Linkbricks Horizon A I Japanese Pro V5 70 B

Introduction

The Linkbricks-Horizon-AI-Japanese-Pro-V5-70B model is a powerful language model designed for Japanese text generation, developed by Saxo, a data scientist at Linkbricks. It leverages advanced techniques and datasets to enhance its capabilities across multiple languages, including Korean, Chinese, and English.

Architecture

The model is based on the Saxo/Linkbricks-Horizon-AI-Japanese-Advanced-V4-70B architecture and utilizes the Transformers library. It features a 128k-context window and supports function calling. The model is designed with DeepSpeed Stage 3, rslora, and BAdam Layer Mode to optimize performance.

Training

The model was fine-tuned using 8 H100-80G GPUs, employing a training process involving SFT (Supervised Fine-Tuning), DPO (Data Parallel Optimization), and MERGE. The training used a 20 million Japanese news and wiki corpus, alongside cross-training data for Japanese, Korean, Chinese, and English languages, as well as logic and mathematics data. The tokenizer remains unchanged from the base model.

Guide: Running Locally

To run the model locally, follow these steps:

1. Install Dependencies: Ensure you have Python and the Transformers library installed. You can install the Transformers library via pip:

   pip install transformers
   

2. Download the Model: Access the model files from the Hugging Face repository.

3. Load the Model: Use the Transformers library to load and run the model:

   from transformers import AutoModelForCausalLM, AutoTokenizer
   
   model_name = "Saxo/Linkbricks-Horizon-AI-Japanese-Pro-V5-70B"
   tokenizer = AutoTokenizer.from_pretrained(model_name)
   model = AutoModelForCausalLM.from_pretrained(model_name)
   

4. Inference: Use the loaded model for text generation tasks.

For optimal performance, it is recommended to use cloud-based GPUs such as those provided by AWS, Google Cloud, or Azure, especially given the model's size and complexity.

License

The Linkbricks-Horizon-AI-Japanese-Pro-V5-70B model is licensed under the Apache 2.0 License. This allows for extensive use, modification, and distribution, provided that proper attribution is given and that any modifications are also shared under the same license.