Qwen2.5-VL-Abliterated-Caption-it-GGUF (3B & 7B)
The Qwen2.5-VL-3B/7B-Abliterated-Caption-it models are fine-tuned variants of Qwen2.5-VL-Instruct architectures, specifically designed for Abliterated (Uncensored) Image Captioning, enabling high-fidelity, richly descriptive captions across diverse and nuanced visual inputs, including sensitive or complex imagery. Building upon the strengths of multimodal Qwen2.5-VL backbones, these models offer robust descriptive power across aspect ratios, artistic and technical visuals, stylized or low-context scenes, and are adaptable to multilingual output via prompt control. Trained on curated datasets combining prithivMLmods/blip3o-caption-mini-arrow, Caption3o-Opt-v2, and private data, they excel at generating unconstrained captions for creative storytelling, dataset enrichment, content moderation research, and generative safety evaluations. While the 3B variant is efficient and lightweight for faster inference, the 7B variant offers enhanced detail, consistency, and reasoning for high-quality outputs. Users should note these models may generate explicit or sensitive captions and are not suited for production with strict content moderation requirements.
Qwen2.5-VL Abliterated Caption (GGUF)
| Model | Link |
|---|---|
| Qwen2.5-VL-3B-Abliterated-Caption-it-GGUF | Hugging Face Repo |
| Qwen2.5-VL-7B-Abliterated-Caption-it-GGUF | Hugging Face Repo |
Model Files
Qwen2.5-VL-3B-Abliterated-Caption-it
| File Name | Quant Type | File Size |
|---|---|---|
| Qwen2.5-VL-3B-Abliterated-Caption-it.IQ4_XS.gguf | IQ4_XS | 1.75 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q2_K.gguf | Q2_K | 1.27 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q3_K_L.gguf | Q3_K_L | 1.71 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q3_K_M.gguf | Q3_K_M | 1.59 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q3_K_S.gguf | Q3_K_S | 1.45 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q4_K_M.gguf | Q4_K_M | 1.93 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q4_K_S.gguf | Q4_K_S | 1.83 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q5_K_M.gguf | Q5_K_M | 2.22 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q5_K_S.gguf | Q5_K_S | 2.17 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q6_K.gguf | Q6_K | 2.54 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.Q8_0.gguf | Q8_0 | 3.29 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.f16.gguf | F16 | 6.18 GB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.mmproj-Q8_0.gguf | MMProj Q8_0 | 845 MB |
| Qwen2.5-VL-3B-Abliterated-Caption-it.mmproj-f16.gguf | MMProj F16 | 1.34 GB |
Qwen2.5-VL-7B-Abliterated-Caption-it
| File Name | Quant Type | File Size |
|---|---|---|
| Qwen2.5-VL-7B-Abliterated-Caption-it.IQ4_XS.gguf | IQ4_XS | 4.25 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q2_K.gguf | Q2_K | 3.02 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q3_K_L.gguf | Q3_K_L | 4.09 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q3_K_M.gguf | Q3_K_M | 3.81 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q3_K_S.gguf | Q3_K_S | 3.49 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q4_K_M.gguf | Q4_K_M | 4.68 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q4_K_S.gguf | Q4_K_S | 4.46 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q5_K_M.gguf | Q5_K_M | 5.44 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q5_K_S.gguf | Q5_K_S | 5.32 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q6_K.gguf | Q6_K | 6.25 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.Q8_0.gguf | Q8_0 | 8.1 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.f16.gguf | F16 | 15.2 GB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.mmproj-Q8_0.gguf | MMProj Q8_0 | 853 MB |
| Qwen2.5-VL-7B-Abliterated-Caption-it.mmproj-f16.gguf | MMProj F16 | 1.35 GB |
Quants Usage
(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)
Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):
