LLMs are like the wild west. Things are moving fast and no one is really documenting things for outsiders. The only way to learn these things is by using them.

LLMs are like the wild west. Things are moving fast and no one is really documenting things for outsiders. The only way to learn these things is by using them. If you tried out Oobabooga, you would know that each LLM needs their own template instructions to understand its input and respond correctly. More recent tools, like Ollama, have these instructions included with the models. However, there are still different types of models available through Ollama. Some of them are suffixed with "instruct" which tells you that you can use them like a chat bot, whereas other "main" models only do completions and spit out gibberish if you ask them a question.