epispasm/qwen3.5-9b-nsfw-captioning-v5_from_oldhag88_epispasmt_trained
The epispasm/qwen3.5-9b-nsfw-captioning-v5_from_oldhag88_epispasmt_trained model is a 9 billion parameter Qwen3.5-based language model developed by epispasm. It was fine-tuned from oldhag88/qwen3.5-9b-nsfw-captioning-v5 using Unsloth and Huggingface's TRL library, resulting in 2x faster training. This model is specifically designed for NSFW captioning tasks and has a context length of 32768 tokens.
Loading preview...
Model Overview
This model, developed by epispasm, is a 9 billion parameter Qwen3.5-based language model fine-tuned for NSFW captioning. It leverages the oldhag88/qwen3.5-9b-nsfw-captioning-v5 as its base and was trained with Unsloth and Huggingface's TRL library, achieving a 2x speedup in the training process. The model supports a context length of 32768 tokens.
Key Characteristics
- Architecture: Qwen3.5-based.
- Parameter Count: 9 billion parameters.
- Training Efficiency: Utilized Unsloth for 2x faster fine-tuning.
- Primary Use Case: Specialized in NSFW captioning.
- Context Length: Supports up to 32768 tokens.
Known Issues & Patching
It's noted that this model may require a patch to function correctly with certain inference engines like LM Studio, due to a "missing tensor 'blk.N.attn_norm.weight'" error in its GGUF format. A provided Python script (patch_gguf_qwen36.py) is available to fix this by updating GGUF metadata, specifically adjusting qwen35.block_count to 32 and qwen35.nextn_predict_layers to 0 for Qwen3.5-9B models. Users are advised to apply this patch for optimal compatibility.