DAY 100 / 210

Launching LLM Inference Fundamentals

Day 100 opens phase-3-inference by grounding the learner in production serving concepts before any optimization work begins. Because no prior days exist, this session creates the baseline mental model needed to evaluate later changes to the Maku flow. The larger arc requires inference fluency to ship reliable AI features inside StartupTribunal.

⏱ 35 min target📝 2 quiz Qs

Resources

readingHugging Face
Transformers Pipelines for Inference
entire page
25 min

Deliverable

First journal entry posted to app/maku/page.tsx documenting one inference misconception and its correction

Quiz · 2 questions

1. Which statement correctly distinguishes training from inference latency concerns?

Training latency is dominated by data movement; inference latency is dominated by KV cache sizeInference is always faster than training because no gradients are computedBoth phases have identical memory bottlenecksInference never requires batching

2. Name one common misconception about model quantization that leads to silent accuracy loss in production.

Journal

Time spent (minutes)

Blockers

Commit / PR links (one per line)