budecosystem/genz-13b-v2

GenZ-13B-v2: An Advanced Llama V2 Fine-tune

GenZ-13B-v2, developed by Bud Ecosystem, is a 13 billion parameter Large Language Model (LLM) built upon Meta's Llama V2 architecture. This model is fine-tuned to serve as a sophisticated AI assistant, focusing on delivering high-quality and informative responses to user prompts. A key differentiator is its accessibility, with quantized versions (4-bit and ggml) enabling inference on devices with limited GPU memory or even CPU-only setups.

Key Capabilities & Features

• Enhanced Assistant: Designed for understanding and responding to user prompts with high-quality, user-focused interactions.
• Versatile Generation: Capable of text summarization, text generation, and chatbot creation, as demonstrated by examples in code, poem, and email generation.
• Improved Performance: GenZ-13B-v2 shows improved evaluation results compared to its predecessor, GenZ-13B V1, particularly in roleplay and mathematical tasks, achieving an MT Bench score of 6.79.
• Open-Source Focus: Part of Bud Ecosystem's initiative to democratize access to fine-tuned LLMs, with plans for a series of models across different parameter counts and quantizations.
• Supervised Fine-Tuning (SFT): Trained using a meticulous SFT process on curated datasets, including OpenAssistant and Thought Source for Chain Of Thought (CoT) approaches.

Good For

• Research: An excellent foundation for research on large language models.
• Specialized Fine-tuning: Ideal for further specialization and fine-tuning for specific use cases.
• Accessible Deployment: Quantized versions allow for deployment and experimentation on personal computers and environments with limited resources.
• Text-based Applications: Suitable for developing applications requiring text summarization, generation, or conversational AI.