NLP Lineage

27 June 2023 · Efficiency and infrastructure

Position interpolation

Extending Context Window of Large Language Models via Positional Interpolation

S. Chen, S. Wong, L. Chen, Y. Tian

Date
27 June 2023
What the date marks
arXiv submission
Kind
Technique
Organisation
Meta
Access
Not applicable
Other dates
YaRN followed 2023-09
Threads
In the guided tour
Chapter 10. Reading more than you remember
Learning phases
1. Transformer architecture; 4. Embeddings and RAG

A milestone: 1 later milestone descends directly from it, and its descendants reach 7 of the 10 threads. 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

A model trained at 2K tokens broke down badly past that length.

The problem

Retraining from scratch at longer context is prohibitively expensive.

What it introduced

Scale RoPE position indices down to fit the trained range, then fine-tune briefly. 32K context from a 2K model in a thousand steps.

Why it matters

It made long context an affordable post-hoc upgrade rather than a training-time commitment.

What modern models inherit

Context extension as a cheap fine-tuning step.

What came next

YaRN and LongRoPE refined it; nearly every long-context open model uses some variant.

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

Position interpolation has 21 documented ancestors across 7 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 10 turning points among them.

Position interpolation has 21 documented ancestors across 7 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

Position interpolation has 29 recorded ancestors, including indirect ones, across 7 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 10 turning points among them.

Position interpolation has 29 recorded ancestors, including indirect ones, across 7 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2023Position interpolationThis milestone.
  2. 2023LLaMA Turning pointPosition interpolation built directly on thisChapter 8. The weights get out
  3. 2022Chinchilla Turning pointLLaMA built directly on thisChapter 5. Scale becomes a science
  4. 2021GopherChinchilla built directly on thisChapter 5. Scale becomes a science
  5. 2021RoPE Turning pointPosition interpolation built directly on thisChapter 10. Reading more than you remember
  6. 2020Scaling laws Turning pointChinchilla built directly on thisChapter 5. Scale becomes a science
  7. 2019GPT-2 Turning pointScaling laws was influenced by thisChapter 4. Pretrain once, use everywhere
  8. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  9. 2019Transformer-XLRoPE was influenced by thisChapter 4. Pretrain once, use everywhere
  10. 2018GPT-1GPT-2 built directly on thisChapter 4. Pretrain once, use everywhere
  11. 2017Transformer Turning pointRoPE built directly on thisChapter 3. Attention, then only attention
  12. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  13. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  14. 2015LM pretraining Turning pointGPT-1 was influenced by thisChapter 2. Words become vectors
  15. 2015BPE tokenizationLLaMA built directly on thisChapter 2. Words become vectors
  16. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  17. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  18. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  19. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  20. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  21. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  22. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2023Position interpolationThis milestone.
  2. 2023LLaMA Turning pointPosition interpolation built directly on thisChapter 8. The weights get out
  3. 2022Chinchilla Turning pointLLaMA built directly on thisChapter 5. Scale becomes a science
  4. 2021GopherChinchilla built directly on thisChapter 5. Scale becomes a science
  5. 2021RoPE Turning pointPosition interpolation built directly on thisChapter 10. Reading more than you remember
  6. 2020Scaling laws Turning pointChinchilla built directly on thisChapter 5. Scale becomes a science
  7. 2019GPT-2 Turning pointScaling laws was influenced by thisChapter 4. Pretrain once, use everywhere
  8. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  9. 2019Transformer-XLRoPE was influenced by thisChapter 4. Pretrain once, use everywhere
  10. 2018GPT-1GPT-2 built directly on thisChapter 4. Pretrain once, use everywhere
  11. 2018ELMoGPT-1 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  12. 2018ULMFiTGPT-1 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  13. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  14. 2017Transformer Turning pointRoPE built directly on thisChapter 3. Attention, then only attention
  15. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  16. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  17. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  18. 2015LM pretraining Turning pointGPT-1 was influenced by thisChapter 2. Words become vectors
  19. 2015BPE tokenizationLLaMA built directly on thisChapter 2. Words become vectors
  20. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  21. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  22. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  23. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  24. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  25. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  26. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  27. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  28. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  29. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  30. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

Position interpolation has 10 documented descendants across 3 chapters, reaching forward to Qwen3.8 in 2026. Showing its direct descendants.

Position interpolation has 10 documented descendants across 3 chapters, reaching forward to Qwen3.8 in 2026. Showing all of them.

Position interpolation has 15 recorded descendants, including indirect ones, across 3 chapters, reaching forward to Qwen3.8 in 2026. Showing its direct descendants.

Position interpolation has 15 recorded descendants, including indirect ones, across 3 chapters, reaching forward to Qwen3.8 in 2026. Showing all of them.

  1. 2023Position interpolationThis milestone.
  2. 2024Llama 3.1 405Bwas influenced by Position interpolationChapter 8. The weights get out
  3. 2024DeepSeek-V3was influenced by Llama 3.1 405BChapter 8. The weights get out
  4. 2025DeepSeek-R1built directly on DeepSeek-V3Chapter 12. Thinking longer
  5. 2025Llama 4succeeded Llama 3.1 405BChapter 8. The weights get out
  6. 2025Qwen3was influenced by DeepSeek-R1Chapter 8. The weights get out
  7. 2025DeepSeek-V3.2built directly on DeepSeek-V3Chapter 10. Reading more than you remember
  8. 2026Qwen3.5built directly on Qwen3Chapter 8. The weights get out
  9. 2026Muse Sparksucceeded Llama 4Chapter 8. The weights get out
  10. 2026DeepSeek-V4was influenced by DeepSeek-V3Chapter 8. The weights get out
  11. 2026Qwen3.8built directly on Qwen3.5Chapter 8. The weights get out
  1. 2023Position interpolationThis milestone.
  2. 2024Llama 3.1 405Bwas influenced by Position interpolationChapter 8. The weights get out
  3. 2024DeepSeek-V3was influenced by Llama 3.1 405BChapter 8. The weights get out
  4. 2025DeepSeek-R1built directly on DeepSeek-V3Chapter 12. Thinking longer
  5. 2025Llama 4succeeded Llama 3.1 405BChapter 8. The weights get out
  6. 2025Qwen3was influenced by DeepSeek-R1Chapter 8. The weights get out
  7. 2025Kimi K2was influenced by DeepSeek-V3 (indirect influence)Chapter 8. The weights get out
  8. 2025gpt-osswas influenced by DeepSeek-R1 (indirect influence)Chapter 8. The weights get out
  9. 2025GLM-4.5was influenced by DeepSeek-R1 (indirect influence)Chapter 8. The weights get out
  10. 2025DeepSeek-V3.2built directly on DeepSeek-V3Chapter 10. Reading more than you remember
  11. 2026Qwen3.5built directly on Qwen3Chapter 8. The weights get out
  12. 2026Muse Sparksucceeded Llama 4Chapter 8. The weights get out
  13. 2026DeepSeek-V4was influenced by DeepSeek-V3Chapter 8. The weights get out
  14. 2026GLM-5.2built directly on GLM-4.5Chapter 8. The weights get out
  15. 2026Kimi K3built directly on Kimi K2Chapter 8. The weights get out
  16. 2026Qwen3.8built directly on Qwen3.5Chapter 8. The weights get out

Direct links and their sources

Where it came from

Built directly on

What it led to

Influenced

Terms