Qwen3-VL-8B-Heretic-Stable

Qwen3-VL-8B-Heretic-Stable is an optimized release built on top of huihui-ai/Huihui-Qwen3-VL-8B-Instruct-abliterated. This version focuses on updated shard sizing, repository optimization, and compatibility improvements for the latest Transformers releases, while preserving the multimodal capabilities of the original model. The result is a stable and efficient 8B vision-language model designed for image and text reasoning with streamlined deployment and inference workflows.

[!IMPORTANT] This model is intended for research and learning purposes only. Any content generated by this model is used at the user’s own risk. The authors and hosting page disclaim any liability for outputs produced by this model. Users are responsible for ensuring safe, ethical, and lawful usage.

Evaluation [Self Reported]

Note: This release does not introduce new benchmark results and primarily focuses on repackaging, sharding updates, and Transformers compatibility improvements over the base model.

Key Highlights

• Latest Transformers Compatibility Re-sharded and optimized for compatibility with recent Transformers releases.

• Optimized Model Sharding Updated shard structure for improved download reliability, storage handling, and inference efficiency.

• Stable Multimodal Pipeline Ensures consistent behavior across image-text inputs using the original Qwen3-VL architecture.

• 8B Vision-Language Architecture Built on Qwen3-VL-8B-Instruct, supporting strong image understanding and multimodal reasoning.

• Improved Deployment Stability Designed for smoother loading and more predictable inference across environments.

• Preserved Model Behavior No changes to core weights or architecture; behavior remains aligned with the base model lineage.

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-VL-8B-Instruct-abliterated

Quick Start with Transformers

pip install transformers==5.9.0
# or
pip install git+https://github.com/huggingface/transformers.git

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

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "prithivMLmods/Qwen3-VL-8B-Heretic-Stable",
    torch_dtype="auto",
    device_map="auto"
)

processor = AutoProcessor.from_pretrained(
    "prithivMLmods/Qwen3-VL-8B-Heretic-Stable"
)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {
                "type": "text",
                "text": "Describe this image in detail."
            },
        ],
    }
]

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

• Multimodal Research Studying vision-language reasoning and model behavior across image-text inputs.

• Dataset Enrichment Generating captions and structured descriptions for visual datasets.

• Evaluation & Benchmarking Testing multimodal performance under varied prompts and image complexity.

• Application Prototyping Building vision-language tools for analysis, captioning, and reasoning workflows.

Limitations & Risks

Important Note: This model inherits the behavior and limitations of its base model.

• Output Variability Responses may vary depending on image quality, prompt design, and sampling parameters.

• Multimodal Constraints Performance depends on resolution, context length, and input complexity.

• Deployment Requirements Requires GPU acceleration for efficient inference, especially for high-resolution inputs.

• General Model Limitations May still produce inaccurate or incomplete outputs in complex scenarios.