TinyLLaVA
Here, we introduce TinyLLaVA-Qwen2.5-3B-SigLIP , which is trained by the TinyLLaVA Factory codebase. For LLM and vision tower, we choose Qwen2.5-3B and siglip-so400m-patch14-384, respectively.
Usage
Execute the following test code:
from tinyllava.eval.run_tiny_llava import eval_model
model_path = 'Zhang199/TinyLLaVA-Qwen2.5-3B-SigLIP'
prompt = "What are the things I should be cautious about when I visit here?"
image_file = "https://llava-vl.github.io/static/images/view.jpg"
conv_mode = "phi" # or llama, gemma, etc
args = type('Args', (), {
"model_path": model_path,
"model": None,
"query": prompt,
"conv_mode": conv_mode,
"image_file": image_file,
"sep": ",",
"temperature": 0,
"top_p": None,
"num_beams": 1,
"max_new_tokens": 512
})()
eval_model(args)
Result
| model_name | vqav2 | gqa | sqa | textvqa | MM-VET | POPE | MME | MMMU |
|---|---|---|---|---|---|---|---|---|
| LLaVA-1.5-7B | 78.5 | 62.0 | 66.8 | 58.2 | 30.5 | 85.9 | 1510.7 | - |
| bczhou/TinyLLaVA-3.1B (our legacy model) | 79.9 | 62.0 | 69.1 | 59.1 | 32.0 | 86.4 | 1464.9 | - |
| tinyllava/TinyLLaVA-Gemma-SigLIP-2.4B | 78.4 | 61.6 | 64.4 | 53.6 | 26.9 | 86.4 | 1339.0 | 31.7 |
| tinyllava/TinyLLaVA-Phi-2-SigLIP-3.1B | 80.1 | 62.1 | 73.0 | 60.3 | 37.5 | 87.2 | 1466.4 | 38.4 |
| Zhang199/TinyLLaVA-Qwen2-0.5B-SigLIP | 72.33 | 55.84 | 60.14 | 45.17 | 19.5 | 86.59 | 1153 | 29.7 |
| Zhang199/TinyLLaVA-Qwen2.5-3B-SigLIP | 79.4 | 62.5 | 74.1 | 58.3 | 34.8 | 87.4 | 1438.7 | 39.9 |
P.S. TinyLLaVA Factory is an open-source modular codebase for small-scale LMMs with a focus on simplicity of code implementations, extensibility of new features, and reproducibility of training results. This code repository provides standard training&evaluating pipelines, flexible data preprocessing&model configurations, and easily extensible architectures. Users can customize their own LMMs with minimal coding effort and less coding mistake.
TinyLLaVA Factory integrates a suite of cutting-edge models and methods.
- LLM currently supports OpenELM, TinyLlama, StableLM, Qwen, Gemma, Phi, and Qwen2.
- Vision tower currently supports CLIP, SigLIP, Dino, and combination of CLIP and Dino.
- Connector currently supports MLP, Qformer, and Resampler.