NLP Lineage

25 May 2023 · Language modelling

Falcon

Falcon / RefinedWeb models

TII (Almazrouei et al.)

Date
25 May 2023
What the date marks
Model release
Kind
Model
Organisation
Technology Innovation Institute, UAE
Access
Apache 2.0
Other dates
Apache 2.0 from 2023-05-31; paper 2023-11-28
Threads
In the guided tour
Chapter 8. The weights get out
Learning phases
2. How LLMs are trained; 8. Training from scratch

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

The best open weights carried non-commercial licences.

The problem

Businesses could not build on LLaMA, and the alternatives were weaker.

What it introduced

A competitive 40B model trained on filtered web data, relicensed to Apache 2.0 within a week of release.

Why it matters

It briefly led the open leaderboards and proved permissive licensing at scale was possible before Llama 2 followed.

What modern models inherit

Permissive licensing as a competitive lever.

What came next

Llama 2's commercial licence, then Mistral's Apache releases.

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

Falcon has 2 documented ancestors across 2 chapters, reaching back to C4 in 2019. Showing all of them.

Falcon has 2 documented ancestors across 2 chapters, reaching back to C4 in 2019. Showing all of them.

Falcon has 23 recorded ancestors, including indirect ones, across 5 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 6 turning points among them.

Falcon has 23 recorded ancestors, including indirect ones, across 5 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2023FalconThis milestone.
  2. 2023RefinedWebFalcon built directly on thisChapter 8. The weights get out
  3. 2019C4 Turning pointRefinedWeb was influenced by thisChapter 4. Pretrain once, use everywhere
  1. 2023FalconThis milestone.
  2. 2023RefinedWebFalcon built directly on thisChapter 8. The weights get out
  3. 2019C4 Turning pointRefinedWeb was influenced by thisChapter 4. Pretrain once, use everywhere
  4. 2019WebTextC4 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  5. 2018GPT-1WebText was influenced by thisChapter 4. Pretrain once, use everywhere
  6. 2018ELMoGPT-1 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  7. 2018ULMFiTGPT-1 was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  8. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  9. 2017Transformer Turning pointGPT-1 built directly on thisChapter 3. Attention, then only attention
  10. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  11. 2016Layer normTransformer built directly on thisChapter 3. Attention, then only attention
  12. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  13. 2015LM pretraining Turning pointGPT-1 was influenced by thisChapter 2. Words become vectors
  14. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  15. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  16. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  17. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  18. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  19. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  20. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  21. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  22. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  23. 1997LSTM Turning pointELMo built directly on thisChapter 1. Language as probability
  24. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

No documented descendants are recorded for Falcon without weak links.

No documented descendants are recorded for Falcon without weak links.

No recorded descendants are recorded for Falcon.

No recorded descendants are recorded for Falcon.

Direct links and their sources

Where it came from

Built directly on

Developed alongside

What it led to

None recorded in the sources used.