From 043b7b157139de07f91366ced261ae17de883f9c Mon Sep 17 00:00:00 2001 From: Alajmah <55954113+Alajmah@users.noreply.github.com> Date: Tue, 29 Sep 2026 14:47:20 +0300 Subject: [PATCH 1/2] Add compressed 2017-2022 LLM foundations --- .github/workflows/validate.yml | 12 ++++++++++++ data/claims/bert-published-by-google.yaml | 9 +++++++++ data/claims/bert-release-of-bert.yaml | 9 +++++++++ data/claims/bert-uses-masked-language-modeling.yaml | 9 +++++++++ data/claims/bert-uses-transformer.yaml | 9 +++++++++ data/claims/chinchilla-published-by-deepmind.yaml | 9 +++++++++ data/claims/chinchilla-release-of-chinchilla.yaml | 9 +++++++++ .../chinchilla-uses-compute-optimal-training.yaml | 9 +++++++++ data/claims/chinchilla-uses-transformer.yaml | 9 +++++++++ data/claims/gpt-1-published-by-openai.yaml | 9 +++++++++ data/claims/gpt-1-release-of-gpt.yaml | 9 +++++++++ data/claims/gpt-1-uses-generative-pretraining.yaml | 9 +++++++++ data/claims/gpt-1-uses-transformer.yaml | 9 +++++++++ data/claims/gpt-2-published-by-openai.yaml | 9 +++++++++ data/claims/gpt-2-release-of-gpt.yaml | 9 +++++++++ data/claims/gpt-2-uses-generative-pretraining.yaml | 9 +++++++++ data/claims/gpt-2-uses-transformer.yaml | 9 +++++++++ data/claims/gpt-3-published-by-openai.yaml | 9 +++++++++ data/claims/gpt-3-release-of-gpt.yaml | 9 +++++++++ data/claims/gpt-3-uses-in-context-learning.yaml | 9 +++++++++ data/claims/gpt-3-uses-transformer.yaml | 9 +++++++++ data/claims/instructgpt-fine-tuned-from-gpt-3.yaml | 9 +++++++++ data/claims/instructgpt-published-by-openai.yaml | 9 +++++++++ data/claims/instructgpt-release-of-gpt.yaml | 9 +++++++++ data/claims/instructgpt-uses-rlhf.yaml | 9 +++++++++ data/claims/palm-published-by-google.yaml | 9 +++++++++ data/claims/palm-release-of-palm.yaml | 9 +++++++++ data/claims/palm-uses-transformer.yaml | 9 +++++++++ .../switch-transformer-published-by-google.yaml | 9 +++++++++ ...h-transformer-release-of-switch-transformer.yaml | 9 +++++++++ data/claims/switch-transformer-uses-sparse-moe.yaml | 9 +++++++++ .../claims/switch-transformer-uses-transformer.yaml | 9 +++++++++ data/claims/t5-published-by-google.yaml | 9 +++++++++ data/claims/t5-release-of-t5.yaml | 9 +++++++++ data/claims/t5-uses-text-to-text-transfer.yaml | 9 +++++++++ data/claims/t5-uses-transformer.yaml | 9 +++++++++ data/events/2018-06-11-gpt-1-announced.yaml | 8 ++++++++ data/events/2018-10-11-bert-paper.yaml | 8 ++++++++ data/events/2019-02-14-gpt-2-announced.yaml | 8 ++++++++ data/events/2019-10-16-rmsnorm-paper.yaml | 8 ++++++++ data/events/2019-10-23-t5-paper.yaml | 8 ++++++++ data/events/2020-01-23-scaling-laws-paper.yaml | 8 ++++++++ data/events/2020-02-12-swiglu-paper.yaml | 8 ++++++++ data/events/2020-05-28-gpt-3-paper.yaml | 8 ++++++++ .../events/2021-01-11-switch-transformer-paper.yaml | 8 ++++++++ data/events/2021-09-03-flan-paper.yaml | 8 ++++++++ data/events/2022-01-28-chain-of-thought-paper.yaml | 8 ++++++++ data/events/2022-03-04-instructgpt-paper.yaml | 8 ++++++++ data/events/2022-03-29-chinchilla-paper.yaml | 8 ++++++++ data/events/2022-04-05-palm-paper.yaml | 8 ++++++++ data/events/2022-05-27-flashattention-paper.yaml | 8 ++++++++ data/families/bert.yaml | 4 ++++ data/families/chinchilla.yaml | 4 ++++ data/families/switch-transformer.yaml | 4 ++++ data/families/t5.yaml | 4 ++++ data/organizations/deepmind.yaml | 4 ++++ data/organizations/google.yaml | 4 ++++ data/relationships/bert-published-by-google.yaml | 7 +++++++ data/relationships/bert-release-of-bert.yaml | 7 +++++++ .../bert-uses-masked-language-modeling.yaml | 7 +++++++ data/relationships/bert-uses-transformer.yaml | 7 +++++++ .../chinchilla-published-by-deepmind.yaml | 7 +++++++ .../chinchilla-release-of-chinchilla.yaml | 7 +++++++ .../chinchilla-uses-compute-optimal-training.yaml | 7 +++++++ data/relationships/chinchilla-uses-transformer.yaml | 7 +++++++ data/relationships/gpt-1-published-by-openai.yaml | 7 +++++++ data/relationships/gpt-1-release-of-gpt.yaml | 7 +++++++ .../gpt-1-uses-generative-pretraining.yaml | 7 +++++++ data/relationships/gpt-1-uses-transformer.yaml | 7 +++++++ data/relationships/gpt-2-published-by-openai.yaml | 7 +++++++ data/relationships/gpt-2-release-of-gpt.yaml | 7 +++++++ .../gpt-2-uses-generative-pretraining.yaml | 7 +++++++ data/relationships/gpt-2-uses-transformer.yaml | 7 +++++++ data/relationships/gpt-3-published-by-openai.yaml | 7 +++++++ data/relationships/gpt-3-release-of-gpt.yaml | 7 +++++++ .../gpt-3-uses-in-context-learning.yaml | 7 +++++++ data/relationships/gpt-3-uses-transformer.yaml | 7 +++++++ .../instructgpt-fine-tuned-from-gpt-3.yaml | 7 +++++++ .../instructgpt-published-by-openai.yaml | 7 +++++++ data/relationships/instructgpt-release-of-gpt.yaml | 7 +++++++ data/relationships/instructgpt-uses-rlhf.yaml | 7 +++++++ data/relationships/palm-published-by-google.yaml | 7 +++++++ data/relationships/palm-release-of-palm.yaml | 7 +++++++ data/relationships/palm-uses-transformer.yaml | 7 +++++++ .../switch-transformer-published-by-google.yaml | 7 +++++++ ...h-transformer-release-of-switch-transformer.yaml | 7 +++++++ .../switch-transformer-uses-sparse-moe.yaml | 7 +++++++ .../switch-transformer-uses-transformer.yaml | 7 +++++++ data/relationships/t5-published-by-google.yaml | 7 +++++++ data/relationships/t5-release-of-t5.yaml | 7 +++++++ .../t5-uses-text-to-text-transfer.yaml | 7 +++++++ data/relationships/t5-uses-transformer.yaml | 7 +++++++ data/releases/bert.yaml | 4 ++++ data/releases/chinchilla.yaml | 4 ++++ data/releases/gpt-1.yaml | 4 ++++ data/releases/gpt-2.yaml | 4 ++++ data/releases/gpt-3.yaml | 4 ++++ data/releases/instructgpt.yaml | 4 ++++ data/releases/palm.yaml | 4 ++++ data/releases/switch-transformer.yaml | 4 ++++ data/releases/t5.yaml | 4 ++++ data/sources/bert-paper.yaml | 8 ++++++++ data/sources/chinchilla-paper.yaml | 8 ++++++++ data/sources/cot-paper.yaml | 8 ++++++++ data/sources/flan-paper.yaml | 8 ++++++++ data/sources/flashattention-paper.yaml | 7 +++++++ data/sources/gpt1-openai-announcement.yaml | 7 +++++++ data/sources/gpt2-openai-announcement.yaml | 7 +++++++ data/sources/gpt3-paper.yaml | 8 ++++++++ data/sources/instructgpt-paper.yaml | 8 ++++++++ data/sources/palm-paper.yaml | 8 ++++++++ data/sources/rmsnorm-paper.yaml | 7 +++++++ data/sources/scaling-laws-paper.yaml | 8 ++++++++ data/sources/swiglu-paper.yaml | 7 +++++++ data/sources/switch-transformer-paper.yaml | 8 ++++++++ data/sources/t5-paper.yaml | 8 ++++++++ data/techniques/chain-of-thought-prompting.yaml | 7 +++++++ data/techniques/compute-optimal-training.yaml | 5 +++++ data/techniques/flashattention.yaml | 5 +++++ data/techniques/generative-pretraining.yaml | 5 +++++ data/techniques/in-context-learning.yaml | 5 +++++ data/techniques/instruction-tuning.yaml | 5 +++++ data/techniques/masked-language-modeling.yaml | 5 +++++ data/techniques/neural-scaling-laws.yaml | 5 +++++ .../reinforcement-learning-from-human-feedback.yaml | 6 ++++++ data/techniques/rmsnorm.yaml | 6 ++++++ data/techniques/swiglu.yaml | 5 +++++ data/techniques/text-to-text-transfer.yaml | 5 +++++ narratives/foundations/README.md | 13 +++++++++++++ 129 files changed, 944 insertions(+) create mode 100644 data/claims/bert-published-by-google.yaml create mode 100644 data/claims/bert-release-of-bert.yaml create mode 100644 data/claims/bert-uses-masked-language-modeling.yaml create mode 100644 data/claims/bert-uses-transformer.yaml create mode 100644 data/claims/chinchilla-published-by-deepmind.yaml create mode 100644 data/claims/chinchilla-release-of-chinchilla.yaml create mode 100644 data/claims/chinchilla-uses-compute-optimal-training.yaml create mode 100644 data/claims/chinchilla-uses-transformer.yaml create mode 100644 data/claims/gpt-1-published-by-openai.yaml create mode 100644 data/claims/gpt-1-release-of-gpt.yaml create mode 100644 data/claims/gpt-1-uses-generative-pretraining.yaml create mode 100644 data/claims/gpt-1-uses-transformer.yaml create mode 100644 data/claims/gpt-2-published-by-openai.yaml create mode 100644 data/claims/gpt-2-release-of-gpt.yaml create mode 100644 data/claims/gpt-2-uses-generative-pretraining.yaml create mode 100644 data/claims/gpt-2-uses-transformer.yaml create mode 100644 data/claims/gpt-3-published-by-openai.yaml create mode 100644 data/claims/gpt-3-release-of-gpt.yaml create mode 100644 data/claims/gpt-3-uses-in-context-learning.yaml create mode 100644 data/claims/gpt-3-uses-transformer.yaml create mode 100644 data/claims/instructgpt-fine-tuned-from-gpt-3.yaml create mode 100644 data/claims/instructgpt-published-by-openai.yaml create mode 100644 data/claims/instructgpt-release-of-gpt.yaml create mode 100644 data/claims/instructgpt-uses-rlhf.yaml create mode 100644 data/claims/palm-published-by-google.yaml create mode 100644 data/claims/palm-release-of-palm.yaml create mode 100644 data/claims/palm-uses-transformer.yaml create mode 100644 data/claims/switch-transformer-published-by-google.yaml create mode 100644 data/claims/switch-transformer-release-of-switch-transformer.yaml create mode 100644 data/claims/switch-transformer-uses-sparse-moe.yaml create mode 100644 data/claims/switch-transformer-uses-transformer.yaml create mode 100644 data/claims/t5-published-by-google.yaml create mode 100644 data/claims/t5-release-of-t5.yaml create mode 100644 data/claims/t5-uses-text-to-text-transfer.yaml create mode 100644 data/claims/t5-uses-transformer.yaml create mode 100644 data/events/2018-06-11-gpt-1-announced.yaml create mode 100644 data/events/2018-10-11-bert-paper.yaml create mode 100644 data/events/2019-02-14-gpt-2-announced.yaml create mode 100644 data/events/2019-10-16-rmsnorm-paper.yaml create mode 100644 data/events/2019-10-23-t5-paper.yaml create mode 100644 data/events/2020-01-23-scaling-laws-paper.yaml create mode 100644 data/events/2020-02-12-swiglu-paper.yaml create mode 100644 data/events/2020-05-28-gpt-3-paper.yaml create mode 100644 data/events/2021-01-11-switch-transformer-paper.yaml create mode 100644 data/events/2021-09-03-flan-paper.yaml create mode 100644 data/events/2022-01-28-chain-of-thought-paper.yaml create mode 100644 data/events/2022-03-04-instructgpt-paper.yaml create mode 100644 data/events/2022-03-29-chinchilla-paper.yaml create mode 100644 data/events/2022-04-05-palm-paper.yaml create mode 100644 data/events/2022-05-27-flashattention-paper.yaml create mode 100644 data/families/bert.yaml create mode 100644 data/families/chinchilla.yaml create mode 100644 data/families/switch-transformer.yaml create mode 100644 data/families/t5.yaml create mode 100644 data/organizations/deepmind.yaml create mode 100644 data/organizations/google.yaml create mode 100644 data/relationships/bert-published-by-google.yaml create mode 100644 data/relationships/bert-release-of-bert.yaml create mode 100644 data/relationships/bert-uses-masked-language-modeling.yaml create mode 100644 data/relationships/bert-uses-transformer.yaml create mode 100644 data/relationships/chinchilla-published-by-deepmind.yaml create mode 100644 data/relationships/chinchilla-release-of-chinchilla.yaml create mode 100644 data/relationships/chinchilla-uses-compute-optimal-training.yaml create mode 100644 data/relationships/chinchilla-uses-transformer.yaml create mode 100644 data/relationships/gpt-1-published-by-openai.yaml create mode 100644 data/relationships/gpt-1-release-of-gpt.yaml create mode 100644 data/relationships/gpt-1-uses-generative-pretraining.yaml create mode 100644 data/relationships/gpt-1-uses-transformer.yaml create mode 100644 data/relationships/gpt-2-published-by-openai.yaml create mode 100644 data/relationships/gpt-2-release-of-gpt.yaml create mode 100644 data/relationships/gpt-2-uses-generative-pretraining.yaml create mode 100644 data/relationships/gpt-2-uses-transformer.yaml create mode 100644 data/relationships/gpt-3-published-by-openai.yaml create mode 100644 data/relationships/gpt-3-release-of-gpt.yaml create mode 100644 data/relationships/gpt-3-uses-in-context-learning.yaml create mode 100644 data/relationships/gpt-3-uses-transformer.yaml create mode 100644 data/relationships/instructgpt-fine-tuned-from-gpt-3.yaml create mode 100644 data/relationships/instructgpt-published-by-openai.yaml create mode 100644 data/relationships/instructgpt-release-of-gpt.yaml create mode 100644 data/relationships/instructgpt-uses-rlhf.yaml create mode 100644 data/relationships/palm-published-by-google.yaml create mode 100644 data/relationships/palm-release-of-palm.yaml create mode 100644 data/relationships/palm-uses-transformer.yaml create mode 100644 data/relationships/switch-transformer-published-by-google.yaml create mode 100644 data/relationships/switch-transformer-release-of-switch-transformer.yaml create mode 100644 data/relationships/switch-transformer-uses-sparse-moe.yaml create mode 100644 data/relationships/switch-transformer-uses-transformer.yaml create mode 100644 data/relationships/t5-published-by-google.yaml create mode 100644 data/relationships/t5-release-of-t5.yaml create mode 100644 data/relationships/t5-uses-text-to-text-transfer.yaml create mode 100644 data/relationships/t5-uses-transformer.yaml create mode 100644 data/releases/bert.yaml create mode 100644 data/releases/chinchilla.yaml create mode 100644 data/releases/gpt-1.yaml create mode 100644 data/releases/gpt-2.yaml create mode 100644 data/releases/gpt-3.yaml create mode 100644 data/releases/instructgpt.yaml create mode 100644 data/releases/palm.yaml create mode 100644 data/releases/switch-transformer.yaml create mode 100644 data/releases/t5.yaml create mode 100644 data/sources/bert-paper.yaml create mode 100644 data/sources/chinchilla-paper.yaml create mode 100644 data/sources/cot-paper.yaml create mode 100644 data/sources/flan-paper.yaml create mode 100644 data/sources/flashattention-paper.yaml create mode 100644 data/sources/gpt1-openai-announcement.yaml create mode 100644 data/sources/gpt2-openai-announcement.yaml create mode 100644 data/sources/gpt3-paper.yaml create mode 100644 data/sources/instructgpt-paper.yaml create mode 100644 data/sources/palm-paper.yaml create mode 100644 data/sources/rmsnorm-paper.yaml create mode 100644 data/sources/scaling-laws-paper.yaml create mode 100644 data/sources/swiglu-paper.yaml create mode 100644 data/sources/switch-transformer-paper.yaml create mode 100644 data/sources/t5-paper.yaml create mode 100644 data/techniques/chain-of-thought-prompting.yaml create mode 100644 data/techniques/compute-optimal-training.yaml create mode 100644 data/techniques/flashattention.yaml create mode 100644 data/techniques/generative-pretraining.yaml create mode 100644 data/techniques/in-context-learning.yaml create mode 100644 data/techniques/instruction-tuning.yaml create mode 100644 data/techniques/masked-language-modeling.yaml create mode 100644 data/techniques/neural-scaling-laws.yaml create mode 100644 data/techniques/reinforcement-learning-from-human-feedback.yaml create mode 100644 data/techniques/rmsnorm.yaml create mode 100644 data/techniques/swiglu.yaml create mode 100644 data/techniques/text-to-text-transfer.yaml create mode 100644 narratives/foundations/README.md diff --git a/.github/workflows/validate.yml b/.github/workflows/validate.yml index 40f18aa..e44cc46 100644 --- a/.github/workflows/validate.yml +++ b/.github/workflows/validate.yml @@ -26,6 +26,18 @@ jobs: run: python tools/validate.py - name: Build chronological timeline run: python tools/build_timeline.py --format json > /tmp/modelmap-timeline.json + - name: Build 2017 year view + run: python tools/build_year.py 2017 --format json > /tmp/modelmap-2017.json + - name: Build 2018 year view + run: python tools/build_year.py 2018 --format json > /tmp/modelmap-2018.json + - name: Build 2019 year view + run: python tools/build_year.py 2019 --format json > /tmp/modelmap-2019.json + - name: Build 2020 year view + run: python tools/build_year.py 2020 --format json > /tmp/modelmap-2020.json + - name: Build 2021 year view + run: python tools/build_year.py 2021 --format json > /tmp/modelmap-2021.json + - name: Build 2022 year view + run: python tools/build_year.py 2022 --format json > /tmp/modelmap-2022.json - name: Build 2023 year view run: python tools/build_year.py 2023 --format json > /tmp/modelmap-2023.json - name: Build 2024 year view diff --git a/data/claims/bert-published-by-google.yaml b/data/claims/bert-published-by-google.yaml new file mode 100644 index 0000000..be82721 --- /dev/null +++ b/data/claims/bert-published-by-google.yaml @@ -0,0 +1,9 @@ +id: claim:bert-published-by-google +type: claim +subject: release:bert +predicate: published_by +object: org:google +status: confirmed_primary +evidence: + - source: source:bert-paper + locator: author affiliations — Google AI Language diff --git a/data/claims/bert-release-of-bert.yaml b/data/claims/bert-release-of-bert.yaml new file mode 100644 index 0000000..3c365a8 --- /dev/null +++ b/data/claims/bert-release-of-bert.yaml @@ -0,0 +1,9 @@ +id: claim:bert-release-of-bert +type: claim +subject: release:bert +predicate: release_of +object: family:bert +status: confirmed_primary +evidence: + - source: source:bert-paper + locator: title and abstract introduce BERT diff --git a/data/claims/bert-uses-masked-language-modeling.yaml b/data/claims/bert-uses-masked-language-modeling.yaml new file mode 100644 index 0000000..d3d79a4 --- /dev/null +++ b/data/claims/bert-uses-masked-language-modeling.yaml @@ -0,0 +1,9 @@ +id: claim:bert-uses-masked-language-modeling +type: claim +subject: release:bert +predicate: uses_technique +object: technique:masked-language-modeling +status: confirmed_primary +evidence: + - source: source:bert-paper + locator: pre-training tasks — masked language model diff --git a/data/claims/bert-uses-transformer.yaml b/data/claims/bert-uses-transformer.yaml new file mode 100644 index 0000000..b44acf8 --- /dev/null +++ b/data/claims/bert-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:bert-uses-transformer +type: claim +subject: release:bert +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:bert-paper + locator: title and abstract — bidirectional Transformers diff --git a/data/claims/chinchilla-published-by-deepmind.yaml b/data/claims/chinchilla-published-by-deepmind.yaml new file mode 100644 index 0000000..72ffe7f --- /dev/null +++ b/data/claims/chinchilla-published-by-deepmind.yaml @@ -0,0 +1,9 @@ +id: claim:chinchilla-published-by-deepmind +type: claim +subject: release:chinchilla +predicate: published_by +object: org:deepmind +status: confirmed_primary +evidence: + - source: source:chinchilla-paper + locator: DeepMind authors and publication diff --git a/data/claims/chinchilla-release-of-chinchilla.yaml b/data/claims/chinchilla-release-of-chinchilla.yaml new file mode 100644 index 0000000..ae30afd --- /dev/null +++ b/data/claims/chinchilla-release-of-chinchilla.yaml @@ -0,0 +1,9 @@ +id: claim:chinchilla-release-of-chinchilla +type: claim +subject: release:chinchilla +predicate: release_of +object: family:chinchilla +status: confirmed_primary +evidence: + - source: source:chinchilla-paper + locator: paper introduces the 70B-parameter Chinchilla model diff --git a/data/claims/chinchilla-uses-compute-optimal-training.yaml b/data/claims/chinchilla-uses-compute-optimal-training.yaml new file mode 100644 index 0000000..e744e5d --- /dev/null +++ b/data/claims/chinchilla-uses-compute-optimal-training.yaml @@ -0,0 +1,9 @@ +id: claim:chinchilla-uses-compute-optimal-training +type: claim +subject: release:chinchilla +predicate: uses_technique +object: technique:compute-optimal-training +status: confirmed_primary +evidence: + - source: source:chinchilla-paper + locator: compute-optimal model/data scaling analysis and Chinchilla training recipe diff --git a/data/claims/chinchilla-uses-transformer.yaml b/data/claims/chinchilla-uses-transformer.yaml new file mode 100644 index 0000000..726475c --- /dev/null +++ b/data/claims/chinchilla-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:chinchilla-uses-transformer +type: claim +subject: release:chinchilla +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:chinchilla-paper + locator: autoregressive Transformer language-model setup diff --git a/data/claims/gpt-1-published-by-openai.yaml b/data/claims/gpt-1-published-by-openai.yaml new file mode 100644 index 0000000..d766b2a --- /dev/null +++ b/data/claims/gpt-1-published-by-openai.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-1-published-by-openai +type: claim +subject: release:gpt-1 +predicate: published_by +object: org:openai +status: confirmed_primary +evidence: + - source: source:gpt1-openai-announcement + locator: OpenAI milestone announcement diff --git a/data/claims/gpt-1-release-of-gpt.yaml b/data/claims/gpt-1-release-of-gpt.yaml new file mode 100644 index 0000000..c334a0f --- /dev/null +++ b/data/claims/gpt-1-release-of-gpt.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-1-release-of-gpt +type: claim +subject: release:gpt-1 +predicate: release_of +object: family:gpt +status: confirmed_primary +evidence: + - source: source:gpt1-openai-announcement + locator: OpenAI presents its generative pre-training Transformer approach diff --git a/data/claims/gpt-1-uses-generative-pretraining.yaml b/data/claims/gpt-1-uses-generative-pretraining.yaml new file mode 100644 index 0000000..2a98c4e --- /dev/null +++ b/data/claims/gpt-1-uses-generative-pretraining.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-1-uses-generative-pretraining +type: claim +subject: release:gpt-1 +predicate: uses_technique +object: technique:generative-pretraining +status: confirmed_primary +evidence: + - source: source:gpt1-openai-announcement + locator: combines transformers with unsupervised pre-training before supervised adaptation diff --git a/data/claims/gpt-1-uses-transformer.yaml b/data/claims/gpt-1-uses-transformer.yaml new file mode 100644 index 0000000..8a1fbb5 --- /dev/null +++ b/data/claims/gpt-1-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-1-uses-transformer +type: claim +subject: release:gpt-1 +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:gpt1-openai-announcement + locator: approach combines transformers and unsupervised pre-training diff --git a/data/claims/gpt-2-published-by-openai.yaml b/data/claims/gpt-2-published-by-openai.yaml new file mode 100644 index 0000000..cdbd336 --- /dev/null +++ b/data/claims/gpt-2-published-by-openai.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-2-published-by-openai +type: claim +subject: release:gpt-2 +predicate: published_by +object: org:openai +status: confirmed_primary +evidence: + - source: source:gpt2-openai-announcement + locator: OpenAI milestone announcement diff --git a/data/claims/gpt-2-release-of-gpt.yaml b/data/claims/gpt-2-release-of-gpt.yaml new file mode 100644 index 0000000..5f6ce9b --- /dev/null +++ b/data/claims/gpt-2-release-of-gpt.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-2-release-of-gpt +type: claim +subject: release:gpt-2 +predicate: release_of +object: family:gpt +status: confirmed_primary +evidence: + - source: source:gpt2-openai-announcement + locator: GPT-2 is described as a successor to GPT diff --git a/data/claims/gpt-2-uses-generative-pretraining.yaml b/data/claims/gpt-2-uses-generative-pretraining.yaml new file mode 100644 index 0000000..df08afe --- /dev/null +++ b/data/claims/gpt-2-uses-generative-pretraining.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-2-uses-generative-pretraining +type: claim +subject: release:gpt-2 +predicate: uses_technique +object: technique:generative-pretraining +status: confirmed_primary +evidence: + - source: source:gpt2-openai-announcement + locator: trained to predict the next word from previous text without task-specific training diff --git a/data/claims/gpt-2-uses-transformer.yaml b/data/claims/gpt-2-uses-transformer.yaml new file mode 100644 index 0000000..17423bc --- /dev/null +++ b/data/claims/gpt-2-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-2-uses-transformer +type: claim +subject: release:gpt-2 +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:gpt2-openai-announcement + locator: GPT-2 is a large transformer-based language model diff --git a/data/claims/gpt-3-published-by-openai.yaml b/data/claims/gpt-3-published-by-openai.yaml new file mode 100644 index 0000000..824d7e5 --- /dev/null +++ b/data/claims/gpt-3-published-by-openai.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-3-published-by-openai +type: claim +subject: release:gpt-3 +predicate: published_by +object: org:openai +status: confirmed_primary +evidence: + - source: source:gpt3-paper + locator: OpenAI author affiliations diff --git a/data/claims/gpt-3-release-of-gpt.yaml b/data/claims/gpt-3-release-of-gpt.yaml new file mode 100644 index 0000000..e67ab57 --- /dev/null +++ b/data/claims/gpt-3-release-of-gpt.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-3-release-of-gpt +type: claim +subject: release:gpt-3 +predicate: release_of +object: family:gpt +status: confirmed_primary +evidence: + - source: source:gpt3-paper + locator: paper presents GPT-3 as a 175B autoregressive language model diff --git a/data/claims/gpt-3-uses-in-context-learning.yaml b/data/claims/gpt-3-uses-in-context-learning.yaml new file mode 100644 index 0000000..2d13f33 --- /dev/null +++ b/data/claims/gpt-3-uses-in-context-learning.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-3-uses-in-context-learning +type: claim +subject: release:gpt-3 +predicate: uses_technique +object: technique:in-context-learning +status: confirmed_primary +evidence: + - source: source:gpt3-paper + locator: zero-shot, one-shot, and few-shot task demonstrations in model context without gradient updates diff --git a/data/claims/gpt-3-uses-transformer.yaml b/data/claims/gpt-3-uses-transformer.yaml new file mode 100644 index 0000000..a97e4e0 --- /dev/null +++ b/data/claims/gpt-3-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-3-uses-transformer +type: claim +subject: release:gpt-3 +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:gpt3-paper + locator: autoregressive Transformer language-model architecture diff --git a/data/claims/instructgpt-fine-tuned-from-gpt-3.yaml b/data/claims/instructgpt-fine-tuned-from-gpt-3.yaml new file mode 100644 index 0000000..1a73c5b --- /dev/null +++ b/data/claims/instructgpt-fine-tuned-from-gpt-3.yaml @@ -0,0 +1,9 @@ +id: claim:instructgpt-fine-tuned-from-gpt-3 +type: claim +subject: release:instructgpt +predicate: fine_tuned_from +object: release:gpt-3 +status: confirmed_primary +evidence: + - source: source:instructgpt-paper + locator: fine-tuning GPT-3 to follow a broad class of written instructions diff --git a/data/claims/instructgpt-published-by-openai.yaml b/data/claims/instructgpt-published-by-openai.yaml new file mode 100644 index 0000000..1eb763e --- /dev/null +++ b/data/claims/instructgpt-published-by-openai.yaml @@ -0,0 +1,9 @@ +id: claim:instructgpt-published-by-openai +type: claim +subject: release:instructgpt +predicate: published_by +object: org:openai +status: confirmed_primary +evidence: + - source: source:instructgpt-paper + locator: OpenAI authors and publication diff --git a/data/claims/instructgpt-release-of-gpt.yaml b/data/claims/instructgpt-release-of-gpt.yaml new file mode 100644 index 0000000..d86cc6d --- /dev/null +++ b/data/claims/instructgpt-release-of-gpt.yaml @@ -0,0 +1,9 @@ +id: claim:instructgpt-release-of-gpt +type: claim +subject: release:instructgpt +predicate: release_of +object: family:gpt +status: confirmed_primary +evidence: + - source: source:instructgpt-paper + locator: InstructGPT models are fine-tuned from GPT-3 diff --git a/data/claims/instructgpt-uses-rlhf.yaml b/data/claims/instructgpt-uses-rlhf.yaml new file mode 100644 index 0000000..ba89d16 --- /dev/null +++ b/data/claims/instructgpt-uses-rlhf.yaml @@ -0,0 +1,9 @@ +id: claim:instructgpt-uses-rlhf +type: claim +subject: release:instructgpt +predicate: uses_technique +object: technique:reinforcement-learning-from-human-feedback +status: confirmed_primary +evidence: + - source: source:instructgpt-paper + locator: supervised demonstrations, preference comparisons, reward model, and PPO fine-tuning diff --git a/data/claims/palm-published-by-google.yaml b/data/claims/palm-published-by-google.yaml new file mode 100644 index 0000000..17874e0 --- /dev/null +++ b/data/claims/palm-published-by-google.yaml @@ -0,0 +1,9 @@ +id: claim:palm-published-by-google +type: claim +subject: release:palm +predicate: published_by +object: org:google +status: confirmed_primary +evidence: + - source: source:palm-paper + locator: Google authors and Pathways publication diff --git a/data/claims/palm-release-of-palm.yaml b/data/claims/palm-release-of-palm.yaml new file mode 100644 index 0000000..005339e --- /dev/null +++ b/data/claims/palm-release-of-palm.yaml @@ -0,0 +1,9 @@ +id: claim:palm-release-of-palm +type: claim +subject: release:palm +predicate: release_of +object: family:palm +status: confirmed_primary +evidence: + - source: source:palm-paper + locator: paper introduces Pathways Language Model (PaLM) diff --git a/data/claims/palm-uses-transformer.yaml b/data/claims/palm-uses-transformer.yaml new file mode 100644 index 0000000..c0f813d --- /dev/null +++ b/data/claims/palm-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:palm-uses-transformer +type: claim +subject: release:palm +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:palm-paper + locator: abstract — 540-billion parameter densely activated Transformer language model diff --git a/data/claims/switch-transformer-published-by-google.yaml b/data/claims/switch-transformer-published-by-google.yaml new file mode 100644 index 0000000..5c9bea6 --- /dev/null +++ b/data/claims/switch-transformer-published-by-google.yaml @@ -0,0 +1,9 @@ +id: claim:switch-transformer-published-by-google +type: claim +subject: release:switch-transformer +predicate: published_by +object: org:google +status: confirmed_primary +evidence: + - source: source:switch-transformer-paper + locator: Google authors and publication diff --git a/data/claims/switch-transformer-release-of-switch-transformer.yaml b/data/claims/switch-transformer-release-of-switch-transformer.yaml new file mode 100644 index 0000000..f645cd3 --- /dev/null +++ b/data/claims/switch-transformer-release-of-switch-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:switch-transformer-release-of-switch-transformer +type: claim +subject: release:switch-transformer +predicate: release_of +object: family:switch-transformer +status: confirmed_primary +evidence: + - source: source:switch-transformer-paper + locator: paper introduces Switch Transformer diff --git a/data/claims/switch-transformer-uses-sparse-moe.yaml b/data/claims/switch-transformer-uses-sparse-moe.yaml new file mode 100644 index 0000000..f855388 --- /dev/null +++ b/data/claims/switch-transformer-uses-sparse-moe.yaml @@ -0,0 +1,9 @@ +id: claim:switch-transformer-uses-sparse-moe +type: claim +subject: release:switch-transformer +predicate: uses_technique +object: technique:sparse-mixture-of-experts +status: confirmed_primary +evidence: + - source: source:switch-transformer-paper + locator: sparse expert routing and sparsely activated model description diff --git a/data/claims/switch-transformer-uses-transformer.yaml b/data/claims/switch-transformer-uses-transformer.yaml new file mode 100644 index 0000000..108acfc --- /dev/null +++ b/data/claims/switch-transformer-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:switch-transformer-uses-transformer +type: claim +subject: release:switch-transformer +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:switch-transformer-paper + locator: Switch Transformer models based on T5 Transformer models diff --git a/data/claims/t5-published-by-google.yaml b/data/claims/t5-published-by-google.yaml new file mode 100644 index 0000000..1d747ac --- /dev/null +++ b/data/claims/t5-published-by-google.yaml @@ -0,0 +1,9 @@ +id: claim:t5-published-by-google +type: claim +subject: release:t5 +predicate: published_by +object: org:google +status: confirmed_primary +evidence: + - source: source:t5-paper + locator: Google authors and publication diff --git a/data/claims/t5-release-of-t5.yaml b/data/claims/t5-release-of-t5.yaml new file mode 100644 index 0000000..c91bcee --- /dev/null +++ b/data/claims/t5-release-of-t5.yaml @@ -0,0 +1,9 @@ +id: claim:t5-release-of-t5 +type: claim +subject: release:t5 +predicate: release_of +object: family:t5 +status: confirmed_primary +evidence: + - source: source:t5-paper + locator: paper introduces the Text-to-Text Transfer Transformer (T5) diff --git a/data/claims/t5-uses-text-to-text-transfer.yaml b/data/claims/t5-uses-text-to-text-transfer.yaml new file mode 100644 index 0000000..98ec33e --- /dev/null +++ b/data/claims/t5-uses-text-to-text-transfer.yaml @@ -0,0 +1,9 @@ +id: claim:t5-uses-text-to-text-transfer +type: claim +subject: release:t5 +predicate: uses_technique +object: technique:text-to-text-transfer +status: confirmed_primary +evidence: + - source: source:t5-paper + locator: unified text-to-text framework diff --git a/data/claims/t5-uses-transformer.yaml b/data/claims/t5-uses-transformer.yaml new file mode 100644 index 0000000..082d606 --- /dev/null +++ b/data/claims/t5-uses-transformer.yaml @@ -0,0 +1,9 @@ +id: claim:t5-uses-transformer +type: claim +subject: release:t5 +predicate: uses_technique +object: technique:transformer +status: confirmed_primary +evidence: + - source: source:t5-paper + locator: Text-to-Text Transfer Transformer architecture diff --git a/data/events/2018-06-11-gpt-1-announced.yaml b/data/events/2018-06-11-gpt-1-announced.yaml new file mode 100644 index 0000000..e4a75a0 --- /dev/null +++ b/data/events/2018-06-11-gpt-1-announced.yaml @@ -0,0 +1,8 @@ +id: event:2018-06-11-gpt-1-announced +type: event +name: OpenAI generative pre-training model announced +event_type: model_announced +subject: release:gpt-1 +occurred_at: '2018-06-11' +evidence: + - source:gpt1-openai-announcement diff --git a/data/events/2018-10-11-bert-paper.yaml b/data/events/2018-10-11-bert-paper.yaml new file mode 100644 index 0000000..d697dac --- /dev/null +++ b/data/events/2018-10-11-bert-paper.yaml @@ -0,0 +1,8 @@ +id: event:2018-10-11-bert-paper +type: event +name: BERT paper published +event_type: paper_published +subject: release:bert +occurred_at: '2018-10-11' +evidence: + - source:bert-paper diff --git a/data/events/2019-02-14-gpt-2-announced.yaml b/data/events/2019-02-14-gpt-2-announced.yaml new file mode 100644 index 0000000..141af4c --- /dev/null +++ b/data/events/2019-02-14-gpt-2-announced.yaml @@ -0,0 +1,8 @@ +id: event:2019-02-14-gpt-2-announced +type: event +name: GPT-2 announced +event_type: model_announced +subject: release:gpt-2 +occurred_at: '2019-02-14' +evidence: + - source:gpt2-openai-announcement diff --git a/data/events/2019-10-16-rmsnorm-paper.yaml b/data/events/2019-10-16-rmsnorm-paper.yaml new file mode 100644 index 0000000..d7c5f92 --- /dev/null +++ b/data/events/2019-10-16-rmsnorm-paper.yaml @@ -0,0 +1,8 @@ +id: event:2019-10-16-rmsnorm-paper +type: event +name: RMSNorm paper published +event_type: paper_published +subject: technique:rmsnorm +occurred_at: '2019-10-16' +evidence: + - source:rmsnorm-paper diff --git a/data/events/2019-10-23-t5-paper.yaml b/data/events/2019-10-23-t5-paper.yaml new file mode 100644 index 0000000..9edafce --- /dev/null +++ b/data/events/2019-10-23-t5-paper.yaml @@ -0,0 +1,8 @@ +id: event:2019-10-23-t5-paper +type: event +name: T5 paper published +event_type: paper_published +subject: release:t5 +occurred_at: '2019-10-23' +evidence: + - source:t5-paper diff --git a/data/events/2020-01-23-scaling-laws-paper.yaml b/data/events/2020-01-23-scaling-laws-paper.yaml new file mode 100644 index 0000000..5f09717 --- /dev/null +++ b/data/events/2020-01-23-scaling-laws-paper.yaml @@ -0,0 +1,8 @@ +id: event:2020-01-23-scaling-laws-paper +type: event +name: Neural language-model scaling laws paper published +event_type: paper_published +subject: technique:neural-scaling-laws +occurred_at: '2020-01-23' +evidence: + - source:scaling-laws-paper diff --git a/data/events/2020-02-12-swiglu-paper.yaml b/data/events/2020-02-12-swiglu-paper.yaml new file mode 100644 index 0000000..60e3444 --- /dev/null +++ b/data/events/2020-02-12-swiglu-paper.yaml @@ -0,0 +1,8 @@ +id: event:2020-02-12-swiglu-paper +type: event +name: GLU variants and SwiGLU paper published +event_type: paper_published +subject: technique:swiglu +occurred_at: '2020-02-12' +evidence: + - source:swiglu-paper diff --git a/data/events/2020-05-28-gpt-3-paper.yaml b/data/events/2020-05-28-gpt-3-paper.yaml new file mode 100644 index 0000000..fa23c62 --- /dev/null +++ b/data/events/2020-05-28-gpt-3-paper.yaml @@ -0,0 +1,8 @@ +id: event:2020-05-28-gpt-3-paper +type: event +name: GPT-3 paper published +event_type: paper_published +subject: release:gpt-3 +occurred_at: '2020-05-28' +evidence: + - source:gpt3-paper diff --git a/data/events/2021-01-11-switch-transformer-paper.yaml b/data/events/2021-01-11-switch-transformer-paper.yaml new file mode 100644 index 0000000..583e1cc --- /dev/null +++ b/data/events/2021-01-11-switch-transformer-paper.yaml @@ -0,0 +1,8 @@ +id: event:2021-01-11-switch-transformer-paper +type: event +name: Switch Transformer paper published +event_type: paper_published +subject: release:switch-transformer +occurred_at: '2021-01-11' +evidence: + - source:switch-transformer-paper diff --git a/data/events/2021-09-03-flan-paper.yaml b/data/events/2021-09-03-flan-paper.yaml new file mode 100644 index 0000000..24da401 --- /dev/null +++ b/data/events/2021-09-03-flan-paper.yaml @@ -0,0 +1,8 @@ +id: event:2021-09-03-flan-paper +type: event +name: FLAN instruction-tuning paper published +event_type: paper_published +subject: technique:instruction-tuning +occurred_at: '2021-09-03' +evidence: + - source:flan-paper diff --git a/data/events/2022-01-28-chain-of-thought-paper.yaml b/data/events/2022-01-28-chain-of-thought-paper.yaml new file mode 100644 index 0000000..5d8f046 --- /dev/null +++ b/data/events/2022-01-28-chain-of-thought-paper.yaml @@ -0,0 +1,8 @@ +id: event:2022-01-28-chain-of-thought-paper +type: event +name: Chain-of-thought prompting paper published +event_type: paper_published +subject: technique:chain-of-thought-prompting +occurred_at: '2022-01-28' +evidence: + - source:cot-paper diff --git a/data/events/2022-03-04-instructgpt-paper.yaml b/data/events/2022-03-04-instructgpt-paper.yaml new file mode 100644 index 0000000..f2fa8f6 --- /dev/null +++ b/data/events/2022-03-04-instructgpt-paper.yaml @@ -0,0 +1,8 @@ +id: event:2022-03-04-instructgpt-paper +type: event +name: InstructGPT paper published +event_type: paper_published +subject: release:instructgpt +occurred_at: '2022-03-04' +evidence: + - source:instructgpt-paper diff --git a/data/events/2022-03-29-chinchilla-paper.yaml b/data/events/2022-03-29-chinchilla-paper.yaml new file mode 100644 index 0000000..ecfde78 --- /dev/null +++ b/data/events/2022-03-29-chinchilla-paper.yaml @@ -0,0 +1,8 @@ +id: event:2022-03-29-chinchilla-paper +type: event +name: Chinchilla compute-optimal training paper published +event_type: paper_published +subject: release:chinchilla +occurred_at: '2022-03-29' +evidence: + - source:chinchilla-paper diff --git a/data/events/2022-04-05-palm-paper.yaml b/data/events/2022-04-05-palm-paper.yaml new file mode 100644 index 0000000..df43b44 --- /dev/null +++ b/data/events/2022-04-05-palm-paper.yaml @@ -0,0 +1,8 @@ +id: event:2022-04-05-palm-paper +type: event +name: PaLM paper published +event_type: paper_published +subject: release:palm +occurred_at: '2022-04-05' +evidence: + - source:palm-paper diff --git a/data/events/2022-05-27-flashattention-paper.yaml b/data/events/2022-05-27-flashattention-paper.yaml new file mode 100644 index 0000000..b6d7c68 --- /dev/null +++ b/data/events/2022-05-27-flashattention-paper.yaml @@ -0,0 +1,8 @@ +id: event:2022-05-27-flashattention-paper +type: event +name: FlashAttention paper published +event_type: paper_published +subject: technique:flashattention +occurred_at: '2022-05-27' +evidence: + - source:flashattention-paper diff --git a/data/families/bert.yaml b/data/families/bert.yaml new file mode 100644 index 0000000..5ef4ccd --- /dev/null +++ b/data/families/bert.yaml @@ -0,0 +1,4 @@ +id: family:bert +type: model_family +name: BERT +status: canonical diff --git a/data/families/chinchilla.yaml b/data/families/chinchilla.yaml new file mode 100644 index 0000000..d1fba5c --- /dev/null +++ b/data/families/chinchilla.yaml @@ -0,0 +1,4 @@ +id: family:chinchilla +type: model_family +name: Chinchilla +status: canonical diff --git a/data/families/switch-transformer.yaml b/data/families/switch-transformer.yaml new file mode 100644 index 0000000..0d7fb98 --- /dev/null +++ b/data/families/switch-transformer.yaml @@ -0,0 +1,4 @@ +id: family:switch-transformer +type: model_family +name: Switch Transformer +status: canonical diff --git a/data/families/t5.yaml b/data/families/t5.yaml new file mode 100644 index 0000000..e765b0d --- /dev/null +++ b/data/families/t5.yaml @@ -0,0 +1,4 @@ +id: family:t5 +type: model_family +name: T5 +status: canonical diff --git a/data/organizations/deepmind.yaml b/data/organizations/deepmind.yaml new file mode 100644 index 0000000..c04dbae --- /dev/null +++ b/data/organizations/deepmind.yaml @@ -0,0 +1,4 @@ +id: org:deepmind +type: organization +name: DeepMind +status: canonical diff --git a/data/organizations/google.yaml b/data/organizations/google.yaml new file mode 100644 index 0000000..8b0a050 --- /dev/null +++ b/data/organizations/google.yaml @@ -0,0 +1,4 @@ +id: org:google +type: organization +name: Google +status: canonical diff --git a/data/relationships/bert-published-by-google.yaml b/data/relationships/bert-published-by-google.yaml new file mode 100644 index 0000000..8335d08 --- /dev/null +++ b/data/relationships/bert-published-by-google.yaml @@ -0,0 +1,7 @@ +id: relationship:bert-published-by-google +type: relationship +subject: release:bert +predicate: published_by +object: org:google +claim_refs: + - claim:bert-published-by-google diff --git a/data/relationships/bert-release-of-bert.yaml b/data/relationships/bert-release-of-bert.yaml new file mode 100644 index 0000000..46b1df5 --- /dev/null +++ b/data/relationships/bert-release-of-bert.yaml @@ -0,0 +1,7 @@ +id: relationship:bert-release-of-bert +type: relationship +subject: release:bert +predicate: release_of +object: family:bert +claim_refs: + - claim:bert-release-of-bert diff --git a/data/relationships/bert-uses-masked-language-modeling.yaml b/data/relationships/bert-uses-masked-language-modeling.yaml new file mode 100644 index 0000000..d8e0901 --- /dev/null +++ b/data/relationships/bert-uses-masked-language-modeling.yaml @@ -0,0 +1,7 @@ +id: relationship:bert-uses-masked-language-modeling +type: relationship +subject: release:bert +predicate: uses_technique +object: technique:masked-language-modeling +claim_refs: + - claim:bert-uses-masked-language-modeling diff --git a/data/relationships/bert-uses-transformer.yaml b/data/relationships/bert-uses-transformer.yaml new file mode 100644 index 0000000..6f02cf2 --- /dev/null +++ b/data/relationships/bert-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:bert-uses-transformer +type: relationship +subject: release:bert +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:bert-uses-transformer diff --git a/data/relationships/chinchilla-published-by-deepmind.yaml b/data/relationships/chinchilla-published-by-deepmind.yaml new file mode 100644 index 0000000..90a4f6f --- /dev/null +++ b/data/relationships/chinchilla-published-by-deepmind.yaml @@ -0,0 +1,7 @@ +id: relationship:chinchilla-published-by-deepmind +type: relationship +subject: release:chinchilla +predicate: published_by +object: org:deepmind +claim_refs: + - claim:chinchilla-published-by-deepmind diff --git a/data/relationships/chinchilla-release-of-chinchilla.yaml b/data/relationships/chinchilla-release-of-chinchilla.yaml new file mode 100644 index 0000000..cfe871a --- /dev/null +++ b/data/relationships/chinchilla-release-of-chinchilla.yaml @@ -0,0 +1,7 @@ +id: relationship:chinchilla-release-of-chinchilla +type: relationship +subject: release:chinchilla +predicate: release_of +object: family:chinchilla +claim_refs: + - claim:chinchilla-release-of-chinchilla diff --git a/data/relationships/chinchilla-uses-compute-optimal-training.yaml b/data/relationships/chinchilla-uses-compute-optimal-training.yaml new file mode 100644 index 0000000..20895b9 --- /dev/null +++ b/data/relationships/chinchilla-uses-compute-optimal-training.yaml @@ -0,0 +1,7 @@ +id: relationship:chinchilla-uses-compute-optimal-training +type: relationship +subject: release:chinchilla +predicate: uses_technique +object: technique:compute-optimal-training +claim_refs: + - claim:chinchilla-uses-compute-optimal-training diff --git a/data/relationships/chinchilla-uses-transformer.yaml b/data/relationships/chinchilla-uses-transformer.yaml new file mode 100644 index 0000000..bee492e --- /dev/null +++ b/data/relationships/chinchilla-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:chinchilla-uses-transformer +type: relationship +subject: release:chinchilla +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:chinchilla-uses-transformer diff --git a/data/relationships/gpt-1-published-by-openai.yaml b/data/relationships/gpt-1-published-by-openai.yaml new file mode 100644 index 0000000..f54ff7e --- /dev/null +++ b/data/relationships/gpt-1-published-by-openai.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-1-published-by-openai +type: relationship +subject: release:gpt-1 +predicate: published_by +object: org:openai +claim_refs: + - claim:gpt-1-published-by-openai diff --git a/data/relationships/gpt-1-release-of-gpt.yaml b/data/relationships/gpt-1-release-of-gpt.yaml new file mode 100644 index 0000000..be7e1a8 --- /dev/null +++ b/data/relationships/gpt-1-release-of-gpt.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-1-release-of-gpt +type: relationship +subject: release:gpt-1 +predicate: release_of +object: family:gpt +claim_refs: + - claim:gpt-1-release-of-gpt diff --git a/data/relationships/gpt-1-uses-generative-pretraining.yaml b/data/relationships/gpt-1-uses-generative-pretraining.yaml new file mode 100644 index 0000000..3019038 --- /dev/null +++ b/data/relationships/gpt-1-uses-generative-pretraining.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-1-uses-generative-pretraining +type: relationship +subject: release:gpt-1 +predicate: uses_technique +object: technique:generative-pretraining +claim_refs: + - claim:gpt-1-uses-generative-pretraining diff --git a/data/relationships/gpt-1-uses-transformer.yaml b/data/relationships/gpt-1-uses-transformer.yaml new file mode 100644 index 0000000..7f31f31 --- /dev/null +++ b/data/relationships/gpt-1-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-1-uses-transformer +type: relationship +subject: release:gpt-1 +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:gpt-1-uses-transformer diff --git a/data/relationships/gpt-2-published-by-openai.yaml b/data/relationships/gpt-2-published-by-openai.yaml new file mode 100644 index 0000000..0a211b5 --- /dev/null +++ b/data/relationships/gpt-2-published-by-openai.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-2-published-by-openai +type: relationship +subject: release:gpt-2 +predicate: published_by +object: org:openai +claim_refs: + - claim:gpt-2-published-by-openai diff --git a/data/relationships/gpt-2-release-of-gpt.yaml b/data/relationships/gpt-2-release-of-gpt.yaml new file mode 100644 index 0000000..1a82379 --- /dev/null +++ b/data/relationships/gpt-2-release-of-gpt.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-2-release-of-gpt +type: relationship +subject: release:gpt-2 +predicate: release_of +object: family:gpt +claim_refs: + - claim:gpt-2-release-of-gpt diff --git a/data/relationships/gpt-2-uses-generative-pretraining.yaml b/data/relationships/gpt-2-uses-generative-pretraining.yaml new file mode 100644 index 0000000..421a251 --- /dev/null +++ b/data/relationships/gpt-2-uses-generative-pretraining.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-2-uses-generative-pretraining +type: relationship +subject: release:gpt-2 +predicate: uses_technique +object: technique:generative-pretraining +claim_refs: + - claim:gpt-2-uses-generative-pretraining diff --git a/data/relationships/gpt-2-uses-transformer.yaml b/data/relationships/gpt-2-uses-transformer.yaml new file mode 100644 index 0000000..cb461bc --- /dev/null +++ b/data/relationships/gpt-2-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-2-uses-transformer +type: relationship +subject: release:gpt-2 +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:gpt-2-uses-transformer diff --git a/data/relationships/gpt-3-published-by-openai.yaml b/data/relationships/gpt-3-published-by-openai.yaml new file mode 100644 index 0000000..8f3ade6 --- /dev/null +++ b/data/relationships/gpt-3-published-by-openai.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-3-published-by-openai +type: relationship +subject: release:gpt-3 +predicate: published_by +object: org:openai +claim_refs: + - claim:gpt-3-published-by-openai diff --git a/data/relationships/gpt-3-release-of-gpt.yaml b/data/relationships/gpt-3-release-of-gpt.yaml new file mode 100644 index 0000000..c874185 --- /dev/null +++ b/data/relationships/gpt-3-release-of-gpt.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-3-release-of-gpt +type: relationship +subject: release:gpt-3 +predicate: release_of +object: family:gpt +claim_refs: + - claim:gpt-3-release-of-gpt diff --git a/data/relationships/gpt-3-uses-in-context-learning.yaml b/data/relationships/gpt-3-uses-in-context-learning.yaml new file mode 100644 index 0000000..426bc8e --- /dev/null +++ b/data/relationships/gpt-3-uses-in-context-learning.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-3-uses-in-context-learning +type: relationship +subject: release:gpt-3 +predicate: uses_technique +object: technique:in-context-learning +claim_refs: + - claim:gpt-3-uses-in-context-learning diff --git a/data/relationships/gpt-3-uses-transformer.yaml b/data/relationships/gpt-3-uses-transformer.yaml new file mode 100644 index 0000000..0f71fc9 --- /dev/null +++ b/data/relationships/gpt-3-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-3-uses-transformer +type: relationship +subject: release:gpt-3 +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:gpt-3-uses-transformer diff --git a/data/relationships/instructgpt-fine-tuned-from-gpt-3.yaml b/data/relationships/instructgpt-fine-tuned-from-gpt-3.yaml new file mode 100644 index 0000000..f9a5356 --- /dev/null +++ b/data/relationships/instructgpt-fine-tuned-from-gpt-3.yaml @@ -0,0 +1,7 @@ +id: relationship:instructgpt-fine-tuned-from-gpt-3 +type: relationship +subject: release:instructgpt +predicate: fine_tuned_from +object: release:gpt-3 +claim_refs: + - claim:instructgpt-fine-tuned-from-gpt-3 diff --git a/data/relationships/instructgpt-published-by-openai.yaml b/data/relationships/instructgpt-published-by-openai.yaml new file mode 100644 index 0000000..5468c84 --- /dev/null +++ b/data/relationships/instructgpt-published-by-openai.yaml @@ -0,0 +1,7 @@ +id: relationship:instructgpt-published-by-openai +type: relationship +subject: release:instructgpt +predicate: published_by +object: org:openai +claim_refs: + - claim:instructgpt-published-by-openai diff --git a/data/relationships/instructgpt-release-of-gpt.yaml b/data/relationships/instructgpt-release-of-gpt.yaml new file mode 100644 index 0000000..15bea74 --- /dev/null +++ b/data/relationships/instructgpt-release-of-gpt.yaml @@ -0,0 +1,7 @@ +id: relationship:instructgpt-release-of-gpt +type: relationship +subject: release:instructgpt +predicate: release_of +object: family:gpt +claim_refs: + - claim:instructgpt-release-of-gpt diff --git a/data/relationships/instructgpt-uses-rlhf.yaml b/data/relationships/instructgpt-uses-rlhf.yaml new file mode 100644 index 0000000..8d6e4df --- /dev/null +++ b/data/relationships/instructgpt-uses-rlhf.yaml @@ -0,0 +1,7 @@ +id: relationship:instructgpt-uses-rlhf +type: relationship +subject: release:instructgpt +predicate: uses_technique +object: technique:reinforcement-learning-from-human-feedback +claim_refs: + - claim:instructgpt-uses-rlhf diff --git a/data/relationships/palm-published-by-google.yaml b/data/relationships/palm-published-by-google.yaml new file mode 100644 index 0000000..d56f749 --- /dev/null +++ b/data/relationships/palm-published-by-google.yaml @@ -0,0 +1,7 @@ +id: relationship:palm-published-by-google +type: relationship +subject: release:palm +predicate: published_by +object: org:google +claim_refs: + - claim:palm-published-by-google diff --git a/data/relationships/palm-release-of-palm.yaml b/data/relationships/palm-release-of-palm.yaml new file mode 100644 index 0000000..3f500c1 --- /dev/null +++ b/data/relationships/palm-release-of-palm.yaml @@ -0,0 +1,7 @@ +id: relationship:palm-release-of-palm +type: relationship +subject: release:palm +predicate: release_of +object: family:palm +claim_refs: + - claim:palm-release-of-palm diff --git a/data/relationships/palm-uses-transformer.yaml b/data/relationships/palm-uses-transformer.yaml new file mode 100644 index 0000000..3fa3f1c --- /dev/null +++ b/data/relationships/palm-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:palm-uses-transformer +type: relationship +subject: release:palm +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:palm-uses-transformer diff --git a/data/relationships/switch-transformer-published-by-google.yaml b/data/relationships/switch-transformer-published-by-google.yaml new file mode 100644 index 0000000..8bcb6c0 --- /dev/null +++ b/data/relationships/switch-transformer-published-by-google.yaml @@ -0,0 +1,7 @@ +id: relationship:switch-transformer-published-by-google +type: relationship +subject: release:switch-transformer +predicate: published_by +object: org:google +claim_refs: + - claim:switch-transformer-published-by-google diff --git a/data/relationships/switch-transformer-release-of-switch-transformer.yaml b/data/relationships/switch-transformer-release-of-switch-transformer.yaml new file mode 100644 index 0000000..3e4d892 --- /dev/null +++ b/data/relationships/switch-transformer-release-of-switch-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:switch-transformer-release-of-switch-transformer +type: relationship +subject: release:switch-transformer +predicate: release_of +object: family:switch-transformer +claim_refs: + - claim:switch-transformer-release-of-switch-transformer diff --git a/data/relationships/switch-transformer-uses-sparse-moe.yaml b/data/relationships/switch-transformer-uses-sparse-moe.yaml new file mode 100644 index 0000000..9bcedc6 --- /dev/null +++ b/data/relationships/switch-transformer-uses-sparse-moe.yaml @@ -0,0 +1,7 @@ +id: relationship:switch-transformer-uses-sparse-moe +type: relationship +subject: release:switch-transformer +predicate: uses_technique +object: technique:sparse-mixture-of-experts +claim_refs: + - claim:switch-transformer-uses-sparse-moe diff --git a/data/relationships/switch-transformer-uses-transformer.yaml b/data/relationships/switch-transformer-uses-transformer.yaml new file mode 100644 index 0000000..08ced30 --- /dev/null +++ b/data/relationships/switch-transformer-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:switch-transformer-uses-transformer +type: relationship +subject: release:switch-transformer +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:switch-transformer-uses-transformer diff --git a/data/relationships/t5-published-by-google.yaml b/data/relationships/t5-published-by-google.yaml new file mode 100644 index 0000000..e18fec6 --- /dev/null +++ b/data/relationships/t5-published-by-google.yaml @@ -0,0 +1,7 @@ +id: relationship:t5-published-by-google +type: relationship +subject: release:t5 +predicate: published_by +object: org:google +claim_refs: + - claim:t5-published-by-google diff --git a/data/relationships/t5-release-of-t5.yaml b/data/relationships/t5-release-of-t5.yaml new file mode 100644 index 0000000..79d7a5c --- /dev/null +++ b/data/relationships/t5-release-of-t5.yaml @@ -0,0 +1,7 @@ +id: relationship:t5-release-of-t5 +type: relationship +subject: release:t5 +predicate: release_of +object: family:t5 +claim_refs: + - claim:t5-release-of-t5 diff --git a/data/relationships/t5-uses-text-to-text-transfer.yaml b/data/relationships/t5-uses-text-to-text-transfer.yaml new file mode 100644 index 0000000..62ad5b4 --- /dev/null +++ b/data/relationships/t5-uses-text-to-text-transfer.yaml @@ -0,0 +1,7 @@ +id: relationship:t5-uses-text-to-text-transfer +type: relationship +subject: release:t5 +predicate: uses_technique +object: technique:text-to-text-transfer +claim_refs: + - claim:t5-uses-text-to-text-transfer diff --git a/data/relationships/t5-uses-transformer.yaml b/data/relationships/t5-uses-transformer.yaml new file mode 100644 index 0000000..ecd9d40 --- /dev/null +++ b/data/relationships/t5-uses-transformer.yaml @@ -0,0 +1,7 @@ +id: relationship:t5-uses-transformer +type: relationship +subject: release:t5 +predicate: uses_technique +object: technique:transformer +claim_refs: + - claim:t5-uses-transformer diff --git a/data/releases/bert.yaml b/data/releases/bert.yaml new file mode 100644 index 0000000..f678814 --- /dev/null +++ b/data/releases/bert.yaml @@ -0,0 +1,4 @@ +id: release:bert +type: model_release +name: BERT +status: canonical diff --git a/data/releases/chinchilla.yaml b/data/releases/chinchilla.yaml new file mode 100644 index 0000000..4a33940 --- /dev/null +++ b/data/releases/chinchilla.yaml @@ -0,0 +1,4 @@ +id: release:chinchilla +type: model_release +name: Chinchilla +status: canonical diff --git a/data/releases/gpt-1.yaml b/data/releases/gpt-1.yaml new file mode 100644 index 0000000..c81e2ae --- /dev/null +++ b/data/releases/gpt-1.yaml @@ -0,0 +1,4 @@ +id: release:gpt-1 +type: model_release +name: GPT-1 +status: canonical diff --git a/data/releases/gpt-2.yaml b/data/releases/gpt-2.yaml new file mode 100644 index 0000000..b4fccd6 --- /dev/null +++ b/data/releases/gpt-2.yaml @@ -0,0 +1,4 @@ +id: release:gpt-2 +type: model_release +name: GPT-2 +status: canonical diff --git a/data/releases/gpt-3.yaml b/data/releases/gpt-3.yaml new file mode 100644 index 0000000..adae5b4 --- /dev/null +++ b/data/releases/gpt-3.yaml @@ -0,0 +1,4 @@ +id: release:gpt-3 +type: model_release +name: GPT-3 +status: canonical diff --git a/data/releases/instructgpt.yaml b/data/releases/instructgpt.yaml new file mode 100644 index 0000000..b4f9034 --- /dev/null +++ b/data/releases/instructgpt.yaml @@ -0,0 +1,4 @@ +id: release:instructgpt +type: model_release +name: InstructGPT +status: canonical diff --git a/data/releases/palm.yaml b/data/releases/palm.yaml new file mode 100644 index 0000000..a30cfb2 --- /dev/null +++ b/data/releases/palm.yaml @@ -0,0 +1,4 @@ +id: release:palm +type: model_release +name: PaLM +status: canonical diff --git a/data/releases/switch-transformer.yaml b/data/releases/switch-transformer.yaml new file mode 100644 index 0000000..a8af072 --- /dev/null +++ b/data/releases/switch-transformer.yaml @@ -0,0 +1,4 @@ +id: release:switch-transformer +type: model_release +name: Switch Transformer +status: canonical diff --git a/data/releases/t5.yaml b/data/releases/t5.yaml new file mode 100644 index 0000000..f71bf97 --- /dev/null +++ b/data/releases/t5.yaml @@ -0,0 +1,4 @@ +id: release:t5 +type: model_release +name: T5 +status: canonical diff --git a/data/sources/bert-paper.yaml b/data/sources/bert-paper.yaml new file mode 100644 index 0000000..873e502 --- /dev/null +++ b/data/sources/bert-paper.yaml @@ -0,0 +1,8 @@ +id: source:bert-paper +type: source +title: 'BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding' +source_type: paper +url: https://arxiv.org/abs/1810.04805 +publisher: Google AI Language +published_at: '2018-10-11' +version: v1 diff --git a/data/sources/chinchilla-paper.yaml b/data/sources/chinchilla-paper.yaml new file mode 100644 index 0000000..1eef381 --- /dev/null +++ b/data/sources/chinchilla-paper.yaml @@ -0,0 +1,8 @@ +id: source:chinchilla-paper +type: source +title: Training Compute-Optimal Large Language Models +source_type: paper +url: https://arxiv.org/abs/2203.15556 +publisher: DeepMind +published_at: '2022-03-29' +version: v1 diff --git a/data/sources/cot-paper.yaml b/data/sources/cot-paper.yaml new file mode 100644 index 0000000..2fe63aa --- /dev/null +++ b/data/sources/cot-paper.yaml @@ -0,0 +1,8 @@ +id: source:cot-paper +type: source +title: Chain-of-Thought Prompting Elicits Reasoning in Large Language Models +source_type: paper +url: https://arxiv.org/abs/2201.11903 +publisher: Google Research +published_at: '2022-01-28' +version: v1 diff --git a/data/sources/flan-paper.yaml b/data/sources/flan-paper.yaml new file mode 100644 index 0000000..8e5d943 --- /dev/null +++ b/data/sources/flan-paper.yaml @@ -0,0 +1,8 @@ +id: source:flan-paper +type: source +title: Finetuned Language Models Are Zero-Shot Learners +source_type: paper +url: https://arxiv.org/abs/2109.01652 +publisher: Google Research +published_at: '2021-09-03' +version: v1 diff --git a/data/sources/flashattention-paper.yaml b/data/sources/flashattention-paper.yaml new file mode 100644 index 0000000..0fae515 --- /dev/null +++ b/data/sources/flashattention-paper.yaml @@ -0,0 +1,7 @@ +id: source:flashattention-paper +type: source +title: 'FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness' +source_type: paper +url: https://arxiv.org/abs/2205.14135 +published_at: '2022-05-27' +version: v1 diff --git a/data/sources/gpt1-openai-announcement.yaml b/data/sources/gpt1-openai-announcement.yaml new file mode 100644 index 0000000..8760b88 --- /dev/null +++ b/data/sources/gpt1-openai-announcement.yaml @@ -0,0 +1,7 @@ +id: source:gpt1-openai-announcement +type: source +title: Improving language understanding with unsupervised learning +source_type: official_announcement +url: https://openai.com/index/language-unsupervised/ +publisher: OpenAI +published_at: '2018-06-11' diff --git a/data/sources/gpt2-openai-announcement.yaml b/data/sources/gpt2-openai-announcement.yaml new file mode 100644 index 0000000..bbff29f --- /dev/null +++ b/data/sources/gpt2-openai-announcement.yaml @@ -0,0 +1,7 @@ +id: source:gpt2-openai-announcement +type: source +title: Better language models and their implications +source_type: official_announcement +url: https://openai.com/index/better-language-models/ +publisher: OpenAI +published_at: '2019-02-14' diff --git a/data/sources/gpt3-paper.yaml b/data/sources/gpt3-paper.yaml new file mode 100644 index 0000000..1d84e81 --- /dev/null +++ b/data/sources/gpt3-paper.yaml @@ -0,0 +1,8 @@ +id: source:gpt3-paper +type: source +title: Language Models are Few-Shot Learners +source_type: paper +url: https://arxiv.org/abs/2005.14165 +publisher: OpenAI +published_at: '2020-05-28' +version: v1 diff --git a/data/sources/instructgpt-paper.yaml b/data/sources/instructgpt-paper.yaml new file mode 100644 index 0000000..8984482 --- /dev/null +++ b/data/sources/instructgpt-paper.yaml @@ -0,0 +1,8 @@ +id: source:instructgpt-paper +type: source +title: Training language models to follow instructions with human feedback +source_type: paper +url: https://arxiv.org/abs/2203.02155 +publisher: OpenAI +published_at: '2022-03-04' +version: v1 diff --git a/data/sources/palm-paper.yaml b/data/sources/palm-paper.yaml new file mode 100644 index 0000000..3ded138 --- /dev/null +++ b/data/sources/palm-paper.yaml @@ -0,0 +1,8 @@ +id: source:palm-paper +type: source +title: 'PaLM: Scaling Language Modeling with Pathways' +source_type: paper +url: https://arxiv.org/abs/2204.02311 +publisher: Google +published_at: '2022-04-05' +version: v1 diff --git a/data/sources/rmsnorm-paper.yaml b/data/sources/rmsnorm-paper.yaml new file mode 100644 index 0000000..65ec590 --- /dev/null +++ b/data/sources/rmsnorm-paper.yaml @@ -0,0 +1,7 @@ +id: source:rmsnorm-paper +type: source +title: Root Mean Square Layer Normalization +source_type: paper +url: https://arxiv.org/abs/1910.07467 +published_at: '2019-10-16' +version: v1 diff --git a/data/sources/scaling-laws-paper.yaml b/data/sources/scaling-laws-paper.yaml new file mode 100644 index 0000000..72ef76d --- /dev/null +++ b/data/sources/scaling-laws-paper.yaml @@ -0,0 +1,8 @@ +id: source:scaling-laws-paper +type: source +title: Scaling Laws for Neural Language Models +source_type: paper +url: https://arxiv.org/abs/2001.08361 +publisher: OpenAI +published_at: '2020-01-23' +version: v1 diff --git a/data/sources/swiglu-paper.yaml b/data/sources/swiglu-paper.yaml new file mode 100644 index 0000000..7cada55 --- /dev/null +++ b/data/sources/swiglu-paper.yaml @@ -0,0 +1,7 @@ +id: source:swiglu-paper +type: source +title: GLU Variants Improve Transformer +source_type: paper +url: https://arxiv.org/abs/2002.05202 +published_at: '2020-02-12' +version: v1 diff --git a/data/sources/switch-transformer-paper.yaml b/data/sources/switch-transformer-paper.yaml new file mode 100644 index 0000000..d108745 --- /dev/null +++ b/data/sources/switch-transformer-paper.yaml @@ -0,0 +1,8 @@ +id: source:switch-transformer-paper +type: source +title: 'Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity' +source_type: paper +url: https://arxiv.org/abs/2101.03961 +publisher: Google +published_at: '2021-01-11' +version: v1 diff --git a/data/sources/t5-paper.yaml b/data/sources/t5-paper.yaml new file mode 100644 index 0000000..546c3e9 --- /dev/null +++ b/data/sources/t5-paper.yaml @@ -0,0 +1,8 @@ +id: source:t5-paper +type: source +title: Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer +source_type: paper +url: https://arxiv.org/abs/1910.10683 +publisher: Google +published_at: '2019-10-23' +version: v1 diff --git a/data/techniques/chain-of-thought-prompting.yaml b/data/techniques/chain-of-thought-prompting.yaml new file mode 100644 index 0000000..1c284fe --- /dev/null +++ b/data/techniques/chain-of-thought-prompting.yaml @@ -0,0 +1,7 @@ +id: technique:chain-of-thought-prompting +type: technique +name: Chain-of-thought prompting +aliases: + - CoT prompting +status: canonical +description: Prompt a language model with intermediate reasoning-step demonstrations to elicit multi-step reasoning. diff --git a/data/techniques/compute-optimal-training.yaml b/data/techniques/compute-optimal-training.yaml new file mode 100644 index 0000000..b5d6552 --- /dev/null +++ b/data/techniques/compute-optimal-training.yaml @@ -0,0 +1,5 @@ +id: technique:compute-optimal-training +type: technique +name: Compute-optimal language-model training +status: canonical +description: Allocate a fixed training-compute budget between model parameters and training tokens to improve compute efficiency. diff --git a/data/techniques/flashattention.yaml b/data/techniques/flashattention.yaml new file mode 100644 index 0000000..f501a64 --- /dev/null +++ b/data/techniques/flashattention.yaml @@ -0,0 +1,5 @@ +id: technique:flashattention +type: technique +name: FlashAttention +status: canonical +description: IO-aware exact attention algorithm that reduces memory traffic through tiling between GPU memory levels. diff --git a/data/techniques/generative-pretraining.yaml b/data/techniques/generative-pretraining.yaml new file mode 100644 index 0000000..46bc75e --- /dev/null +++ b/data/techniques/generative-pretraining.yaml @@ -0,0 +1,5 @@ +id: technique:generative-pretraining +type: technique +name: Generative language-model pretraining +status: canonical +description: Pretrain an autoregressive language model on unlabeled text before adapting it to downstream tasks. diff --git a/data/techniques/in-context-learning.yaml b/data/techniques/in-context-learning.yaml new file mode 100644 index 0000000..fdbedd6 --- /dev/null +++ b/data/techniques/in-context-learning.yaml @@ -0,0 +1,5 @@ +id: technique:in-context-learning +type: technique +name: In-context learning +status: canonical +description: Adapt model behavior from instructions or demonstrations supplied in the prompt without gradient updates. diff --git a/data/techniques/instruction-tuning.yaml b/data/techniques/instruction-tuning.yaml new file mode 100644 index 0000000..46032bc --- /dev/null +++ b/data/techniques/instruction-tuning.yaml @@ -0,0 +1,5 @@ +id: technique:instruction-tuning +type: technique +name: Instruction tuning +status: canonical +description: Fine-tune a pretrained language model across tasks expressed through natural-language instructions to improve generalization to unseen instructions. diff --git a/data/techniques/masked-language-modeling.yaml b/data/techniques/masked-language-modeling.yaml new file mode 100644 index 0000000..d5eaaf4 --- /dev/null +++ b/data/techniques/masked-language-modeling.yaml @@ -0,0 +1,5 @@ +id: technique:masked-language-modeling +type: technique +name: Masked language modeling +status: canonical +description: Pretraining objective that predicts masked tokens using bidirectional context. diff --git a/data/techniques/neural-scaling-laws.yaml b/data/techniques/neural-scaling-laws.yaml new file mode 100644 index 0000000..57428bb --- /dev/null +++ b/data/techniques/neural-scaling-laws.yaml @@ -0,0 +1,5 @@ +id: technique:neural-scaling-laws +type: technique +name: Neural language-model scaling laws +status: canonical +description: Empirical power-law relationships between language-model loss, model size, dataset size, and training compute. diff --git a/data/techniques/reinforcement-learning-from-human-feedback.yaml b/data/techniques/reinforcement-learning-from-human-feedback.yaml new file mode 100644 index 0000000..827129a --- /dev/null +++ b/data/techniques/reinforcement-learning-from-human-feedback.yaml @@ -0,0 +1,6 @@ +id: technique:reinforcement-learning-from-human-feedback +type: technique +name: Reinforcement learning from human feedback +aliases: + - RLHF +status: canonical diff --git a/data/techniques/rmsnorm.yaml b/data/techniques/rmsnorm.yaml new file mode 100644 index 0000000..e2da508 --- /dev/null +++ b/data/techniques/rmsnorm.yaml @@ -0,0 +1,6 @@ +id: technique:rmsnorm +type: technique +name: Root Mean Square Layer Normalization +aliases: + - RMSNorm +status: canonical diff --git a/data/techniques/swiglu.yaml b/data/techniques/swiglu.yaml new file mode 100644 index 0000000..94aa329 --- /dev/null +++ b/data/techniques/swiglu.yaml @@ -0,0 +1,5 @@ +id: technique:swiglu +type: technique +name: SwiGLU +status: canonical +description: Gated feed-forward activation using Swish in a GLU-style Transformer feed-forward block. diff --git a/data/techniques/text-to-text-transfer.yaml b/data/techniques/text-to-text-transfer.yaml new file mode 100644 index 0000000..6876155 --- /dev/null +++ b/data/techniques/text-to-text-transfer.yaml @@ -0,0 +1,5 @@ +id: technique:text-to-text-transfer +type: technique +name: Text-to-text transfer +status: canonical +description: Frame diverse NLP tasks as text input to text output under one model and objective interface. diff --git a/narratives/foundations/README.md b/narratives/foundations/README.md new file mode 100644 index 0000000..ca433f9 --- /dev/null +++ b/narratives/foundations/README.md @@ -0,0 +1,13 @@ +# 2017–2022 foundation narrative scaffold + +This directory is for editorial synthesis derived from canonical ModelMap records. Do not duplicate canonical dates, model identities, organization links, or technique edges here as a second source of truth. + +The foundation chapter should stay compressed and answer only the questions needed to understand the high-resolution 2023–2026 history: + +- How did Transformer pretraining evolve from task-specific fine-tuning toward in-context learning? +- How did BERT/T5 establish alternative bidirectional and text-to-text transfer branches? +- How did empirical scaling laws and Chinchilla change how model size, data, and compute were allocated? +- How did sparse MoE, RMSNorm, SwiGLU, RoPE, and FlashAttention establish efficiency primitives later reused by open LLMs? +- How did instruction tuning, RLHF, and chain-of-thought prompting create the post-training and reasoning foundations of later systems? + +Editorial prose should distinguish first publication dates from later model availability and should not imply direct conceptual ancestry unless the evidence supports it. From 0e9639d742bfd2cfa2f5c1d0084859e128090d80 Mon Sep 17 00:00:00 2001 From: Alajmah <55954113+Alajmah@users.noreply.github.com> Date: Tue, 29 Sep 2026 14:49:53 +0300 Subject: [PATCH 2/2] Refine foundation naming and scale claims --- data/claims/chinchilla-parameter-count.yaml | 9 +++++++++ data/claims/gpt-1-release-of-gpt.yaml | 5 ++++- data/claims/gpt-2-parameter-count.yaml | 9 +++++++++ data/claims/gpt-2-succeeds-gpt-1.yaml | 9 +++++++++ data/claims/gpt-3-parameter-count.yaml | 9 +++++++++ data/claims/palm-parameter-count.yaml | 9 +++++++++ data/claims/switch-transformer-max-parameter-count.yaml | 9 +++++++++ data/relationships/gpt-2-succeeds-gpt-1.yaml | 7 +++++++ data/releases/gpt-1.yaml | 1 + 9 files changed, 66 insertions(+), 1 deletion(-) create mode 100644 data/claims/chinchilla-parameter-count.yaml create mode 100644 data/claims/gpt-2-parameter-count.yaml create mode 100644 data/claims/gpt-2-succeeds-gpt-1.yaml create mode 100644 data/claims/gpt-3-parameter-count.yaml create mode 100644 data/claims/palm-parameter-count.yaml create mode 100644 data/claims/switch-transformer-max-parameter-count.yaml create mode 100644 data/relationships/gpt-2-succeeds-gpt-1.yaml diff --git a/data/claims/chinchilla-parameter-count.yaml b/data/claims/chinchilla-parameter-count.yaml new file mode 100644 index 0000000..435e326 --- /dev/null +++ b/data/claims/chinchilla-parameter-count.yaml @@ -0,0 +1,9 @@ +id: claim:chinchilla-parameter-count +type: claim +subject: release:chinchilla +predicate: parameter_count +value: 70000000000 +status: confirmed_primary +evidence: + - source: source:chinchilla-paper + locator: abstract — Chinchilla is a 70B parameter model diff --git a/data/claims/gpt-1-release-of-gpt.yaml b/data/claims/gpt-1-release-of-gpt.yaml index c334a0f..23a4dae 100644 --- a/data/claims/gpt-1-release-of-gpt.yaml +++ b/data/claims/gpt-1-release-of-gpt.yaml @@ -6,4 +6,7 @@ object: family:gpt status: confirmed_primary evidence: - source: source:gpt1-openai-announcement - locator: OpenAI presents its generative pre-training Transformer approach + locator: OpenAI's 2018 generative pre-training Transformer model + - source: source:gpt2-openai-announcement + locator: GPT-2 is explicitly described as a successor to GPT +notes: GPT-1 is the retrospective label used here for the 2018 model. diff --git a/data/claims/gpt-2-parameter-count.yaml b/data/claims/gpt-2-parameter-count.yaml new file mode 100644 index 0000000..ca3cff2 --- /dev/null +++ b/data/claims/gpt-2-parameter-count.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-2-parameter-count +type: claim +subject: release:gpt-2 +predicate: parameter_count +value: 1500000000 +status: confirmed_primary +evidence: + - source: source:gpt2-openai-announcement + locator: GPT-2 is described as a 1.5 billion parameter Transformer language model diff --git a/data/claims/gpt-2-succeeds-gpt-1.yaml b/data/claims/gpt-2-succeeds-gpt-1.yaml new file mode 100644 index 0000000..6110a6d --- /dev/null +++ b/data/claims/gpt-2-succeeds-gpt-1.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-2-succeeds-gpt-1 +type: claim +subject: release:gpt-2 +predicate: succeeds +object: release:gpt-1 +status: confirmed_primary +evidence: + - source: source:gpt2-openai-announcement + locator: GPT-2 is described as a successor to GPT and a direct scale-up diff --git a/data/claims/gpt-3-parameter-count.yaml b/data/claims/gpt-3-parameter-count.yaml new file mode 100644 index 0000000..3b24ef0 --- /dev/null +++ b/data/claims/gpt-3-parameter-count.yaml @@ -0,0 +1,9 @@ +id: claim:gpt-3-parameter-count +type: claim +subject: release:gpt-3 +predicate: parameter_count +value: 175000000000 +status: confirmed_primary +evidence: + - source: source:gpt3-paper + locator: abstract — 175 billion parameters diff --git a/data/claims/palm-parameter-count.yaml b/data/claims/palm-parameter-count.yaml new file mode 100644 index 0000000..4677220 --- /dev/null +++ b/data/claims/palm-parameter-count.yaml @@ -0,0 +1,9 @@ +id: claim:palm-parameter-count +type: claim +subject: release:palm +predicate: parameter_count +value: 540000000000 +status: confirmed_primary +evidence: + - source: source:palm-paper + locator: abstract — 540-billion parameter densely activated Transformer language model diff --git a/data/claims/switch-transformer-max-parameter-count.yaml b/data/claims/switch-transformer-max-parameter-count.yaml new file mode 100644 index 0000000..46a639a --- /dev/null +++ b/data/claims/switch-transformer-max-parameter-count.yaml @@ -0,0 +1,9 @@ +id: claim:switch-transformer-max-parameter-count +type: claim +subject: release:switch-transformer +predicate: maximum_reported_parameter_count +value: 1000000000000 +status: reported_primary +evidence: + - source: source:switch-transformer-paper + locator: abstract — pre-training up to trillion-parameter models diff --git a/data/relationships/gpt-2-succeeds-gpt-1.yaml b/data/relationships/gpt-2-succeeds-gpt-1.yaml new file mode 100644 index 0000000..c2127b4 --- /dev/null +++ b/data/relationships/gpt-2-succeeds-gpt-1.yaml @@ -0,0 +1,7 @@ +id: relationship:gpt-2-succeeds-gpt-1 +type: relationship +subject: release:gpt-2 +predicate: succeeds +object: release:gpt-1 +claim_refs: + - claim:gpt-2-succeeds-gpt-1 diff --git a/data/releases/gpt-1.yaml b/data/releases/gpt-1.yaml index c81e2ae..d310abb 100644 --- a/data/releases/gpt-1.yaml +++ b/data/releases/gpt-1.yaml @@ -1,4 +1,5 @@ id: release:gpt-1 type: model_release name: GPT-1 +description: Retrospective name for OpenAI's 2018 generative pre-training Transformer model, referred to as GPT by the 2019 GPT-2 announcement. status: canonical