Doc needs to be refined and discussed #20

• LLM Internal Observability Infrastructure or something? e.g., DMI is an observability layer for LLM inference, giving real-time access to ANY internal model states with minimal overhead.
• starting scripts need to be unified, e.g., including model name or independent of models.
• adding a road map w/ call for contributions.
• add a section of "why should I care", as the placeholders for potential uses.
• a short visual demo or screenshots are helpful
• missing performance claims. adding overhead numbers. Ideally, a comparison table to Transformer Lens, PyTorch Hooks, and others.
• In addition to a script, maybe adding a short 10-line python example of usage.
• When doc is a bit long, adding table of contents.
• No need to specifically mention ClickHouse. Mentioning of visualizer/grafana.

e.g., some headline pitch

If you're:

• debugging hallucinations and model bugs in production
• studying safety/interpretability and steering
• studying attention collapse
• building distillation datasets
  and more...

You need internal visibility without rewriting your model or slowing your model by 10x.