Feature Request: Add LlamaCppModel Support to smolagents

[ryantzr1]

Motivation Behind This Feature

As the landscape of language models continues to evolve, integrating diverse model architectures becomes crucial for enhancing the versatility and applicability of libraries like smolagents. The llama.cpp framework offers an efficient and optimized way to run large language models with reduced resource consumption, making it an attractive option for developers and researchers.

Current Challenges:

• Limited Model Support: While smolagents currently supports models like those from Hugging Face's Transformers library, there's a growing demand for integrating models managed by llama.cpp.
• Performance Optimization: llama.cpp provides optimized performance for running large language models on resource-constrained environments, which is beneficial for users who require high efficiency without compromising on model capabilities.

Proposed Solution:
Introduce a new LlamaCppModel class that seamlessly integrates llama.cpp models into the smolagents ecosystem, ensuring proper parameter handling and conditional tool usage.

Detailed Description

The LlamaCppModel class is designed to interact with llama.cpp models, providing robust parameter management and the ability to utilize tools only when explicitly provided. This integration ensures that users can leverage the efficiency of llama.cpp while maintaining the flexibility and functionality that smolagents offers.

Key Features:

• Flexible Model Loading: Supports loading models from a local path or directly from a Hugging Face repository.
• Parameter Management: Allows customization of GPU layers, context size, and maximum token generation.
• Conditional Tool Integration: Integrates tools seamlessly when they are passed, ensuring optimized performance.