NLP Lineage

12 September 2023 · Efficiency and infrastructure

vLLM

Efficient Memory Management for LLM Serving with PagedAttention

W. Kwon, Z. Li, S. Zhuang, et al. (9 authors)

Date
12 September 2023
What the date marks
arXiv submission
Kind
Infrastructure
Organisation
UC Berkeley / Stanford / UCSD
Access
Apache 2.0
Other dates
SOSP 2023; open source 2023-06
Threads
In the guided tour
Chapter 9. Making it run anywhere
Learning phases
6. Local models and deployment

Notable: no later milestone in the data is documented as descending from it. How levels are decided

Read the primary source (arxiv.org)

Full summary in progress. Until then, this page shows the milestone's record from the research data.

What it changed

Before it

Serving systems reserved a contiguous block of memory per request for the KV cache.

The problem

Requests vary in length, so most of that reserved memory sat unused while other requests queued.

What it introduced

Manage the KV cache in pages like virtual memory, so blocks are allocated on demand and shared between requests.

Why it matters

Two to four times the throughput at the same cost, and it became the default open serving engine.

What modern models inherit

Paged KV cache and continuous batching in production serving.

What came next

The standard deployment path for open models, including the largest MoE models.

Before and after

The milestones it descends from, and the milestones that descend from it. Work developed alongside it is not counted. Weak links are left out unless you include them.

Before: where it came from

vLLM has 11 documented ancestors across 4 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 4 turning points among them.

vLLM has 11 documented ancestors across 4 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

vLLM has 12 recorded ancestors, including indirect ones, across 4 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 4 turning points among them.

vLLM has 12 recorded ancestors, including indirect ones, across 4 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2023vLLMThis milestone.
  2. 2019Multi-query attentionvLLM was influenced by thisChapter 9. Making it run anywhere
  3. 2017Transformer Turning pointMulti-query attention built directly on thisChapter 3. Attention, then only attention
  4. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  5. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  6. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  7. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  8. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  9. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  10. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  11. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  12. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2023vLLMThis milestone.
  2. 2019Multi-query attentionvLLM was influenced by thisChapter 9. Making it run anywhere
  3. 2017Transformer Turning pointMulti-query attention built directly on thisChapter 3. Attention, then only attention
  4. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  5. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  6. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  7. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  8. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  9. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  10. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  11. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  12. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  13. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

No documented descendants are recorded for vLLM without weak links.

No documented descendants are recorded for vLLM without weak links.

No recorded descendants are recorded for vLLM.

No recorded descendants are recorded for vLLM.

Direct links and their sources

Where it came from

Influenced by

Developed alongside

What it led to

None recorded in the sources used.