NLP Lineage

22 April 2024 · Scale and data

Phi-3

Phi-3 Technical Report

M. Abdin, et al.

Date
22 April 2024
What the date marks
arXiv submission
Kind
Model
Organisation
Microsoft
Access
MIT
Threads
In the guided tour
Chapter 8. The weights get out
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

Small models were assumed to be weak, and useful mainly for cheap tasks.

The problem

Phones and laptops cannot run a 70B model, but their users want capable assistance.

What it introduced

A 3.8B model trained on heavily filtered and synthetic data, performing near models ten times its size.

Why it matters

It pushed the data-quality thesis to its edge and made on-device assistants credible.

What modern models inherit

On-device capability through data curation rather than scale.

What came next

The phi series continued; every lab now ships a small model tuned on synthetic data.

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

Phi-3 has 36 documented ancestors across 8 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 15 turning points among them.

Phi-3 has 36 documented ancestors across 8 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

Phi-3 has 39 recorded ancestors, including indirect ones, across 8 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 15 turning points among them.

Phi-3 has 39 recorded ancestors, including indirect ones, across 8 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2024Phi-3This milestone.
  2. 2023phi-1Phi-3 built directly on thisChapter 8. The weights get out
  3. 2022GPT-3.5 Turning pointphi-1 was influenced by thisChapter 7. The assistant goes public
  4. 2022InstructGPT Turning pointGPT-3.5 built directly on thisChapter 6. Teaching models what people want
  5. 2021FLANInstructGPT was influenced by thisChapter 6. Teaching models what people want
  6. 2020RLHF for summarisation Turning pointInstructGPT built directly on thisChapter 6. Teaching models what people want
  7. 2020GPT-3 Turning pointGPT-3.5 built directly on thisChapter 5. Scale becomes a science
  8. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  9. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  10. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  11. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  12. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  13. 2019GPT-2 Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  14. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  15. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  16. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  17. 2018GPT-1GPT-2 built directly on thisChapter 4. Pretrain once, use everywhere
  18. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  19. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  20. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  21. 2017RLHF originRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  22. 2017Transformer Turning pointGPT-2 built directly on this
    Sparse Transformer built directly on this
    Scaling laws built directly on this    Chapter 3. Attention, then only attention
  23. 2016Layer normGPT-2 built directly on thisChapter 3. Attention, then only attention
  24. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  25. 2015LM pretraining Turning pointGPT-1 was influenced by thisChapter 2. Words become vectors
  26. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  27. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  28. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  29. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  30. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  31. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  32. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  33. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  34. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  35. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  36. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  37. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2024Phi-3This milestone.
  2. 2023phi-1Phi-3 built directly on thisChapter 8. The weights get out
  3. 2022Self-Instructphi-1 was influenced by this (indirect influence)Chapter 6. Teaching models what people want
  4. 2022GPT-3.5 Turning pointphi-1 was influenced by thisChapter 7. The assistant goes public
  5. 2022InstructGPT Turning pointGPT-3.5 built directly on thisChapter 6. Teaching models what people want
  6. 2021FLANSelf-Instruct was influenced by thisChapter 6. Teaching models what people want
  7. 2020RLHF for summarisation Turning pointInstructGPT built directly on thisChapter 6. Teaching models what people want
  8. 2020GPT-3 Turning pointGPT-3.5 built directly on this
    Self-Instruct built directly on this    Chapter 5. Scale becomes a science
  9. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  10. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  11. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  12. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  13. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  14. 2019GPT-2 Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  15. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  16. 2019Transformer-XLSparse Transformer was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  17. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  18. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  19. 2018GPT-1GPT-2 built directly on thisChapter 4. Pretrain once, use everywhere
  20. 2018ELMoBERT was influenced by this
    GPT-1 was influenced by this (indirect influence)    Chapter 4. Pretrain once, use everywhere
  21. 2018ULMFiTBERT was influenced by this
    GPT-1 was influenced by this (indirect influence)    Chapter 4. Pretrain once, use everywhere
  22. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  23. 2017Transformer Turning pointGPT-2 built directly on this
    Sparse Transformer built directly on this
    T5 built directly on this    Chapter 3. Attention, then only attention
  24. 2017RLHF originRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  25. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  26. 2016Layer normGPT-2 built directly on thisChapter 3. Attention, then only attention
  27. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  28. 2015LM pretraining Turning pointBERT was influenced by this
    GPT-1 was influenced by this    Chapter 2. Words become vectors
  29. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  30. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  31. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  32. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  33. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  34. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  35. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  36. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  37. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  38. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  39. 1997LSTM Turning pointSeq2seq built directly on this
    ELMo built directly on this    Chapter 1. Language as probability
  40. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

No documented descendants are recorded for Phi-3 without weak links.

No documented descendants are recorded for Phi-3 without weak links.

No recorded descendants are recorded for Phi-3.

No recorded descendants are recorded for Phi-3.

Direct links and their sources

Where it came from

Built directly on

What it led to

None recorded in the sources used.