NLP Lineage

5 January 2024 · Scale and data

DeepSeek LLM

DeepSeek LLM: Scaling Open-Source Language Models with Longtermism

DeepSeek-AI

Date
5 January 2024
What the date marks
arXiv submission
Kind
Research paper
Organisation
DeepSeek
Access
MIT code, DeepSeek model licence
Threads
In the guided tour
Chapter 8. The weights get out
Learning phases
2. How LLMs are trained

A milestone: 1 later milestone descends directly from it, and its descendants reach 8 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

Scaling laws were published by Western labs and taken on trust by everyone else.

The problem

The published coefficients did not transfer cleanly to other data and hyperparameter regimes.

What it introduced

Their own scaling study on their own data, used to size 7B and 67B open models.

Why it matters

The opening move of the lab that would later ship R1 and V3, and a reminder that scaling laws are data-dependent.

What modern models inherit

Scaling laws as something you measure for your own setup.

What came next

DeepSeekMoE, DeepSeekMath and the V2 and V3 architectures.

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

DeepSeek LLM has 42 documented ancestors across 9 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 19 turning points among them.

DeepSeek LLM has 42 documented ancestors across 9 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

DeepSeek LLM has 43 recorded ancestors, including indirect ones, across 9 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 19 turning points among them.

DeepSeek LLM has 43 recorded ancestors, including indirect ones, across 9 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2024DeepSeek LLMThis milestone.
  2. 2023Llama 2 Turning pointDeepSeek LLM was influenced by thisChapter 8. The weights get out
  3. 2023Grouped-query attention Turning pointLlama 2 built directly on thisChapter 9. Making it run anywhere
  4. 2023LLaMA Turning pointLlama 2 built directly on thisChapter 8. The weights get out
  5. 2022Chinchilla Turning pointDeepSeek LLM was influenced by thisChapter 5. Scale becomes a science
  6. 2022InstructGPT Turning pointLlama 2 was influenced by thisChapter 6. Teaching models what people want
  7. 2021GopherChinchilla built directly on thisChapter 5. Scale becomes a science
  8. 2021FLANInstructGPT was influenced by thisChapter 6. Teaching models what people want
  9. 2021RoPE Turning pointLLaMA built directly on thisChapter 10. Reading more than you remember
  10. 2020RLHF for summarisation Turning pointInstructGPT built directly on thisChapter 6. Teaching models what people want
  11. 2020GPT-3 Turning pointInstructGPT built directly on thisChapter 5. Scale becomes a science
  12. 2020Scaling laws Turning pointDeepSeek LLM built directly on thisChapter 5. Scale becomes a science
  13. 2019Multi-query attentionGrouped-query attention built directly on thisChapter 9. Making it run anywhere
  14. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  15. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  16. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  17. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  18. 2019GPT-2 Turning pointScaling laws was influenced by thisChapter 4. Pretrain once, use everywhere
  19. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  20. 2019Transformer-XLRoPE was influenced by thisChapter 4. Pretrain once, use everywhere
  21. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  22. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  23. 2018GPT-1GPT-2 built directly on thisChapter 4. Pretrain once, use everywhere
  24. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  25. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  26. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  27. 2017Transformer Turning pointScaling laws built directly on thisChapter 3. Attention, then only attention
  28. 2017RLHF originRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  29. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  30. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  31. 2015LM pretraining Turning pointGPT-1 was influenced by thisChapter 2. Words become vectors
  32. 2015BPE tokenizationGPT-2 built directly on this
    LLaMA built directly on this    Chapter 2. Words become vectors
  33. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  34. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  35. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  36. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  37. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  38. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  39. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  40. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  41. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  42. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  43. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2024DeepSeek LLMThis milestone.
  2. 2023Llama 2 Turning pointDeepSeek LLM was influenced by thisChapter 8. The weights get out
  3. 2023Grouped-query attention Turning pointLlama 2 built directly on thisChapter 9. Making it run anywhere
  4. 2023LLaMA Turning pointLlama 2 built directly on thisChapter 8. The weights get out
  5. 2022Chinchilla Turning pointDeepSeek LLM was influenced by thisChapter 5. Scale becomes a science
  6. 2022InstructGPT Turning pointLlama 2 was influenced by thisChapter 6. Teaching models what people want
  7. 2021GopherChinchilla built directly on thisChapter 5. Scale becomes a science
  8. 2021FLANInstructGPT was influenced by thisChapter 6. Teaching models what people want
  9. 2021RoPE Turning pointLLaMA built directly on thisChapter 10. Reading more than you remember
  10. 2020RLHF for summarisation Turning pointInstructGPT built directly on thisChapter 6. Teaching models what people want
  11. 2020GPT-3 Turning pointInstructGPT built directly on thisChapter 5. Scale becomes a science
  12. 2020Scaling laws Turning pointDeepSeek LLM built directly on thisChapter 5. Scale becomes a science
  13. 2019Multi-query attentionGrouped-query attention built directly on thisChapter 9. Making it run anywhere
  14. 2019T5 Turning pointFLAN was influenced by thisChapter 4. Pretrain once, use everywhere
  15. 2019C4 Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  16. 2019RLHF for LMsRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  17. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  18. 2019GPT-2 Turning pointScaling laws was influenced by thisChapter 4. Pretrain once, use everywhere
  19. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  20. 2019Transformer-XLRoPE was influenced by thisChapter 4. Pretrain once, use everywhere
  21. 2018BERT Turning pointT5 was influenced by thisChapter 4. Pretrain once, use everywhere
  22. 2018SentencePieceT5 built directly on thisChapter 4. Pretrain once, use everywhere
  23. 2018GPT-1GPT-2 built directly on thisChapter 4. Pretrain once, use everywhere
  24. 2018ELMoGPT-1 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  25. 2018ULMFiTGPT-1 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  26. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  27. 2017Transformer Turning pointScaling laws built directly on thisChapter 3. Attention, then only attention
  28. 2017RLHF originRLHF for summarisation built directly on thisChapter 6. Teaching models what people want
  29. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  30. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  31. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  32. 2015LM pretraining Turning pointGPT-1 was influenced by thisChapter 2. Words become vectors
  33. 2015BPE tokenizationGPT-2 built directly on this
    LLaMA built directly on this    Chapter 2. Words become vectors
  34. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  35. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  36. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  37. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  38. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  39. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  40. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  41. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  42. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  43. 1997LSTM Turning pointSeq2seq built directly on thisChapter 1. Language as probability
  44. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

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

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

DeepSeek LLM has 13 recorded descendants, including indirect ones, across 3 chapters, reaching forward to Qwen3.8 in 2026. Showing its direct descendants.

DeepSeek LLM has 13 recorded descendants, including indirect ones, across 3 chapters, reaching forward to Qwen3.8 in 2026. Showing all of them.

  1. 2024DeepSeek LLMThis milestone.
  2. 2024GRPObuilt directly on DeepSeek LLMChapter 12. Thinking longer
  3. 2024DeepSeek-V3built directly on GRPOChapter 8. The weights get out
  4. 2025DeepSeek-R1built directly on GRPOChapter 12. Thinking longer
  5. 2025Qwen3was influenced by DeepSeek-R1Chapter 8. The weights get out
  6. 2025Kimi K2was influenced by GRPOChapter 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. 2026DeepSeek-V4was influenced by DeepSeek-V3Chapter 8. The weights get out
  10. 2026Kimi K3built directly on Kimi K2Chapter 8. The weights get out
  11. 2026Qwen3.8built directly on Qwen3.5Chapter 8. The weights get out
  1. 2024DeepSeek LLMThis milestone.
  2. 2024GRPObuilt directly on DeepSeek LLMChapter 12. Thinking longer
  3. 2024DeepSeek-V3built directly on GRPOChapter 8. The weights get out
  4. 2025DeepSeek-R1built directly on GRPOChapter 12. Thinking longer
  5. 2025Qwen3was influenced by DeepSeek-R1Chapter 8. The weights get out
  6. 2025Kimi K2was influenced by GRPOChapter 8. The weights get out
  7. 2025gpt-osswas influenced by DeepSeek-R1 (indirect influence)Chapter 8. The weights get out
  8. 2025GLM-4.5was influenced by DeepSeek-R1 (indirect influence)Chapter 8. The weights get out
  9. 2025DeepSeek-V3.2built directly on DeepSeek-V3Chapter 10. Reading more than you remember
  10. 2026Qwen3.5built directly on Qwen3Chapter 8. The weights get out
  11. 2026DeepSeek-V4was influenced by DeepSeek-V3Chapter 8. The weights get out
  12. 2026GLM-5.2built directly on GLM-4.5Chapter 8. The weights get out
  13. 2026Kimi K3built directly on Kimi K2Chapter 8. The weights get out
  14. 2026Qwen3.8built directly on Qwen3.5Chapter 8. The weights get out

Direct links and their sources

Where it came from

Built directly on

Influenced by

What it led to

Built on directly by

Terms