Add Youtu-VL

[MollySophia]

HF: https://huggingface.co/tencent/Youtu-VL-4B-Instruct

# molly @ MOLLYJI-MB1 in ~/workspace/mlx-vlm on git:main x .venv [11:03:33] 
$ mlx_vlm.convert \        
    --hf-path /Users/molly/workspace-youtulm-ncnn/Youtu-VL-4B-Instruct \
    --mlx-path /Users/molly/workspace/Youtu-VL-4B-Instruct-4bit-test \
    -q --q-bits 4 --trust-remote-code
[INFO] Loading
[INFO] Using dtype: bfloat16
[INFO] Quantizing
[INFO] Quantized model with 5.460 bits per weight.

# molly @ MOLLYJI-MB1 in ~/workspace/mlx-vlm on git:main x .venv [11:11:41] 
$ mlx_vlm.generate --model ../Youtu-VL-4B-Instruct-4bit-test --max-tokens 512 --prompt "Please describe this image shortly" --trust-remote-code --image /Users/molly/workspace/Youtu-VL-4B-Instruct-6bit-mlx/assets/youtu-vl-logo.png
==========
Files: ['/Users/molly/workspace/Youtu-VL-4B-Instruct-6bit-mlx/assets/youtu-vl-logo.png'] 

Prompt: <|begin_of_text|>system
You are a helpful assistant.<|end_of_text|>
<|begin_of_text|>user
<|vision_start|><|image_pad|><|vision_end|>Please describe this image shortly<|end_of_text|>
<|begin_of_text|>assistant

This image is a promotional banner for a project called “Youtu-VL.”

It features:

- A cute cartoon penguin on the left, wearing headphones and a colorful scarf, suggesting a fun or accessible AI tool.
- The main title “Youtu-VL” in large, gray, sans-serif font centered on a black background.
- The tagline below: “Unleashing Visual Potential via Unified Vision-Language Supervision,” indicating the project’s focus on combining visual and language understanding.
- A colorful abstract logo on the right, composed of three interconnected shapes in blue, pink, and teal — likely representing the Youtu Lab’s branding.

The overall design is clean, modern, and tech-oriented, aimed at conveying innovation and accessibility in vision-language AI.
==========
Prompt: 348 tokens, 131.744 tokens-per-sec
Generation: 165 tokens, 42.101 tokens-per-sec
Peak memory: 4.887 GB