Feat: Add MPS fallback to CUDA support and uv-compatible CLI for FastVLM inference

[Deepchavda007]

Summary

1. Automatic device selection: Inference now automatically selects the best available hardware — preferring Apple Silicon (MPS), falling back to CUDA if available, and then to CPU.
2. Modern Python environment support via uv: Added instructions and compatibility for using uv, a fast dependency manager designed for Python 3.10+, as an alternative to Conda.

---

• predict.py

  • Replaced hardcoded "mps" device usage with a dynamic fallback:

    if torch.backends.mps.is_available():
        DEVICE = torch.device("mps")
        logger.info("Using MPS device")
    elif torch.cuda.is_available():
        DEVICE = torch.device("cuda")
        logger.info("Using CUDA device")
    else:
        DEVICE = torch.device("cpu")
        logger.info("Using CPU device")

---

Type of change

• Improvement
• Documentation update
• Bug fix

---

Checklist

• Inference tested on MacBook (MPS)
• Inference tested with CUDA-enabled GPU
• Tested with uv environment setup and pip fallback
• Backward-compatible with conda environments

---