What is an AI model?

You need a model and an app

Think of the model as the part that learned patterns, and the app as the part you use. Downloading a model does not install a chat app. Ollama, LM Studio and the Hugging Bay app can load supported model files.

Five words you will see

• Weights: the learned numbers stored in the model files.
• Parameters: those numbers, counted. 1B means about one billion; it does not mean a 1 GB download.
• Quantization: storing numbers with less precision to reduce size. Q4 is a common four-bit family; the exact file size still matters.
• Context: the text a model can work with at once, including the conversation and its reply. Longer context needs more memory.
• Instruct or Chat: a model adapted to follow instructions or hold a conversation.

The file type must match the app

GGUF is a model file format used by llama.cpp and supported by runners such as Ollama and LM Studio. Safetensors and MLX model folders need a compatible loader and may need several files. Renaming a file does not convert it.

For a first text chat, choose a complete chat model in a format your app supports. A tokenizer, image adapter or one part of a split download is not a complete model.

Try a small, ordinary question

Ask: “Explain what a rainbow is in three short sentences.” A reply shows that the model loaded; it does not prove that every answer is correct. Check facts before relying on them.