Ghost 8B Beta: Multilingual, Knowledge-Rich, and Cost-Efficient LLM

Ghost 8B Beta is an 8 billion parameter large language model developed by Ghost X and Hieu Lam, based on the Llama 3 architecture. It is designed for excellent multilingual support, superior knowledge capabilities, and cost-effectiveness in production environments. The model is available in two context length versions: 8K and 128K tokens, and supports multilingual function tools by default.

Key Capabilities & Differentiators

• Multilingual Proficiency: Supports 9+ popular languages including English, French, Italian, Spanish, Portuguese, German, Vietnamese, Korean, and Chinese.
• Function Tool Support: Integrates function calling capabilities, allowing for advanced interaction and automation.
• Performance Benchmarks: Outperforms Llama 3 8B Instruct and GPT 3.5 Turbo in AlpacaEval 2.0's length-controlled win rates. Achieves scores comparable to GPT 3.5 Turbo and Claude v1 on MT Bench, and surpasses xAI Grok 1, OpenAI GPT 3.5, and Mistral Mixtral 8x7b on GSM8K (zero-shot).
• Cost-Effective Optimization: Developed with a focus on achieving high capabilities with a modest size, reducing GPU costs for deployment and operation.
• Reproducible Training: Utilizes a unique "Teach the little boy how to cook Saigon Pho" recipe for continual pre-training and fine-tuning, ensuring reproducibility of the model.

Ideal Use Cases

• Multilingual Chatbots: For applications requiring robust conversational AI across diverse languages.
• Complex Task Solving: Excels in tasks requiring reasoning and knowledge, as indicated by strong performance in mathematical (GSM8K) and general knowledge (GPQA) benchmarks.
• Function Calling Applications: For integrating external tools and services within an AI workflow.
• Cost-Sensitive Deployments: Offers a powerful solution for businesses and startups seeking high performance without extensive computational resources.