Alajmah · Sep 29, 2026 · Sep 29, 2026
diff --git 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
@@ -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
@@ -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
@@ -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
@@ -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-parameter-count.yaml 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/chinchilla-published-by-deepmind.yaml 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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -0,0 +1,12 @@
+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'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-1-uses-generative-pretraining.yaml 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
@@ -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-parameter-count.yaml 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-published-by-openai.yaml 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
@@ -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-succeeds-gpt-1.yaml 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-2-uses-generative-pretraining.yaml 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
@@ -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-parameter-count.yaml 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/gpt-3-published-by-openai.yaml 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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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
@@ -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-parameter-count.yaml 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/palm-published-by-google.yaml 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
@@ -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
@@ -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-max-parameter-count.yaml 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/claims/switch-transformer-published-by-google.yaml 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
@@ -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
@@ -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
@@ -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