GitMylo/nsfwvision-qwen3-vl-8b-v3-safetensors
GitMylo/nsfwvision-qwen3-vl-8b-v3-safetensors is an 8 billion parameter vision-language model, fine-tuned from Qwen3-VL-8B-Instruct-abliterated-v2 with a context length of 32768 tokens. This model's vision tower was specifically trained to learn concepts related to NSFW content, while its language capabilities were preserved by resetting the base model weights after training. Its primary use case is for vision-based tasks involving the detection and understanding of NSFW imagery.
Loading preview...
NSFWVision Qwen3 VL 8B V3
This model, developed by GitMylo, is an 8 billion parameter vision-language model (VLM) based on the Qwen3-VL-8B-Instruct-abliterated-v2 architecture. It features a substantial context length of 32768 tokens, making it suitable for processing detailed visual and textual inputs.
Key Capabilities
- Specialized Vision Training: The model's vision tower underwent specific fine-tuning to recognize and understand concepts related to NSFW (Not Safe For Work) content.
- Preserved Language Abilities: A unique training methodology was employed where the base model's weights were reset after the vision tower's training. This "soaking" process allowed the vision component to learn effectively without biasing the model's original writing style or prompt-following capabilities.
- High Context Vision-Language: Leveraging the Qwen3-VL architecture, it combines robust language understanding with advanced visual processing.
Good for
- NSFW Content Detection: Ideal for applications requiring the identification and analysis of NSFW imagery.
- Content Moderation: Can be integrated into systems for automated content filtering and moderation.
- Research in Vision-Language Models: Provides a specialized VLM for exploring fine-tuning techniques that isolate and enhance specific visual understanding without altering core language generation.