VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
u_Unrolled
D12: Inference Serving & vLLM · The Inference Reading List — The Papers Behind Every Serving Engine
Phase 6/Domain 12/Lesson 16

The Inference Reading List — The Papers Behind Every Serving Engine

15 min read·Staff+ Tier· Unrolled Pro
The Inference Reading List — The Papers Behind Every Serving Engine

You already know: every mechanism this list describes, from the lessons of this domain. This lesson is the map back to the sources.

Why this matters for Atlas. Engines change every few months; the ideas underneath them change slowly. A staff engineer reads a new serving paper by asking which bottleneck it attacks, what single mechanism removes it, what the headline number was measured against, and where it does not help. This lesson gives that method and then a short, annotated list of the papers that built modern LLM serving, each tied to the lesson that teaches it. The numbers quoted are the papers' own reported results on their workloads, not guarantees for yours.

How to use this list. Read each paper's abstract, introduction and evaluation set-up first; reproduce one number with the lesson's code; only then read the method in full.


1. How to read a systems paper

The question. A new paper claims "3× higher throughput". What do you check before believing it applies to you?

Start from nothing. Five questions, in order:

That's the preview. The full lesson, with its code and exercises, is part of Unrolled Pro.
Staff AI & Pro Tier Lesson

Unlock "The Inference Reading List — The Papers Behind Every Serving Engine"

The foundations are free forever. Unrolled Pro unlocks the complete lesson, production architectures, and 55 battle-tested ML system design case studies.

Everything included with Pro:
All 234+ Deep-Dive Lessons across transformer internals, inference, RAG & agents
55 Staff-Level ML & System Design Case Studies with mathematical capacity estimates
AI Engineer Sheet & Exercises with worked solutions, self-checks, and code
100 Landmark Research Papers guided walkthroughs & progress tracking
14-Day Money-Back Guarantee — 100% full refund if you aren't satisfied
View Plans & Unlock All Lessons

From ₹167/mo (~$2/mo) · Cancel anytime · Instant activation

100% Risk-Free: 14-day no-questions-asked money-back guarantee
Build Atlas · Part 9 of 9: Serve it fastUse this lesson in the project: Add a KV cache to TinyGPT, measure what batching buys, simulate static against continuous batching on real traffic, and turn throughput into a cost per token.