Please enable JavaScript.

Coggle requires JavaScript to display documents.

IRCC, AI Clusters, Large Language Models (LLMs), Applications/Users -…

- - - - Meta LLaMA Models (LLaMA 2 / LLaMA 3)
        
        • Strengths: General-purpose reasoning, multilingual capabilities, strong base for fine-tuning.
        
        • Best for: Custom chatbots, research, instruction following.
        
        • Vicuna – Chat-focused.
        
        • Alpaca – Instruction-following.
        
        • OpenHermes – Balanced, conversational and general reasoning.
    - - Mistral 7B / Mixtral (Mixture of Experts)
        
        • Strengths: Fast, efficient, and very strong performance for their size.
        
        • Best for: Instruction-following, chatbots, and even coding tasks.
        
        • Mistral 7B – Lightweight and powerful.
        
        • Mixtral – Uses only part of the model per query (efficient Mixture of Experts), great for multitask dialogue and cost-effective serving.
    - - Phi-1.5 / Phi-2 (Microsoft)
        
        • Strengths: Small model, trained on textbook-quality data.
        
        • Best for: Lightweight tutoring bots, educational tools, reasoning on a tight budget.
        
        • Phi-2 can do math, logic puzzles, and basic reasoning surprisingly well for its 1.3B size.
    - - GPT-J (6B) / GPT-NeoX (20B)
        
        • Strengths: Early open-weight models, good at natural language generation.
        
        • Best for: Writing, summarization, general content generation.
        
        • Note: GPT-J is easier to run locally; NeoX has higher performance but needs more resources.