license: apache-2.0
tags:

• text-generation-inference
• uncensored
• abliterated
• unfiltered
• unredacted
• refusal-ablated
• vllm
• pytorch
• bf16
• max
• alignment-modified
• reasoning
• v2-MAX
• llama.cpp
  language:
• en
  base_model:
• Qwen/Qwen3.5-9B
  pipeline_tag: image-text-to-text
  library_name: transformers

Qwen3.5-9B-abliterated-v2-MAX

Qwen3.5-9B-abliterated-v2-MAX is an optimized release built on top of huihui-ai/Huihui-Qwen3.5-9B-abliterated. This version focuses on improved model sharding, packaging consistency, and compatibility with modern Transformers and inference stacks, while preserving the reasoning and instruction-following capabilities of the base model. The result is a highly capable 9B parameter language model designed for efficient deployment, stable inference, and research-oriented experimentation.

[!IMPORTANT]
This model is intended strictly 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 model-generated content. Users must ensure safe, ethical, and lawful usage.

Compression for the Model

Qwen3.5-9B-abliterated-v2-MAX

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-Qwen3.5-9B-abliterated

Key Highlights

• Optimized Packaging & Sharding
  Improved repository structure for smoother downloads, loading, and deployment across environments.

• Stable Transformers Compatibility
  Updated layout for better compatibility with modern Transformers versions and inference pipelines.

• 9B Parameter Architecture
  Built on Qwen3.5-9B, balancing efficiency and capability for local and research use.

• Efficient Deployment Design
  Designed for lightweight inference, experimentation, and scalable integration.

• Preserved Model Behavior
  No changes to weights or core architecture; performance remains consistent with the original base model lineage.

• Improved Reliability in Loading
  Reduced friction in model initialization and multi-device inference setups.

Quick Start with Transformers

pip install transformers==5.4.0
# or
pip install git+https://github.com/huggingface/transformers.git

from transformers import Qwen3_5ForConditionalGeneration, AutoProcessor
import torch

model = Qwen3_5ForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen3.5-9B-abliterated-v2-MAX",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained(
    "prithivMLmods/Qwen3.5-9B-abliterated-v2-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)

generated_ids_trimmed = [
    out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]

output_text = processor.batch_decode(
    generated_ids_trimmed,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False
)

print(output_text)

Intended Use

• Multimodal and Language Research
  Studying behavior of compact 9B-scale transformer models under different inference settings.

• Red-Teaming & Evaluation
  Testing robustness across adversarial prompts and edge-case inputs.

• Efficient Local Deployment
  Running lightweight yet capable models on consumer GPUs or optimized cloud setups.

• Research Prototyping
  Exploring model behavior, alignment, and inference optimization techniques.

Limitations & Risks

Important Note: This model inherits behavior from its base model with minimal modification.

• Output Variability
  Responses may vary depending on sampling strategy and prompt formulation.

• Resource Dependency
  While efficient, GPU acceleration is recommended for optimal performance.

• No Architectural Changes
  Improvements are limited to packaging and compatibility, not core model capabilities.

• General Model Limitations
  May still produce incorrect, incomplete, or inconsistent outputs in complex scenarios.