NLP Lineage

3 September 2021 · Alignment

FLAN

Finetuned Language Models Are Zero-Shot Learners

J. Wei, M. Bosma, V. Zhao, K. Guu, A. Yu, B. Lester, N. Du, A. Dai, et al. (9 authors)

Date
3 September 2021
What the date marks
arXiv submission
Kind
Training method
Organisation
Google Research
Access
Apache 2.0
Other dates
ICLR 2022
Threads
In the guided tour
Chapter 6. Teaching models what people want
Learning phases
2. How LLMs are trained

A milestone: 3 later milestones descend directly from it, and its descendants reach 10 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

Few-shot prompting worked, but models did not reliably follow a plain instruction.

The problem

A pretrained model completes text; it has no particular reason to do what you ask.

What it introduced

Fine-tune on many tasks phrased as natural-language instructions, then test on held-out task types.

Why it matters

Instruction following turned out to be a learnable skill that generalises. This is the step before RLHF.

What modern models inherit

Instruction tuning as the first stage of every assistant's post-training.

What came next

T0 in parallel, Flan-T5 at scale, and InstructGPT's supervised stage.

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

FLAN has 29 documented ancestors across 5 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 12 turning points among them.

FLAN has 29 documented ancestors across 5 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

FLAN has 31 recorded ancestors, including indirect ones, across 5 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 12 turning points among them.

FLAN has 31 recorded ancestors, including indirect ones, across 5 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2021FLANThis milestone.
  2. 2020GPT-3 Turning pointFLAN was influenced by thisChapter 5. Scale becomes a science
  3. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  4. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  5. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  6. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  7. 2019GPT-2 Turning pointT5 was influenced by this
    GPT-3 built directly on this    Chapter 4. Pretrain once, use everywhere
  8. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  9. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  10. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  11. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  12. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  13. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  14. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  15. 2017Transformer Turning pointT5 built directly on thisChapter 3. Attention, then only attention
  16. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  17. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  18. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  19. 2015BPE tokenizationGPT-2 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 this
    ELMo built directly on this    Chapter 1. Language as probability
  30. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2021FLANThis milestone.
  2. 2020GPT-3 Turning pointFLAN was influenced by thisChapter 5. Scale becomes a science
  3. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  4. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  5. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  6. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  7. 2019GPT-2 Turning pointT5 was influenced by this
    GPT-3 built directly on this    Chapter 4. Pretrain once, use everywhere
  8. 2019WebTextGPT-2 was influenced by this
    C4 was influenced by this (indirect influence)    Chapter 4. Pretrain once, use everywhere
  9. 2019Transformer-XLSparse Transformer was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  10. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  11. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  12. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  13. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  14. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  15. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  16. 2017Transformer Turning pointT5 built directly on thisChapter 3. Attention, then only attention
  17. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  18. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  19. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  20. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  21. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  22. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  23. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  24. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  25. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  26. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  27. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  28. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  29. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  30. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  31. 1997LSTM Turning pointSeq2seq built directly on this
    ELMo built directly on this    Chapter 1. Language as probability
  32. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

FLAN has 59 documented descendants across 8 chapters, reaching forward to Claude Opus 5.5 in 2026. Showing its direct descendants and the 4 turning points among them.

FLAN has 59 documented descendants across 8 chapters, reaching forward to Claude Opus 5.5 in 2026. Showing all of them.

FLAN has 71 recorded descendants, including indirect ones, across 8 chapters, reaching forward to Claude Opus 5.5 in 2026. Showing its direct descendants and the 4 turning points among them.

FLAN has 71 recorded descendants, including indirect ones, across 8 chapters, reaching forward to Claude Opus 5.5 in 2026. Showing all of them.

  1. 2021FLANThis milestone.
  2. 2022InstructGPT Turning pointwas influenced by FLANChapter 6. Teaching models what people want
  3. 2022Flan-T5 / Flan-PaLMbuilt directly on FLANChapter 6. Teaching models what people want
  4. 2022ChatGPTbuilt directly on InstructGPTChapter 7. The assistant goes public
  5. 2022GPT-3.5 Turning pointbuilt directly on InstructGPTChapter 7. The assistant goes public
  6. 2022Self-Instructwas influenced by FLANChapter 6. Teaching models what people want
  7. 2023Alpacabuilt directly on Self-InstructChapter 8. The weights get out
  8. 2023GPT-4 Turning pointbuilt directly on InstructGPTChapter 7. The assistant goes public
  9. 2023AutoGPTwas influenced by GPT-4Chapter 13. Models that act
  10. 2023LLaVAbuilt directly on GPT-4Chapter 11. Seeing and hearing
  11. 2023DPObuilt directly on InstructGPTChapter 6. Teaching models what people want
  12. 2023phi-1was influenced by GPT-3.5Chapter 8. The weights get out
  13. 2023Llama 2 Turning pointwas influenced by InstructGPTChapter 8. The weights get out
  14. 2023GPT-4Vbuilt directly on GPT-4Chapter 11. Seeing and hearing
  15. 2023SWE-benchwas influenced by GPT-4Chapter 13. Models that act
  16. 2024DeepSeek LLMwas influenced by Llama 2Chapter 8. The weights get out
  17. 2024OLMowas influenced by Llama 2Chapter 8. The weights get out
  18. 2024GRPOwas influenced by InstructGPTChapter 12. Thinking longer
  19. 2024Gemmawas influenced by Llama 2Chapter 8. The weights get out
  20. 2024Claude 3was influenced by GPT-4Chapter 14. Where the threads meet now
  21. 2024Chatbot Arenawas influenced by ChatGPTChapter 14. Where the threads meet now
  22. 2024Devinwas influenced by GPT-4Chapter 13. Models that act
  23. 2024Llama 3built directly on Llama 2Chapter 8. The weights get out
  24. 2024Phi-3built directly on phi-1Chapter 8. The weights get out
  25. 2024SWE-agentbuilt directly on SWE-benchChapter 13. Models that act
  26. 2024GPT-4obuilt directly on GPT-4Chapter 11. Seeing and hearing
  27. 2024Claude 3.5 Sonnetbuilt directly on Claude 3Chapter 14. Where the threads meet now
  28. 2024FineWebwas influenced by phi-1Chapter 8. The weights get out
  29. 2024Llama 3.1 405Bbuilt directly on DPOChapter 8. The weights get out
  30. 2024Computer usewas influenced by GPT-4VChapter 13. Models that act
  31. 2024DeepSeek-V3built directly on GRPOChapter 8. The weights get out
  32. 2025DeepSeek-R1built directly on GRPOChapter 12. Thinking longer
  33. 2025Claude Codewas influenced by Computer useChapter 13. Models that act
  34. 2025Claude 3.7built directly on Claude 3.5 SonnetChapter 12. Thinking longer
  35. 2025GPT-4.5built directly on GPT-4oChapter 14. Where the threads meet now
  36. 2025Llama 4succeeded Llama 3.1 405BChapter 8. The weights get out
  37. 2025Qwen3was influenced by DeepSeek-R1Chapter 8. The weights get out
  38. 2025Claude 4succeeded Claude 3
    was influenced by SWE-bench    Chapter 13. Models that act
  39. 2025Kimi K2was influenced by GRPOChapter 8. The weights get out
  40. 2025GPT-5succeeded GPT-4oChapter 14. Where the threads meet now
  41. 2025DeepSeek-V3.2built directly on DeepSeek-V3Chapter 10. Reading more than you remember
  42. 2025Claude Sonnet 4.5built directly on Claude 4Chapter 13. Models that act
  43. 2025GPT-5.1built directly on GPT-5Chapter 14. Where the threads meet now
  44. 2025Claude Opus 4.5was influenced by SWE-benchChapter 13. Models that act
  45. 2025GPT-5.2built directly on GPT-5.1Chapter 14. Where the threads meet now
  46. 2026Claude Opus 4.6built directly on Claude Opus 4.5Chapter 14. Where the threads meet now
  47. 2026GPT-5.3-Codexbuilt directly on GPT-5.2Chapter 13. Models that act
  48. 2026Qwen3.5built directly on Qwen3Chapter 8. The weights get out
  49. 2026GPT-5.4built directly on GPT-5.3-CodexChapter 14. Where the threads meet now
  50. 2026Muse Sparksucceeded Llama 4Chapter 8. The weights get out
  51. 2026Claude Opus 4.7built directly on Claude Opus 4.6Chapter 14. Where the threads meet now
  52. 2026GPT-5.5built directly on GPT-5.4Chapter 14. Where the threads meet now
  53. 2026DeepSeek-V4was influenced by DeepSeek-V3Chapter 8. The weights get out
  54. 2026Claude Opus 4.8built directly on Claude Opus 4.7Chapter 14. Where the threads meet now
  55. 2026Claude Sonnet 5succeeded Claude Opus 4.8Chapter 14. Where the threads meet now
  56. 2026GPT-5.6 Solbuilt directly on GPT-5.5Chapter 14. Where the threads meet now
  57. 2026Kimi K3built directly on Kimi K2Chapter 8. The weights get out
  58. 2026Claude Opus 5succeeded Claude Opus 4.8Chapter 14. Where the threads meet now
  59. 2026Qwen3.8built directly on Qwen3.5Chapter 8. The weights get out
  60. 2026Claude Opus 5.5built directly on Claude Opus 5Chapter 14. Where the threads meet now
  1. 2021FLANThis milestone.
  2. 2022InstructGPT Turning pointwas influenced by FLANChapter 6. Teaching models what people want
  3. 2022Flan-T5 / Flan-PaLMbuilt directly on FLANChapter 6. Teaching models what people want
  4. 2022ChatGPTbuilt directly on InstructGPTChapter 7. The assistant goes public
  5. 2022GPT-3.5 Turning pointbuilt directly on InstructGPTChapter 7. The assistant goes public
  6. 2022Self-Instructwas influenced by FLANChapter 6. Teaching models what people want
  7. 2023Alpacabuilt directly on Self-InstructChapter 8. The weights get out
  8. 2023GPT-4 Turning pointbuilt directly on InstructGPTChapter 7. The assistant goes public
  9. 2023AutoGPTwas influenced by GPT-4Chapter 13. Models that act
  10. 2023LLaVAbuilt directly on GPT-4Chapter 11. Seeing and hearing
  11. 2023DPObuilt directly on InstructGPTChapter 6. Teaching models what people want
  12. 2023phi-1was influenced by Self-Instruct (indirect influence)Chapter 8. The weights get out
  13. 2023Llama 2 Turning pointwas influenced by InstructGPTChapter 8. The weights get out
  14. 2023GPT-4Vbuilt directly on GPT-4Chapter 11. Seeing and hearing
  15. 2023SWE-benchwas influenced by GPT-4Chapter 13. Models that act
  16. 2024DeepSeek LLMwas influenced by Llama 2Chapter 8. The weights get out
  17. 2024OLMowas influenced by Llama 2Chapter 8. The weights get out
  18. 2024GRPOwas influenced by InstructGPTChapter 12. Thinking longer
  19. 2024Gemmawas influenced by Llama 2Chapter 8. The weights get out
  20. 2024Claude 3was influenced by GPT-4Chapter 14. Where the threads meet now
  21. 2024Chatbot Arenawas influenced by ChatGPTChapter 14. Where the threads meet now
  22. 2024Devinwas influenced by GPT-4Chapter 13. Models that act
  23. 2024Llama 3built directly on Llama 2Chapter 8. The weights get out
  24. 2024Phi-3built directly on phi-1Chapter 8. The weights get out
  25. 2024SWE-agentbuilt directly on SWE-benchChapter 13. Models that act
  26. 2024GPT-4obuilt directly on GPT-4Chapter 11. Seeing and hearing
  27. 2024Claude 3.5 Sonnetbuilt directly on Claude 3Chapter 14. Where the threads meet now
  28. 2024FineWebwas influenced by phi-1Chapter 8. The weights get out
  29. 2024Llama 3.1 405Bbuilt directly on DPOChapter 8. The weights get out
  30. 2024Computer usewas influenced by GPT-4VChapter 13. Models that act
  31. 2024Gemini 2.0was influenced by Computer use (indirect influence)Chapter 14. Where the threads meet now
  32. 2024DeepSeek-V3built directly on GRPOChapter 8. The weights get out
  33. 2025DeepSeek-R1built directly on GRPOChapter 12. Thinking longer
  34. 2025Claude Codewas influenced by Computer use
    was influenced by SWE-agent (indirect influence)    Chapter 13. Models that act
  35. 2025Claude 3.7built directly on Claude 3.5 SonnetChapter 12. Thinking longer
  36. 2025GPT-4.5built directly on GPT-4oChapter 14. Where the threads meet now
  37. 2025Gemini 2.5built directly on Gemini 2.0Chapter 12. Thinking longer
  38. 2025Llama 4succeeded Llama 3.1 405BChapter 8. The weights get out
  39. 2025Qwen3was influenced by DeepSeek-R1Chapter 8. The weights get out
  40. 2025Claude 4succeeded Claude 3
    was influenced by SWE-bench    Chapter 13. Models that act
  41. 2025Kimi K2was influenced by GRPOChapter 8. The weights get out
  42. 2025IMO goldbuilt directly on Gemini 2.5Chapter 12. Thinking longer
  43. 2025gpt-osswas influenced by DeepSeek-R1 (indirect influence)Chapter 8. The weights get out
  44. 2025GPT-5succeeded GPT-4oChapter 14. Where the threads meet now
  45. 2025GLM-4.5was influenced by DeepSeek-R1 (indirect influence)Chapter 8. The weights get out
  46. 2025DeepSeek-V3.2built directly on DeepSeek-V3Chapter 10. Reading more than you remember
  47. 2025Claude Sonnet 4.5built directly on Claude 4Chapter 13. Models that act
  48. 2025IMO-ProofBenchwas influenced by IMO goldChapter 12. Thinking longer
  49. 2025GPT-5.1built directly on GPT-5Chapter 14. Where the threads meet now
  50. 2025Gemini 3built directly on Gemini 2.5Chapter 14. Where the threads meet now
  51. 2025Claude Opus 4.5was influenced by SWE-benchChapter 13. Models that act
  52. 2025GPT-5.2built directly on GPT-5.1Chapter 14. Where the threads meet now
  53. 2026GPT-5.3-Codexwas influenced by Computer use (indirect influence)Chapter 13. Models that act
  54. 2026Claude Opus 4.6built directly on Claude Opus 4.5Chapter 14. Where the threads meet now
  55. 2026Gemini 3 Deep Thinkbuilt directly on Gemini 3
    was influenced by IMO gold    Chapter 12. Thinking longer
  56. 2026Qwen3.5built directly on Qwen3Chapter 8. The weights get out
  57. 2026Gemini 3.1built directly on Gemini 3Chapter 14. Where the threads meet now
  58. 2026GPT-5.4built directly on GPT-5.3-CodexChapter 14. Where the threads meet now
  59. 2026Project Glasswingwas influenced by Computer use (indirect influence)Chapter 14. Where the threads meet now
  60. 2026Muse Sparksucceeded Llama 4Chapter 8. The weights get out
  61. 2026Claude Opus 4.7built directly on Claude Opus 4.6Chapter 14. Where the threads meet now
  62. 2026GPT-5.5built directly on GPT-5.4Chapter 14. Where the threads meet now
  63. 2026DeepSeek-V4was influenced by DeepSeek-V3Chapter 8. The weights get out
  64. 2026Claude Opus 4.8built directly on Claude Opus 4.7Chapter 14. Where the threads meet now
  65. 2026Claude Sonnet 5succeeded Claude Opus 4.8Chapter 14. Where the threads meet now
  66. 2026Fable 5 / Mythos 5was influenced by Project GlasswingChapter 14. Where the threads meet now
  67. 2026GLM-5.2built directly on GLM-4.5Chapter 8. The weights get out
  68. 2026GPT-5.6 Solbuilt directly on GPT-5.5Chapter 14. Where the threads meet now
  69. 2026Kimi K3built directly on Kimi K2Chapter 8. The weights get out
  70. 2026Claude Opus 5succeeded Claude Opus 4.8Chapter 14. Where the threads meet now
  71. 2026Qwen3.8built directly on Qwen3.5Chapter 8. The weights get out
  72. 2026Claude Opus 5.5built directly on Claude Opus 5Chapter 14. Where the threads meet now

Direct links and their sources

Where it came from

Influenced by

What it led to

Built on directly by

Influenced

Developed alongside

Terms