Athithya117 commited on
Commit
2b94875
·
verified ·
1 Parent(s): 04b8ffe

medium: scale snapshot

Browse files
DATA_SOURCES.md ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Upstream sources and attribution
2
+
3
+ | Directory | Upstream | License | Changes applied |
4
+ |---|---|---|---|
5
+ | corpus/opinions.parquet | HF: docketx/us-caselaw | public-domain-edict-of-government | Streamed, capped at scale preset, deduplicated on cl_opinion_id |
6
+ | graph/citations_filtered.parquet | CourtListener citation-map-2026-06-30 | Public Domain Mark 1.0 | bz2 -> Parquet -> filtered to edges where both endpoints are in the corpus |
7
+ | metadata/clusters_filtered.parquet | CourtListener opinion-clusters-2026-06-30 | Public Domain Mark 1.0 | bz2 -> clean Parquet (rows with non-integer `id` dropped) -> filtered to corpus clusters |
8
+ | eval_isolated/hojicha/ | HF: feyninc/hojicha | CC BY 4.0 | Unmodified, exported to parquet |
9
+ | eval_isolated/synthesized_graph_qa/ | derived from graph/citations_filtered.parquet | CC0 1.0 (derived) | 200 citation-lookup queries; ground truth = citing set |
10
+
11
+ Retrieved: 2026-11-02. CourtListener bulk snapshot: 2026-06-30.
12
+
13
+ ## Schema notes
14
+
15
+ `court_id` is **not** on `opinion-clusters`. CourtListener's hierarchy is:
16
+
17
+ opinions.cluster_id -> clusters.id
18
+ clusters.docket_id -> dockets.id
19
+ dockets.court_id -> courts.id
20
+
21
+ Court metadata is taken from the corpus (`court`, `jurisdiction`) rather than
22
+ from the clusters export, so no `dockets` download is required.
LICENSE ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ CC0 1.0 Universal (CC0 1.0) Public Domain Dedication
2
+
3
+ - docketx/us-caselaw corpus: public-domain-edict-of-government
4
+ - CourtListener bulk data: Public Domain Mark 1.0
5
+ - feyninc/hojicha: CC BY 4.0 (retained as-is in eval_isolated/hojicha/)
6
+
7
+ https://creativecommons.org/publicdomain/zero/1.0/
MANIFEST.json ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "meta": {
3
+ "scale": "medium",
4
+ "corpus_cap": 100000,
5
+ "generated_at": "2026-11-02T13:21:10.495680+00:00",
6
+ "bulk_snapshot": "2026-06-30",
7
+ "graph_summary": {
8
+ "edges": 426462,
9
+ "clusters": 92157,
10
+ "eval_queries": 200
11
+ },
12
+ "source_versions": {
13
+ "corpus": "docketx/us-caselaw@streamed",
14
+ "graph": "courtlistener/citation-map-2026-06-30",
15
+ "metadata": "courtlistener/opinion-clusters-2026-06-30",
16
+ "hojicha": "feyninc/hojicha@latest",
17
+ "synthesized_eval": "derived from graph"
18
+ }
19
+ },
20
+ "files": {
21
+ "MANIFEST.json": {
22
+ "sha256": "0a93fd0f1b4c4718c14d11628809e46ac1680a08e316c09e740bda1352f85f9c",
23
+ "bytes": 1630
24
+ },
25
+ "eval_isolated/hojicha/corpus.parquet": {
26
+ "sha256": "c7d8267356b168dcd23ca98d4c88fffc02cd65d447b334ce3c2a8958d0fc44a7",
27
+ "bytes": 4034847,
28
+ "rows": 194
29
+ },
30
+ "eval_isolated/hojicha/questions.parquet": {
31
+ "sha256": "65f45d09779e64dfbc84f8ef6a8c00dc1b4999ce891ba3670a1e4190a6e57592",
32
+ "bytes": 306494,
33
+ "rows": 1568
34
+ },
35
+ "eval_isolated/synthesized_graph_qa/queries.parquet": {
36
+ "sha256": "c98c3486bd66b625b56ed131fb6026c0a011f0a46309f40eb6c65aa8bd1d02b3",
37
+ "bytes": 124753,
38
+ "rows": 200
39
+ },
40
+ "graph/citations_filtered.parquet": {
41
+ "sha256": "e67edce85bd837758679613bc4cb06b35555e1ebc571fd27b836d4443fa3e122",
42
+ "bytes": 2672347,
43
+ "rows": 426462
44
+ },
45
+ "metadata/clusters_filtered.parquet": {
46
+ "sha256": "f90ff6a7fd6d398ccb14de4ad1aa54378c34724797c3b346538512c990b1adb9",
47
+ "bytes": 3936993,
48
+ "rows": 92157
49
+ },
50
+ "corpus/opinions.parquet": {
51
+ "sha256": "385a567be8286b19c52bd51a837784e5ffd2f8da05ed56902b1e04255eda06d8",
52
+ "bytes": 418034116,
53
+ "rows": 100000
54
+ }
55
+ }
56
+ }
README.md ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc0-1.0
3
+ language:
4
+ - en
5
+ tags:
6
+ - legal
7
+ - us-case-law
8
+ - rag
9
+ - graph-rag
10
+ - citation-graph
11
+ - adaptive-retrieval
12
+ size_categories:
13
+ - 1K<n<10K
14
+ configs:
15
+ - config_name: corpus_opinions
16
+ data_files:
17
+ - split: train
18
+ path: corpus/opinions.parquet
19
+ - config_name: graph_citations
20
+ data_files:
21
+ - split: train
22
+ path: graph/citations_filtered.parquet
23
+ - config_name: metadata_clusters
24
+ data_files:
25
+ - split: train
26
+ path: metadata/clusters_filtered.parquet
27
+ - config_name: eval_hojicha_corpus
28
+ data_files:
29
+ - split: train
30
+ path: eval_isolated/hojicha/corpus.parquet
31
+ - config_name: eval_hojicha_questions
32
+ data_files:
33
+ - split: train
34
+ path: eval_isolated/hojicha/questions.parquet
35
+ - config_name: eval_synthesized_graph_qa
36
+ data_files:
37
+ - split: train
38
+ path: eval_isolated/synthesized_graph_qa/queries.parquet
39
+ ---
40
+
41
+ # adaptive-graphrag-datasets
42
+
43
+ US CourtListener GraphRAG dataset snapshot (scale=medium).
44
+
45
+ See DATA_SOURCES.md for attribution and MANIFEST.json for SHA-256.
46
+
47
+ ## Scale presets
48
+
49
+ | Scale | Corpus rows | Approx. vectors (@ ~5 chunks/doc) |
50
+ |---|---|---|
51
+ | min | 5,000 | ~25K |
52
+ | medium | 100,000 | ~500K |
53
+ | max | 400,000 | ~2M |
54
+
55
+ Graph and metadata are filtered to the sampled corpus, so they scale with it.
56
+
57
+ ## Join keys
58
+
59
+ corpus.opinions.cl_opinion_id <-> graph.citations.citing_opinion_id
60
+ corpus.opinions.cl_opinion_id <-> graph.citations.cited_opinion_id
61
+ corpus.opinions.cluster_id <-> metadata.clusters.id
62
+
63
+ metadata.clusters.docket_id -> dockets.id (not shipped)
64
+ dockets.court_id -> courts.id (court metadata is on corpus)
65
+
66
+ ## Quick start
67
+
68
+ from huggingface_hub import snapshot_download
69
+ snapshot_download(
70
+ repo_id="<your-username>/adaptive-graphrag-datasets-max",
71
+ repo_type="dataset",
72
+ local_dir="data/raw",
73
+ )
corpus/opinions.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:385a567be8286b19c52bd51a837784e5ffd2f8da05ed56902b1e04255eda06d8
3
+ size 418034116
eval_isolated/hojicha/corpus.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7d8267356b168dcd23ca98d4c88fffc02cd65d447b334ce3c2a8958d0fc44a7
3
+ size 4034847
eval_isolated/hojicha/questions.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:65f45d09779e64dfbc84f8ef6a8c00dc1b4999ce891ba3670a1e4190a6e57592
3
+ size 306494
eval_isolated/synthesized_graph_qa/queries.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c98c3486bd66b625b56ed131fb6026c0a011f0a46309f40eb6c65aa8bd1d02b3
3
+ size 124753
graph/citation-map-2026-06-30.csv.bz2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e51bfd93fbbffdd9eb1289971419ccffa3a42023205f50aa40abe28e7e7f4ae6
3
+ size 526206545
graph/citations_filtered.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e67edce85bd837758679613bc4cb06b35555e1ebc571fd27b836d4443fa3e122
3
+ size 2672347
metadata/clusters_filtered.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f90ff6a7fd6d398ccb14de4ad1aa54378c34724797c3b346538512c990b1adb9
3
+ size 3936993
metadata/opinion-clusters-2026-06-30.csv.bz2 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bddbd5ca78377ef2fe56ed2ef506e12afe291b6fc7fc31fac6721f4de33b182f
3
+ size 2457231057