C4AI Command R+ 08-2024 Model Card

Introduction

C4AI Command R+ 08-2024 is an advanced 104 billion parameter model designed for multilingual tasks and complex tool use, such as Retrieval Augmented Generation (RAG). It supports 23 languages and is optimized for reasoning, summarization, and question answering. This model is part of a series of open weight releases by Cohere For AI.

Architecture

C4AI Command R+ 08-2024 is an auto-regressive language model utilizing an optimized transformer architecture, enhanced with supervised fine-tuning and preference training. It features a context length of 128K and utilizes Grouped Query Attention (GQA) for faster inference. The model supports grounded generation, incorporating citations and source information, and is trained for both single-step and multi-step tool use.

Training

The model is trained on 23 languages and evaluated on 10, encompassing languages such as English, French, and Simplified Chinese. It employs grounded generation, using documents to produce cited and verified responses. The training includes supervised fine-tuning and preference adjustments to align with human preferences for helpfulness and safety.

Guide: Running Locally

Environment Setup: Ensure you have Python and pip installed.
Install Transformers: pip install 'transformers>=4.39.1'

Load Model:

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "CohereForAI/c4ai-command-r-plus-08-2024"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

Generate Text: Use the model to generate responses using the chat template.
Cloud GPUs: For optimal performance, consider using cloud GPU services like AWS, Azure, or Google Cloud.

License

C4AI Command R+ 08-2024 is released under the CC-BY-NC 4.0 license, which restricts use to non-commercial purposes. Users must also comply with the C4AI Acceptable Use Policy.

For more information, visit the C4AI License Agreement and Acceptable Use Policy.