Understand the practical vocabulary you need before choosing a local model
You should be able to demonstrate the result described in the lesson and explain why each step was necessary.
Step-by-step
Tokens are the model input units
A model does not read text as whole sentences. Text is split into tokens. Token counts matter because context limits and API usage are measured around them.
Context is working memory
The context window is the amount of tokenized information the model can consider in one request. Larger is not automatically better; memory use and speed usually increase.
Parameters describe model scale
Parameter count is one rough indicator of model size, but architecture and training quality matter too. Do not choose a model by the biggest number alone.
Quantization trades precision for size
Quantized weights reduce memory use so a model can fit on consumer hardware. Lower-bit quantization can increase speed and reduce VRAM use, but may reduce quality on some tasks.
Benchmark your own workload
Test the exact prompts, coding tasks, or extraction jobs you care about. A model that tops a generic chart may be worse for your application.
You should be able to demonstrate the result described in the lesson and explain why each step was necessary.
Troubleshooting
If a command or screen differs from the lesson, stop and verify the version you are using before forcing a change.
What you learned
You now understand the purpose of this part of the system and have a concrete result you can verify before continuing to the next lesson.