Add more representative models for diverse as part of GA++ models

[mergennachin]

🚀 The feature, motivation and pitch

• Object Detection: yolov[8-10], detr-resnet-50, yolos, EfficientDet, MobileViT, OWL-ViT
• Depth Estimation: Depth Anything 2
• TTS: kokoro, parler-tts, microsoft/speecht5_tts
• Audio: Whisper, Wav2Vec2, AST, CLAP
• OCR: TrOCR, PaddlePaddle/PP-OCRv5_mobile_rec
• Image super resolution: Swin2SR, Real-ESRGAN
• Image Understanding: OpenCLIP, SegFormer
• Image reasoning: SmolVLM-256M-Instruct
• Semantic Text Search: sentence-transformers/all-MiniLM-L6-v2, BAAI/bge-base-en-v1.5
• Text Sentiment Analysis: RoBERTa

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

[mergennachin]

cc @digantdesai @kimishpatel @larryliu0820 @JacobSzwejbka @guangy10

[mergennachin]

#12340

#12398

[guangy10]

@mergennachin it looks like #12340 and #12398 cover at least one model in each category, except for SmolVLM-256M-Instruct, which is the model I've been looking already. If there are any model that I could help with please let me know.

[jackzhxng]

@guangy10 We need a TTS model as well. I'm looking into Kokoro since that's by far and away the most popular one but it would be great if you could take a look at microsoft/speecht5_tts, it's already modeled in transformers.

[jackzhxng]

Also just to make sure none of us are doing duplicate work here on the image reasoning front, since @larryliu0820 has recently been looking into Gemma 3's image encoder.