🤖 Autoupdate: Epoch (#5594)

Author: owidbot

etl/steps/data/garden/artificial_intelligence/2025-03-12/epoch.meta.yml

@@ -48,7 +48,7 @@ tables:
           zeroDay: '1949-01-01'
           yearIsDay: true
 
-      training_dataset_size__gradients:
+      training_dataset_size__total:
         title: Training dataset size
         unit: 'unique datapoints'
         description_short: The number of unique data points used to train the model. Each domain has a specific data point unit; for example, for vision it is images, for language it is words, and for games it is timesteps. This means systems can only be compared directly within the same domain.

etl/steps/data/garden/artificial_intelligence/2025-03-12/epoch.py

@@ -112,7 +112,7 @@ def simplify_entry(entry):
     tb["training_computation_petaflop"] = tb["training_compute__flop"] / 1e15
 
     # Convert training dataset size to numeric
-    tb["training_dataset_size__gradients"] = pd.to_numeric(tb["training_dataset_size__gradients"], errors="coerce")
+    tb["training_dataset_size__total"] = pd.to_numeric(tb["training_dataset_size__total"], errors="coerce")
 
     # Convert publication date to a datetime objects
     tb["publication_date"] = pd.to_datetime(tb["publication_date"])
@@ -142,7 +142,7 @@ def simplify_entry(entry):
     )
     tb = tb.format(["days_since_1949", "model"])
     # Add metadata to the publication date column
-    for col in ["publication_date", "training_dataset_size__gradients"]:
+    for col in ["publication_date", "training_dataset_size__total"]:
         tb[col].metadata.origins = tb["domain"].metadata.origins
 
     #

etl/steps/data/garden/artificial_intelligence/2025-03-12/epoch_aggregates_affiliation.py

@@ -30,7 +30,7 @@ def run() -> None:
     unused_columns = [
         "days_since_1949",
         "parameters",
-        "training_dataset_size__gradients",
+        "training_dataset_size__total",
         "domain",
         "training_computation_petaflop",
     ]

etl/steps/data/garden/artificial_intelligence/2025-03-12/epoch_aggregates_domain.py

@@ -40,7 +40,7 @@ def run() -> None:
         "organization_categorization",
         "parameters",
         "training_compute__flop",
-        "training_dataset_size__gradients",
+        "training_dataset_size__total",
         "notability_criteria",
     ]
     # Drop the unused columns

etl/steps/data/garden/artificial_intelligence/2025-03-12/epoch_regressions.py

@@ -91,7 +91,7 @@ def run_regression(tb):
         f"{DL_ERA_START}–{int(END_DATE)}": np.array([DL_ERA_START, END_DATE]),
     }
 
-    metrics = ["training_computation_petaflop", "parameters", "training_dataset_size__gradients"]
+    metrics = ["training_computation_petaflop", "parameters", "training_dataset_size__total"]
     new_tables = []
 
     for metric in metrics:

etl/steps/data/garden/artificial_intelligence/2026-01-30/frontiermath.py

@@ -11,7 +11,7 @@
 # Pre-compile regex patterns for performance
 PATTERNS = {
     "date": re.compile(r"(\d{4})-?(\d{2})-?(\d{2})"),
-    "context_size": re.compile(r"\d+K$"),
+    "context_size": re.compile(r"(\d+K|max)$"),
     "claude_numeric": re.compile(r"claude-(\d+)-(\d+)-(\w+)", re.IGNORECASE),
     "claude_variant": re.compile(r"claude-(\w+)-(\d+(?:\.\d+)?)", re.IGNORECASE),
     "gpt_version": re.compile(r"gpt-([\d.]+)(?:-(\w+))?", re.IGNORECASE),

etl/steps/data/grapher/artificial_intelligence/2025-03-12/epoch.py

@@ -28,7 +28,7 @@ def run() -> None:
     # For visualization purposes I am adding the rows with the maximum values of compute, data, and parameters in each year to the table as a separate "model". I don't want to do this in garden as it'd affect other datasets that depend on this one.
     columns = {
         "training_computation_petaflop": "compute",
-        "training_dataset_size__gradients": "data",
+        "training_dataset_size__total": "data",
         "parameters": "parameters",
     }
     # Find maximum values for a given column (compute, data, params) per year, label them, and add summary rows.

etl/steps/data/meadow/artificial_intelligence/2025-03-12/epoch.py

@@ -36,7 +36,7 @@ def run() -> None:
         "Publication date",
         "Parameters",
         "Training compute (FLOP)",
-        "Training dataset size (gradients)",
+        "Training dataset size (total)",
         "Notability criteria",
     ]
 

etl/steps/data/meadow/artificial_intelligence/2025-10-10/epoch_gpus.py

@@ -28,7 +28,7 @@ def run() -> None:
         "FP32 (single precision) performance (FLOP/s)",
         "FP16 (half precision) performance (FLOP/s)",
         "INT8 performance (OP/s)",
-        "Memory size per board (Byte)",
+        "Memory (bytes)",
         "Memory bandwidth (byte/s)",
         "TDP (W)",
         "Process size (nm)",

snapshots/artificial_intelligence/2025-03-12/epoch.csv.dvc

@@ -30,13 +30,13 @@ meta:
     # Files
     url_main: https://epoch.ai/mlinputs/visualization
     url_download: https://epoch.ai/data/epochdb/notable_ai_models.csv
-    date_accessed: '2026-01-26'
+    date_accessed: '2026-02-27'
 
     # License
     license:
       name: CC BY 4.0
       url: https://creativecommons.org/licenses/by/4.0/
 outs:
-  - md5: 1149d0b75df9f65144cdbe1f881353e1
-    size: 2101444
+  - md5: e387c13e7a6e35d505ced9a0a1d968f2
+    size: 2107224
     path: epoch.csv