slahiriandClaude Opus 4.5
238ef20a59
Add 14 captioning models organized into Fast and Balanced tiers: Fast tier (~1-8GB VRAM): - BLIP-2 (Salesforce/blip2-opt-2.7b) - Florence-2-base-PromptGen-v1.5 - Florence-2-large-PromptGen-v2.0 - Qwen2-VL-2B, Qwen2.5-VL-3B, Qwen3-VL-2B, Qwen3-VL-4B Balanced tier (~8-17GB VRAM): - JoyCaption Alpha-Two and Beta-One - MiniCPM-V-2.6 - Qwen2-VL-7B, Qwen2.5-VL-7B variants, Qwen3-VL-8B Features: - Config-driven model architecture with JSON configs - Model classes for each architecture (Florence, Qwen2/2.5/3, JoyCaption, BLIP-2, MiniCPM-V) - Factory pattern with caching and automatic cleanup on model switch - Precision modes: Auto, FP16, FP8, 4-Bit quantization - Tier-prefixed dropdown for easy model selection 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
25 lines
1000 B
JSON
25 lines
1000 B
JSON
{
|
|
"name": "JoyCaption-Beta-One",
|
|
"description": "JoyCaption Beta One - LLaVA-based image captioning model, trained on 2.4M samples",
|
|
"model_id": "fancyfeast/llama-joycaption-beta-one-hf-llava",
|
|
"model_type": "joycaption",
|
|
"dtype": "bf16",
|
|
"trust_remote_code": true,
|
|
"prompts": {
|
|
"detailed": "Write a detailed description of this image, focusing on the subject, composition, lighting, colors, mood, and artistic style.",
|
|
"short": "Write a brief one-sentence caption for this image.",
|
|
"tags": "List descriptive tags for this image, separated by commas. Include subject, style, mood, colors, and composition.",
|
|
"analyze": "Analyze this image's composition, lighting, color palette, and artistic style. What makes this image visually interesting?"
|
|
},
|
|
"system_prompt": "You are a helpful image captioner.",
|
|
"generation": {
|
|
"max_tokens": 512,
|
|
"do_sample": true,
|
|
"temperature": 0.6
|
|
},
|
|
"download": {
|
|
"use_symlinks": false,
|
|
"subfolder": "LLM"
|
|
}
|
|
}
|