Dataset Viewer
Auto-converted to Parquet Duplicate
complete
bool
1 class
files_total
int64
1
1.56k
files_processed
int64
1
1.56k
skipped_files
int64
0
0
error_files
int64
0
0
chunks_stored
int64
196k
598k
timestamp
stringdate
2026-08-08 23:31:26
2026-08-30 20:32:40
true
1,557
1,557
0
0
196,211
2026-08-10T11:38:08.620116
true
1
1
0
0
196,210
2026-08-22T15:15:46.687008
true
1,557
1,557
0
0
196,211
2026-08-10T11:38:08.620076
true
1,557
1,557
0
0
196,211
2026-08-10T11:38:08.619956
true
1
1
0
0
230,024
2026-08-09T00:21:21.140358
true
1
1
0
0
230,024
2026-08-22T04:23:12.683159
true
1
1
0
0
230,024
2026-08-09T00:15:36.558553
true
1
1
0
0
230,024
2026-08-09T00:10:17.484508
true
1,557
1,557
0
0
598,191
2026-08-10T11:17:53.011342
true
1
1
0
0
598,191
2026-08-22T04:48:10.777994
true
1,557
1,557
0
0
598,191
2026-08-30T19:51:33.973944
true
1,557
1,557
0
0
598,191
2026-08-10T11:17:53.011192
true
1,557
1,557
0
0
205,097
2026-08-10T10:29:21.447180
true
1
1
0
0
205,097
2026-08-22T04:33:28.144567
true
1,557
1,557
0
0
205,097
2026-08-10T10:29:21.447142
true
1,557
1,557
0
0
205,097
2026-08-10T10:29:21.447020
true
1,557
1,557
0
0
344,285
2026-08-10T10:08:14.931440
true
1
1
0
0
344,284
2026-08-22T15:10:55.016117
true
1,557
1,557
0
0
344,285
2026-08-10T10:08:14.931403
true
1,557
1,557
0
0
344,285
2026-08-10T10:08:14.931286
true
1,557
1,557
0
0
196,211
2026-08-10T09:38:11.692207
true
1
1
0
0
196,210
2026-08-22T15:01:42.072907
true
1,557
1,557
0
0
196,211
2026-08-10T09:38:11.692168
true
1,557
1,557
0
0
196,211
2026-08-10T09:38:11.692041
true
1
1
0
0
230,024
2026-08-09T00:04:46.226760
true
1
1
0
0
230,024
2026-08-22T03:47:47.536084
true
1
1
0
0
230,024
2026-08-08T23:59:00.884288
true
1
1
0
0
230,024
2026-08-08T23:53:40.435138
true
1,557
1,557
0
0
598,191
2026-08-10T09:18:48.982547
true
1
1
0
0
598,191
2026-08-22T04:12:58.262633
true
1,557
1,557
0
0
598,191
2026-08-30T19:34:27.643210
true
1,557
1,557
0
0
598,191
2026-08-10T09:18:48.982380
true
1,557
1,557
0
0
205,097
2026-08-09T23:05:48.954255
true
1
1
0
0
205,097
2026-08-22T03:58:09.876139
true
1,557
1,557
0
0
205,097
2026-08-09T23:05:48.954218
true
1,557
1,557
0
0
205,097
2026-08-09T23:05:48.954097
true
1,557
1,557
0
0
344,285
2026-08-09T22:43:52.986213
true
1
1
0
0
344,284
2026-08-22T14:56:49.532401
true
1,557
1,557
0
0
344,285
2026-08-09T22:43:52.986173
true
1,557
1,557
0
0
344,285
2026-08-09T22:43:52.986044
true
1,557
1,557
0
0
196,211
2026-08-09T20:49:10.944972
true
1
1
0
0
196,210
2026-08-22T14:38:03.133665
true
1,557
1,557
0
0
196,211
2026-08-09T20:49:10.944934
true
1,557
1,557
0
0
196,211
2026-08-09T20:49:10.944805
true
1
1
0
0
230,024
2026-08-08T23:36:54.827558
true
1
1
0
0
230,024
2026-08-22T03:03:21.049137
true
1
1
0
0
230,024
2026-08-08T23:33:59.345281
true
1
1
0
0
230,024
2026-08-08T23:31:26.964069
true
1,557
1,557
0
0
598,191
2026-08-09T20:36:55.376806
true
1
1
0
0
598,191
2026-08-22T03:15:51.666963
true
1,557
1,557
0
0
598,191
2026-08-30T19:05:33.872715
true
1,557
1,557
0
0
598,191
2026-08-09T20:36:55.376654
true
1,557
1,557
0
0
205,097
2026-08-09T20:09:52.839164
true
1
1
0
0
205,097
2026-08-22T03:08:13.023039
true
1,557
1,557
0
0
205,097
2026-08-09T20:09:52.839122
true
1,557
1,557
0
0
205,097
2026-08-09T20:09:52.838999
true
1,557
1,557
0
0
344,285
2026-08-09T19:57:46.119013
true
1
1
0
0
344,284
2026-08-22T14:35:29.154677
true
1,557
1,557
0
0
344,285
2026-08-09T19:57:46.118975
true
1,557
1,557
0
0
344,285
2026-08-09T19:57:46.118860
true
1,557
1,557
0
0
196,211
2026-08-09T22:13:42.781133
true
1
1
0
0
196,210
2026-08-22T14:47:37.836766
true
1,557
1,557
0
0
196,211
2026-08-09T22:13:42.781093
true
1,557
1,557
0
0
196,211
2026-08-09T22:13:42.780967
true
1
1
0
0
230,024
2026-08-08T23:48:06.706674
true
1
1
0
0
230,024
2026-08-22T03:22:26.203434
true
1
1
0
0
230,024
2026-08-08T23:44:15.175015
true
1
1
0
0
230,024
2026-08-08T23:40:42.834058
true
1,557
1,557
0
0
598,191
2026-08-09T21:59:15.389679
true
1
1
0
0
598,191
2026-08-22T03:38:51.291645
true
1,557
1,557
0
0
598,191
2026-08-30T19:17:16.846022
true
1,557
1,557
0
0
598,191
2026-08-09T21:59:15.389530
true
1,557
1,557
0
0
205,097
2026-08-09T21:25:37.781937
true
1
1
0
0
205,097
2026-08-22T03:29:11.851332
true
1,557
1,557
0
0
205,097
2026-08-09T21:25:37.781899
true
1,557
1,557
0
0
205,097
2026-08-09T21:25:37.781789
true
1,557
1,557
0
0
344,285
2026-08-09T21:10:36.982130
true
1
1
0
0
344,284
2026-08-22T14:44:21.049559
true
1,557
1,557
0
0
344,285
2026-08-09T21:10:36.982093
true
1,557
1,557
0
0
344,285
2026-08-09T21:10:36.981983
true
1,557
1,557
0
0
196,211
2026-08-10T16:56:20.309218
true
1
1
0
0
196,210
2026-08-22T15:52:37.568916
true
1,557
1,557
0
0
196,211
2026-08-10T16:56:20.309179
true
1,557
1,557
0
0
196,211
2026-08-10T16:56:20.309061
true
1
1
0
0
230,024
2026-08-09T01:05:57.357655
true
1
1
0
0
230,024
2026-08-22T05:27:40.163728
true
1
1
0
0
230,024
2026-08-09T00:54:30.141018
true
1
1
0
0
230,024
2026-08-09T00:43:29.326426
true
1,557
1,557
0
0
598,191
2026-08-10T16:19:47.641282
true
1
1
0
0
598,191
2026-08-22T06:14:32.402676
true
1,557
1,557
0
0
598,191
2026-08-30T20:32:41.653860
true
1,557
1,557
0
0
598,191
2026-08-10T16:19:47.641130
true
1,557
1,557
0
0
205,097
2026-08-10T14:51:42.061530
true
1
1
0
0
205,097
2026-08-22T05:49:27.889661
true
1,557
1,557
0
0
205,097
2026-08-10T14:51:42.061486
true
1,557
1,557
0
0
205,097
2026-08-10T14:51:42.061346
true
1,557
1,557
0
0
344,285
2026-08-10T14:07:08.893840
true
1
1
0
0
344,284
2026-08-22T15:42:59.335924
true
1,557
1,557
0
0
344,285
2026-08-10T14:07:08.893801
true
1,557
1,557
0
0
344,285
2026-08-10T14:07:08.893684
End of preview. Expand in Data Studio

Telco-Retrieve Chunks — 3GPP Rel-19 Vector Database Grid

120 pre-built retrieval indexes over the 3GPP Release-19 corpus — one per (embedding model × chunking strategy × enrichment strategy) configuration. Each is a complete ChromaDB dense index plus a parallel SQLite FTS5 (BM25) index, ready to query with no re-ingestion.

Part of the GSMA Open Telco AI initiative. Companion to GSMA/telco-retrieve-QnA (the evaluation questions) and the open-telco-rag pipeline that reads these indexes directly.

~1.05 TB total. Individual configs range from ~1 GB to ~25 GB. Download the one config you need, not the repo.

Dataset Summary

Ingesting a large standards corpus into a vector store is the expensive, non-reproducible step in a RAG benchmark. This dataset ships that step's output for every point in a 120-cell grid so that retrieval and generation can be evaluated identically across configurations, on any machine, with no GPU and no ingestion run.

Each configuration was produced by open-telco-rag ingest 3gpp and pushed with push-hf.

The grid (6 × 5 × 4 = 120)

Axis Values
Embedding model (6) minilm (all-MiniLM-L6-v2, 384d) · mpnet (all-mpnet-base-v2, 768d) · e5 (intfloat/e5-large-v2, 1024d) · bge (BAAI/bge-large-en-v1.5, 1024d) · otel-109m (OTel-Embedding-109M, telecom fine-tuned, 768d) · otel-0.6b (OTel-Embedding-0.6B, telecom fine-tuned, 1024d)
Chunking strategy (5) text_baseline (1024 char / 200 overlap) · sliding_window_tokens (512 tok / 50) · parent_child (2048 parent / 512 child) · hierarchical_markdown (2048 / 200) · lumber_chunker (LLM semantic chunking, t<theta>)
Enrichment strategy (4) none · metadata_tagging · acronym_expansion · llm_metadata (LLM-generated per-chunk metadata)

Repository Structure

One folder per configuration, named:

3gpp-r19_<model>_<chunking>_<enrichment>_<params>/

Examples:

  • 3gpp-r19_bge_text_baseline_none_c1024_o200/
  • 3gpp-r19_otel-0.6b_parent_child_llm_metadata_p2048_c512/
  • 3gpp-r19_minilm_lumber_chunker_none_t550/

Inside each folder:

Path What
chroma.sqlite3 ChromaDB metadata + document store
<uuid>/data_level0.bin, header.bin, length.bin, link_lists.bin, index_metadata.pickle HNSW dense index for the collection
<collection>_fts.db (-wal, -shm) SQLite FTS5 BM25 index (parallel keyword search)
_ingest_complete.json Per-config ingestion manifest — complete, files_total, files_processed, skipped_files, error_files, chunks_stored, timestamp

Dataset viewer note: Hugging Face auto-builds the viewer from the 120 _ingest_complete.json files, so it shows a 120-row ingestion-manifest table, not the chunk text. The chunk text lives inside chroma.sqlite3 / the FTS DBs and is reached through the retrieval library.

Usage

pip install "open-telco-rag[local-embed]"

# pull one configuration
open-telco-rag pull-hf GSMA/telco-retrieve-chunks \
  3gpp-r19_bge_parent_child_llm_metadata_p2048_c512 --local-dir ./db

# query it — dense, BM25, or hybrid — no ingestion needed
open-telco-rag search "handover procedure for conditional PSCell change" \
  --source-type 3gpp --db-path ./db --top-k 5

# or run the full evaluation against GSMA/telco-retrieve-QnA
open-telco-rag evaluate 3gpp --db-path ./db

Raw access without the library: open chroma.sqlite3 with the chromadb client, or query <collection>_fts.db with any SQLite client (FTS5 MATCH).

Source Data & Curation

  • Corpus: 3GPP specifications, Release 19 (all series), the same underlying text as GSMA/telco-retrieve-QnA. Sourced via the TSpec-LLM dataset and the 3GPP spec archive (3gpp.org/ftp/Specs/archive).
  • Pipeline: open-telco-rag — per-SDO source loader → clean → chunk (per strategy) → enrich (per strategy) → embed (per model) → ChromaStore + SQLite FTS5. See the open-telco-rag README for stage details.
  • Curation rationale: a fully-crossed grid lets an evaluation isolate the effect of each axis (embedding model, chunking, enrichment) on retrieval quality, holding the corpus and question set fixed — which is the point of the companion GSMA/telco-retrieve-QnA benchmark.

Considerations & Limitations

  • Scale. ~1.05 TB. Pull individual config folders; never clone the whole repo.
  • Single corpus / single release. 3GPP Rel-19 only. No other SDOs, no other releases.
  • Point-in-time. Reflects Rel-19 spec text as ingested in August 2026.
  • Reproducibility, not novelty. These are derived indexes; the value is identical retrieval state across configs, not new source data.
  • Format lock-in. ChromaDB + SQLite-FTS5 on-disk formats; readable via open-telco-rag or the chromadb / sqlite3 clients, not as plain Parquet/JSONL.
  • .db-wal / .db-shm sidecar files may be present; SQLite recreates them on open, so a consumer can ignore them.

Licensing

  • Index artifacts and manifests: Apache-2.0, consistent with the open-telco-rag toolkit. (Set here as apache-2.0; change if GSMA requires a different licence for this repo.)
  • Stored chunk text (inside chroma.sqlite3 / the FTS DBs) consists of excerpts of 3GPP specifications. That text remains subject to 3GPP's copyright and redistribution terms and those of the upstream source (the TSpec-LLM dataset / the 3GPP spec archive). Downstream users are responsible for compliance.
Downloads last month
58