Yi 1.5 34 B Chat

Introduction

Yi-1.5 is an enhanced version of the original Yi model, featuring continuous pre-training on a high-quality corpus of 500 billion tokens. It is fine-tuned on 3 million diverse samples, offering improved performance in coding, math, reasoning, and instruction-following, while maintaining strong capabilities in language understanding, commonsense reasoning, and reading comprehension. The model offers context lengths of 4K, 16K, and 32K with 3.6 trillion pre-trained tokens.

Architecture

The models are divided into two categories: chat models and base models. Chat models include variations like Yi-1.5-34B-Chat and Yi-1.5-9B-Chat, which are available on platforms such as Hugging Face, ModelScope, and wisemodel. Base models are also available in different sizes and configurations, emphasizing versatility and adaptability for various applications.

Training

Yi-1.5 models have been pre-trained with an extensive corpus and fine-tuned using a diverse set of samples to enhance their performance across various benchmarks. Yi-1.5-34B-Chat and Yi-1.5-9B-Chat are noted for their competitive performance, often surpassing larger models in specific benchmarks, while the base models also excel in their respective categories.

Guide: Running Locally

To run Yi-1.5 models locally, follow these basic steps:

1. Environment Setup: Ensure you have Python and necessary libraries like PyTorch installed.
2. Model Download: Access the desired model version from platforms like Hugging Face or ModelScope.
3. Loading the Model: Use libraries such as Transformers to load and interact with the model.
4. Inference: Implement the model for your specific task, such as text generation or comprehension evaluation.

For optimal performance, especially with larger models like Yi-1.5-34B, consider using cloud GPUs from providers such as AWS, Google Cloud, or Azure.

License

Yi-1.5 is released under the Apache 2.0 License, allowing for both personal and commercial use with minimal restrictions.