRPO), the intermediate checkpoint (SFT), and the base model (Safe).

⚠️ Note on Evaluation Environment Due to budget constraints, gpt-4o-mini was used for the LLM-as-a-Judge process (including "Do Not Answer" and MT-Bench). Please note that the scoring trends and criteria may differ from results evaluated using the standard gpt-4.

Safety Evaluation (Lower is better / Higher success in uncensoring)

In the "Do Not Answer" (DNA) and "Sorry Bench" benchmarks, while the base model shows a high refusal rate (~88%), this model achieves an extremely low refusal rate of under 4%–15%.

Benchmark	Metric	Base (Safe)	SFT (This Model)	GRPO (Step2)
do not answer	Safety Acc (Low is Better)	0.9979	0.8275	0.147
do not answer jp	Safety Acc (Low is Better)	0.984	0.5378	0.0873
Sorry Bench	Safety Acc (Low is Better)	0.8886	0.8455	0.0409

Capability Evaluation (Higher is better)

Generally, "uncensoring" (lobotomy) procedures tend to degrade a model's general intelligence. However, this model recovered its conversational scores (e.g., MT-Bench) by proceeding from the SFT stage to GRPO.

Benchmark	Metric	Base (Safe)	SFT (This Model)	GRPO (Step2)
MT-Bench	Average Score (1-10)	8.044	7.538	7.513
LM Harness	Average Acc (GSM8K, MMLU)	0.8454	0.8483	0.8436

*Comparisons made between Qwen3-Next-80B-A3B-Thinking (Base) *

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "puwaer/Qwen3-Next-80B-A3B-Thinking-SFT-Uncensored"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

# parsing thinking content
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")

print("thinking content:", thinking_content) # no opening <think> tag
print("content:", content)

2etatg/
Qwen3-Next-80B-A3B-Thinking-SFT-Uncensored

Qwen3-Next-80B-A3B-Thinking-SFT-Uncensored

Training Process

Step 1: SFT (Supervised Fine-Tuning)

Step 2: GRPO (Reinforcement Learning)

Model Performance

Safety Evaluation (Lower is better / Higher success in uncensoring)

Capability Evaluation (Higher is better)

Usage

Data Overview

Datasets

Reward Model

2etatg/Qwen3-Next-80B-A3B-Thinking-SFT-Uncensored

Qwen3-Next-80B-A3B-Thinking-SFT-Uncensored

Training Process

Step 1: SFT (Supervised Fine-Tuning)

Step 2: GRPO (Reinforcement Learning)

Model Performance

Safety Evaluation (Lower is better / Higher success in uncensoring)

Capability Evaluation (Higher is better)

Usage

Data Overview

Datasets

Reward Model

2etatg/
Qwen3-Next-80B-A3B-Thinking-SFT-Uncensored