Expose root tables as named dataset configs

README.md

@@ -9,6 +9,20 @@ tags:
   - pretraining
   - v-jepa
   - deduplication
+configs:
+  - config_name: samples
+    data_files:
+      - split: train
+        path: "samples/**/*.parquet"
+    default: true
+  - config_name: assets
+    data_files:
+      - split: train
+        path: "assets/**/*.parquet"
+  - config_name: memberships
+    data_files:
+      - split: train
+        path: "memberships/**/*.parquet"
 ---
 
 # V-JEPA Reproduce Recipe
@@ -39,14 +53,13 @@ remains reachable.
 ## Layout
 
 ```text
-CURRENT.json
-runs/<run_id>/manifest.json
-runs/<run_id>/samples/dedup_bucket=00..ff/*.parquet
-runs/<run_id>/assets/dedup_bucket=00..ff/*.parquet
-runs/<run_id>/memberships/dataset_id=<name>/dedup_bucket=00..ff/*.parquet
-runs/<run_id>/dataset_roles.parquet
-runs/<run_id>/dataset_stats.parquet
-runs/<run_id>/source_coverage.parquet
+manifest.json
+samples/dedup_bucket=00..ff/*.parquet
+assets/dedup_bucket=00..ff/*.parquet
+memberships/dataset_id=<name>/dedup_bucket=00..ff/*.parquet
+dataset_roles.parquet
+dataset_stats.parquet
+source_coverage.parquet
 ```
 
 - `samples`: exact `(platform, video_id, millisecond interval)` identities.
@@ -69,9 +82,10 @@ from this release.
 ## Load with Hugging Face Datasets
 
 Authentication is required because the repository is private; export a token
-with read access as `HF_TOKEN`. Bare `load_dataset(repo_id)` is unsupported
-because the repository intentionally contains tables with different schemas.
-Always select explicit `data_files`.
+with read access as `HF_TOKEN`. The repository exposes three named
+configurations. `samples` is the default, so both the bare call and an explicit
+`"samples"` select the deduplicated interval table. Use `"assets"` or
+`"memberships"` for the other schemas.
 
 Assign the 256 hash buckets across workers so each worker opens only its own
 shards:
@@ -80,21 +94,7 @@ shards:
 from datasets import load_dataset
 
 repo_id = "BootsofLagrangian/V-JEPA-Reproduce-Recipe"
-revision = "e3b4f02a422f760995406a0b8e1221e3882ec7d7"
-run_id = "823483f781094a668cf6682828723d16"
-bucket = "00"
-
-samples = load_dataset(
-    "parquet",
-    data_files={
-        "train": (
-            f"hf://datasets/{repo_id}@{revision}/runs/{run_id}/samples/"
-            f"dedup_bucket={bucket}/*.parquet"
-        )
-    },
-    split="train",
-    streaming=True,
-)
+samples = load_dataset(repo_id, "samples", split="train", streaming=True)
 
 web_train = samples.filter(
     lambda row: (
@@ -105,7 +105,9 @@ web_train = samples.filter(
 )
 ```
 
-For source lineage, load only the required dataset partition:
+For source lineage, load the memberships configuration and filter by
+`dataset_id`, or address a source partition directly when minimizing shard
+discovery matters:
 
 ```python
 dataset_id = "howto100m"
@@ -113,7 +115,7 @@ memberships = load_dataset(
     "parquet",
     data_files={
         "train": (
-            f"hf://datasets/{repo_id}@{revision}/runs/{run_id}/memberships/"
+            f"hf://datasets/{repo_id}/memberships/"
             f"dataset_id={dataset_id}/**/*.parquet"
         )
     },
@@ -122,6 +124,12 @@ memberships = load_dataset(
 )
 ```
 
+`revision=` is optional for ordinary use because `main` is the current release.
+For reproducible training, pass the immutable data commit shown below to
+`load_dataset(repo_id, "samples", ..., revision=revision)` or add
+`@<revision>` to an `hf://` URL. The manifest run ID remains inside
+`manifest.json` as provenance; it is not part of the Hub path.
+
 ## Exact-dedup boundary
 
 This release removes exact repeated memberships at identical platform, video
@@ -132,10 +140,10 @@ dataset membership.
 
 ## Publication verification
 
-Data commit `e3b4f02a422f760995406a0b8e1221e3882ec7d7` is the immutable revision for
-this release. The upload was checked as 3,294 data objects totaling
-41,557,605,260 bytes including the two JSON files: 3,292 Parquet shards plus
-`manifest.json` and `CURRENT.json`. Every local object size and SHA-256 matched
+Data commit `ROOT_DATA_COMMIT_PENDING` is the immutable revision for this
+release. The upload was checked as 3,293 data objects totaling
+41,557,605,070 bytes: 3,292 Parquet shards plus `manifest.json`. Every local
+object size and SHA-256 matched
 its remote Hub object. Pinned streaming loads each returned a row from one
 representative `samples` shard, one `assets` shard, and one `memberships`
 partition.