---
license: apache-2.0
base_model:
- prithivMLmods/Qwen3-VL-32B-Instruct-abliterated-v1
pipeline_tag: image-text-to-text
tags:
- text-generation-inference
- uncensored
- abliterated
- unfiltered
- unredacted
- vllm
- pytorch
- fp8
- f8_e4m3
- max
- agent
language:
- en
library_name: transformers
---

![1](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/arngxpHFh7IIjFKtwtfjM.png)

# **Qwen3-VL-32B-Instruct-Unredacted-MAX-FP8**

> **Qwen3-VL-32B-Instruct-Unredacted-MAX-FP8** is an FP8-compressed evolution built on top of **prithivMLmods/Qwen3-VL-32B-Instruct-abliterated-v1**. This variant leverages **BF16 · FP8 (F8_E4M3)** precision formats to significantly reduce memory footprint and improve inference efficiency, while preserving the unredacted multimodal reasoning strengths of the original 32B architecture.
> The result is a highly capable 32B vision-language model optimized for unrestricted, detailed reasoning and captioning across complex visual inputs, with enhanced hardware efficiency.

> [!important]
> FP8 (8-bit floating point) weight and activation quantization using hardware acceleration on GPUs – [FP8 W8A8](https://docs.vllm.ai/en/stable/features/quantization/fp8/). Quantization W8A8 FP8-dynamic recipe – [examples](https://github.com/vllm-project/llm-compressor/tree/main/examples/quantization_w8a8_fp8).


## Key Highlights

* **BF16 · FP8 (F8_E4M3) Compression**: Transformer Engine–based FP8 quantization reduces VRAM usage and improves throughput while maintaining strong multimodal reasoning fidelity.
* **Unredacted MAX Training**: Retains the abliterated fine-tuning strategy designed to minimize internal refusal behaviors and improve instruction adherence.
* **32B Parameter Architecture**: Built on top of `prithivMLmods/Qwen3-VL-32B-Instruct-abliterated-v1` (Derived from Qwen/Qwen3-VL-32B-Instruct)., delivering substantially stronger reasoning capacity while benefiting from FP8 efficiency gains.
* **Unrestricted Multimodal Reasoning**: Designed for deep analysis of artistic, forensic, technical, or abstract visual content without standard safety-driven refusals.
* **High-Fidelity Captions**: Produces dense, descriptive outputs suitable for dataset generation, metadata enrichment, or accessibility use cases.
* **Dynamic Resolution Support**: Retains Qwen3-VL’s ability to process varying image resolutions and aspect ratios effectively.
* **Optimized Deployment**: FP8 compression enables smoother deployment on Hopper and compatible GPU architectures.

## Quick Start with Transformers

```python
from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch

# Load the 32B Instruct Unredacted MAX FP8 model
model = Qwen3VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen3-VL-32B-Instruct-Unredacted-MAX-FP8",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained(
    "prithivMLmods/Qwen3-VL-32B-Instruct-Unredacted-MAX-FP8"
)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Provide a detailed caption and reasoning for this image."},
        ],
    }
]

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

image_inputs, video_inputs = process_vision_info(messages)

inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    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

* **Advanced Red-Teaming**: Evaluating multimodal robustness and probing behavioral edge cases.
* **Complex Data Archiving**: Generating detailed captions for medical, artistic, historical, or research datasets.
* **Refusal Mechanism Research**: Studying behavioral shifts in vision-language models after abliterated fine-tuning.
* **Creative Storytelling**: Producing detailed visual descriptions for narrative and world-building projects.

## Limitations & Risks

> **Critical Note**: This model is designed to minimize built-in refusal mechanisms.

* **Sensitive Content Exposure**: The model may generate explicit or controversial descriptions if prompted accordingly.
* **User Responsibility**: Generated outputs must be handled responsibly and used within ethical and legal boundaries.
* **Hardware Requirements**: While lighter than full-precision 32B variants, the FP8 architecture still requires compatible GPU support and sufficient VRAM for high-resolution image processing and extended generations.

## Acknowledgements
 
I would like to thank the works of the following:

* Uncensor any LLM with abliteration – [Maxime Labonne](https://huggingface.co/mlabonne)
* Using FP8 and FP4 with Transformer Engine – [docs.nvidia](https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/examples/fp8_primer.html)
* Remove Refusals with Transformers – [Sumandora](https://github.com/Sumandora/remove-refusals-with-transformers)
* LLM Compressor – [vllm-project](https://github.com/vllm-project/llm-compressor)
* FP8 Floating-Point 8: An Introduction to Efficient, Lower-Precision AI Training – [nvidia](https://developer.nvidia.com/blog/floating-point-8-an-introduction-to-efficient-lower-precision-ai-training/)