Arch Function 7 B

Introduction

The Katanemo Arch-Function collection is a set of state-of-the-art (SOTA) large language models (LLMs) designed for function calling tasks. These models are capable of understanding complex function signatures, identifying required parameters, and generating accurate function call outputs based on natural language prompts. They are optimized for scenarios where automated API interaction and function execution are crucial.

Architecture

The Arch-Function models are designed for optimal performance in function calling, with features including single, parallel, and multiple function calling. They offer high generalization across various use cases and are optimized for low-latency, high-throughput environments. The core LLM is integrated into the open-source Arch Gateway, facilitating seamless interaction between user prompts and developer APIs.

Training

The models in the Arch-Function collection are built on the Qwen 2.5 architecture. Detailed technical information about the training process and model capabilities will be published in an upcoming blog.

Guide: Running Locally

1. Install Requirements: Ensure you have the Hugging Face transformers library version 4.37.0 or higher.

   pip install transformers>=4.37.0
   

2. Load Model and Tokenizer:

   from transformers import AutoModelForCausalLM, AutoTokenizer
   
   model_name = "katanemo/Arch-Function-7B"
   model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype="auto", trust_remote_code=True)
   tokenizer = AutoTokenizer.from_pretrained(model_name)
   

3. Run Examples:

   • Use the provided prompt format for the best performance.
   • Implement the example code to perform single and multi-turn function calling tasks.

4. Cloud GPUs: For enhanced performance and efficiency, consider using cloud GPUs available on platforms like AWS, GCP, or Azure.

License

The Katanemo Arch-Function collection is distributed under the Katanemo license. For more details, refer to the license document.