Qwen3.5-2B-Unredacted-MAX

Qwen3.5-2B-Unredacted-MAX is an optimized release built on top of huihui-ai/Huihui-Qwen3.5-2B-abliterated. This version focuses on improved repository structure, loading stability, and compatibility with modern Transformers inference pipelines, while preserving the reasoning and instruction-following behavior of the base model. The result is a lightweight 2B parameter language model designed for efficient deployment, experimentation, and research workflows.

This model is intended for research and learning purposes only. Any outputs generated by this model are the sole responsibility of the user. The authors and hosting platform disclaim all liability for generated content. Users must ensure safe, ethical, and lawful usage.

---

Base Model Signatures:

This model has been re-sharded and optimized for the latest Transformers version from the base model: https://huggingface.co/huihui-ai/Huihui-Qwen3.5-2B-abliterated

---

Evaluation Report (Self-Reported)

Model: Qwen3.5-2B-Unredacted-MAX

• Abliteration Rate (Non-Refusal Rate): 91.500
• Refusal Rate: 8.500

The evaluation was conducted using 2000 prompts across multiple runs to measure response behavior consistency. Results are averaged and may vary depending on sampling strategy, prompt distribution, and evaluation methodology.

Evaluation Summary (YAML)

evaluation:
  model_name: Qwen3.5-2B-Unredacted-MAX
  total_test_prompts: 2000
  evaluation_runs: 10
  prompts_per_run: 200
  evaluation_type: response_behavior_analysis

results:
  refusal_rate: 8.500
  non_refusal_rate: 91.500
  abliteration_rate: 91.500

Note: These results are self-reported and should be interpreted as approximate behavioral indicators rather than strict benchmarks.

---

Key Highlights

• Optimized Model Packaging Improved repository structure for easier deployment and loading.

• Stable Transformers Compatibility Designed for modern Hugging Face Transformers versions and inference workflows.

• 2B Parameter Architecture Lightweight model suitable for resource-constrained environments.

• Efficient Instruction Following Maintains consistent behavior across structured prompts.

• Fast Local Inference Optimized for low-latency deployment on consumer hardware.

---

Quick Start with Transformers

pip install transformers==5.3.0
# or
pip install git+https://github.com/huggingface/transformers.git

from transformers import Qwen3_5ForConditionalGeneration, AutoProcessor
import torch

model = Qwen3_5ForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen3.5-2B-Unredacted-MAX",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained(
    "prithivMLmods/Qwen3.5-2B-Unredacted-MAX"
)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "text", "text": "Explain how transformer models work in simple terms."}
        ],
    }
]

text = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)

inputs = processor(
    text=[text],
    padding=True,
    return_tensors="pt"
).to("cuda")

generated_ids = model.generate(**inputs, max_new_tokens=256)

output_text = processor.batch_decode(
    [out[len(inp):] for inp, out in zip(inputs.input_ids, generated_ids)],
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False
)

print(output_text)

---

Intended Use

• Research into transformer behavior and lightweight model performance
• Edge-device and CPU-friendly AI deployment
• Red-teaming and robustness testing
• Rapid prototyping of NLP applications

---

Limitations & Risks

Important Note: This model inherits limitations from its base architecture.

• Output quality varies significantly with prompt design
• Limited long-context reasoning compared to larger models
• Requires careful tuning for optimal performance
• May produce incorrect or inconsistent outputs in complex tasks