Make vLLM and transformers optional dependencies

[Copilot]

This PR makes vLLM and transformers optional dependencies to enable lighter installations of rank_llm based on user needs.

Problem

Previously, installing rank_llm required all heavy dependencies including vLLM and transformers, even if users only needed core functionality like API-based ranking. This resulted in:

• Large installation size and long install times
• Dependency conflicts in constrained environments
• Unnecessary overhead for users not using local models

Solution

New Installation Options

Core installation (lightweight, API-only functionality):

pip install rank_llm

With vLLM support (for local model inference):

pip install rank_llm[vllm]

With transformers support (for T5-based models like DuoT5, MonoT5):

pip install rank_llm[transformers]

Full installation (all features):

pip install rank_llm[all]

Implementation Details

All modules that depend on optional libraries now use conditional imports with helpful error messages:

try:
    import vllm
    VLLM_AVAILABLE = True
except ImportError:
    VLLM_AVAILABLE = False
    vllm = None

class VllmHandler:
    def __init__(self, ...):
        if not VLLM_AVAILABLE:
            raise ImportError(
                "vLLM is not installed. Please install it with: pip install rank_llm[vllm]"
            )

Changes Made

• Dependency configuration: Moved vLLM and transformers from core to optional dependencies in pyproject.toml
• Code updates: Added conditional imports to 10+ modules including VllmHandler, DuoT5, MonoT5, RankFiD, and inference handlers
• Error handling: Clear error messages guide users to install missing dependencies
• Documentation: Updated README and added comprehensive usage guide
• Testing: Added unit tests to verify optional dependency behavior

Backward Compatibility

Existing installations continue to work. Users can migrate by installing the appropriate optional dependencies:

• For vLLM users: pip install rank_llm[vllm]
• For T5 model users: pip install rank_llm[transformers]

This change enables more flexible deployment while maintaining full functionality for users who need the complete feature set.

Fixes #293.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[Copilot]

@ronakice 👋 This repository doesn't have Copilot instructions. With Copilot instructions, I can understand the repository better, work faster and produce higher quality PRs.

I can generate a .github/copilot-instructions.md file for you automatically. Click here to open a pre-filled issue and assign it to me. I'll write the instructions, and then tag you for review.

[sahel-sh]

Let's keep vllm for now, a lot of people still use their own gpus for small open source models. Plus I want to make sure that our PyPI package works out side of the box for rankVicuna/Zephyr and FirstMistral without relying on open router.

[ronakice]

Fair point.