Serving Large Language Models
Trace a worked example
From goal to measurement to ship-or-abort for LLM serving on the multi-tenant chat API.
1Learn the idea
Read
Trace
See it
Training
Inference
Training = long study · Inference = quick answer from what it already learned
Goal: improve the multi-tenant chat API using LLM serving without breaking protected slices.
Mechanism reminder for LLM serving: Load weights into GPU/CPU memory, batch prefill/decode, manage KV cache, admit/queue requests, stream tokens, apply auth quotas.
Baseline and shock: approximate weight memory for 7B parameters at 16 bits = 7B × 2 bytes ≈ 14 GB, before KV cache and runtime overhead
Tradeoff in play for LLM serving: Larger batches improve GPU utilization and tokens per second but can worsen queueing and per-user latency. Quantization reduces memory and may increase throughput with possible quality loss. Longer contexts expand usefulness while sharply increasing KV memory and prefill work.
Ship decision: Admit long prompts to a separate pool; keep interactive p95 TTFT under SLO at target QPS.
Rollback triggers for LLM serving must cite tokens/sec, p95 TTFT, p95 TPOT, queue time, OOM rate, cost/1M tokens. If you cannot name a tolerated regression on the multi-tenant chat API, do not promote the change.
Read
Make it operational
Keep the LLM serving decision record beside the multi-tenant chat API code paths that implement it. Future you will not remember why a default exists unless the evidence is linked from the config for LLM serving.
Also pin one numeric memory from this LLM serving chapter: approximate weight memory for 7B parameters at 16 bits = 7B × 2 bytes ≈ 14 GB, before KV cache and runtime overhead That number is not decoration; it is a template for how claims about LLM serving on the multi-tenant chat API should look in design docs. Scoped specifically to LLM serving / multi-tenant chat API / worked-trace.
Read
Common mix-ups
People confuse LLM serving with neighboring buzzwords when debugging the multi-tenant chat API. Before changing prompts, ask whether the broken stage was evidence gathering, the LLM serving judgment itself, validation, or the product action. Fixing the wrong stage creates folklore (“we tried LLM serving and it failed”) that blocks the next team on the multi-tenant chat API. Scoped specifically to LLM serving / multi-tenant chat API / worked-trace.
Read
Rehearsal (serving-llms/worked-trace)
Write a five-line artifact for this page: goal, inputs, check, owner, stop rule. Invent one fluent failure that the check would catch. Keep details specific to serving llms rather than generic AI advice.
Read
Rehearsal (serving-llms/worked-trace)
Write a five-line artifact for this page: goal, inputs, check, owner, stop rule. Invent one fluent failure that the check would catch. Keep details specific to serving llms rather than generic AI advice.
Read
Rehearsal (serving-llms/worked-trace)
Write a five-line artifact for this page: goal, inputs, check, owner, stop rule. Invent one fluent failure that the check would catch. Keep details specific to serving llms rather than generic AI advice.
Read
Chapter close 1
For serving llms, add acceptance test 1: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 2
For serving llms, add acceptance test 2: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 3
For serving llms, add acceptance test 3: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 4
For serving llms, add acceptance test 4: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 5
For serving llms, add acceptance test 5: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 6
For serving llms, add acceptance test 6: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 7
For serving llms, add acceptance test 7: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 8
For serving llms, add acceptance test 8: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 9
For serving llms, add acceptance test 9: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 10
For serving llms, add acceptance test 10: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 11
For serving llms, add acceptance test 11: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 12
For serving llms, add acceptance test 12: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 13
For serving llms, add acceptance test 13: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 14
For serving llms, add acceptance test 14: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 15
For serving llms, add acceptance test 15: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 16
For serving llms, add acceptance test 16: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 17
For serving llms, add acceptance test 17: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 18
For serving llms, add acceptance test 18: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 19
For serving llms, add acceptance test 19: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 20
For serving llms, add acceptance test 20: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 21
For serving llms, add acceptance test 21: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 22
For serving llms, add acceptance test 22: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 23
For serving llms, add acceptance test 23: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 24
For serving llms, add acceptance test 24: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 25
For serving llms, add acceptance test 25: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 26
For serving llms, add acceptance test 26: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 27
For serving llms, add acceptance test 27: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 28
For serving llms, add acceptance test 28: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 29
For serving llms, add acceptance test 29: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 30
For serving llms, add acceptance test 30: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 31
For serving llms, add acceptance test 31: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 32
For serving llms, add acceptance test 32: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 33
For serving llms, add acceptance test 33: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 34
For serving llms, add acceptance test 34: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Read
Chapter close 35
For serving llms, add acceptance test 35: evidence source, threshold, and signer. Keep it unique to this chapter's scenario.
Go deeper
Before you start
Why this matters
List the constraints (latency, cost, privacy, review capacity) that any LLM serving change must respect for the multi-tenant chat API.
In the wild
See how this idea shows up as a product and a company — then come back to the lesson. Skills transfer across vendors.
Related lessons
Check your understanding
Page assessment
Answer from memory. Completion is saved from this evidence, not from opening the next page.
All responses are required.