Add Microsoft KOSMOS-2.5 model support to VLM evaluation application

[Copilot]

This PR adds support for Microsoft's KOSMOS-2.5 vision-language model to the VLM evaluation application, enabling users to test and compare this model alongside the existing options.

Changes Made

Model Integration

• Added KOSMOS-2.5 to model choices: Updated MODEL_CHOICES dictionary to include "microsoft/kosmos-2.5": "main"
• Added required imports: Imported Kosmos2ForConditionalGeneration and Kosmos2Processor from the transformers library
• Implemented model loading logic: Added specific loading configuration for KOSMOS-2.5 in the load_model_and_processor() function with proper device mapping and authentication
• Implemented text extraction logic: Added input processing for KOSMOS-2.5 in the extract_text_from_image() function with appropriate prompt formatting using the <grounding> tag format

Documentation Updates

• Updated README.md: Added KOSMOS-2.5 to the list of supported models in the project documentation

Technical Implementation Details

The KOSMOS-2.5 integration follows the established patterns in the codebase:

# Model loading with proper device mapping and authentication
model = Kosmos2ForConditionalGeneration.from_pretrained(
    model_name,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto" if torch.cuda.is_available() else None,
    use_auth_token=api_token
)
processor = Kosmos2Processor.from_pretrained(model_name, use_auth_token=api_token)

# Proper prompt formatting for KOSMOS-2 models
formatted_prompt = f"<grounding>An image of {prompt}"
inputs = processor(text=formatted_prompt, images=image, return_tensors='pt').to(device)

UI Changes

The KOSMOS-2.5 model is now available in the model selection dropdown alongside all existing models:

Testing

• ✅ Application imports successfully with new KOSMOS-2.5 support
• ✅ Model appears correctly in the UI dropdown
• ✅ Flask application runs without errors
• ✅ Integration follows existing code patterns and conventions

The implementation is minimal and surgical, adding only the necessary code to support KOSMOS-2.5 without modifying existing functionality for other models.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• huggingface.co
  • Triggering command: python -c import requests; print(requests.get('REDACTED').json()) (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.