farbodtavakkoli commited on
Commit
feb9388
·
verified ·
1 Parent(s): a45d4d5

Copy of farbodtavakkoli/OTel-Reranker at a10a437

Browse files
Files changed (4) hide show
  1. .gitattributes +2 -0
  2. OTel-Reranker-Data.jsonl +3 -0
  3. README.md +104 -0
  4. croissant.json +190 -0
.gitattributes CHANGED
@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ rod_farbod_merged_with_qwen3_reranker_negs_with_answer_field_deduped_upload.jsonl filter=lfs diff=lfs merge=lfs -text
62
+ OTel-Reranker-Data.jsonl filter=lfs diff=lfs merge=lfs -text
OTel-Reranker-Data.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e263512521c43e834013a022fb0fa8c30fc18399c1bf775a7ecc57a39f808ca
3
+ size 1030885989
README.md ADDED
@@ -0,0 +1,104 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ task_categories:
4
+ - text-classification
5
+ language:
6
+ - en
7
+ tags:
8
+ - telecommunications
9
+ - telecom
10
+ - 3gpp
11
+ - oran
12
+ - gsma
13
+ - rag
14
+ - reranker
15
+ - cross-encoder
16
+ - retrieval
17
+ - domain-specific
18
+ - otel
19
+ size_categories:
20
+ - 100K<n<1M
21
+ pretty_name: OTel-Reranker
22
+ dataset_info:
23
+ features:
24
+ - name: sentence_0
25
+ dtype: string
26
+ - name: sentence_1
27
+ dtype: string
28
+ - name: label
29
+ dtype: float64
30
+ ---
31
+
32
+ # OTel-Reranker
33
+
34
+ ## Dataset Summary
35
+
36
+ OTel-Reranker is a pointwise reranking dataset for training cross-encoder models that score the relevance of a passage to a given query. It is part of the Open Telco (OTel) AI project, the largest open-source AI initiative in telecommunications, curated by over 100 domain experts from industry and academia.
37
+
38
+ Each sample consists of a query-passage pair and a continuous relevance score. The dataset is designed to train cross-encoder reranker models that refine the candidate set returned by a first-stage retriever, forming the reranking component of end-to-end RAG pipelines in telecommunications.
39
+
40
+ ## Dataset Structure
41
+
42
+ Each sample contains the following fields:
43
+
44
+ | Field | Type | Description |
45
+ |---|---|---|
46
+ | `sentence_0` | string | The query or information need. |
47
+ | `sentence_1` | string | A candidate passage to be scored for relevance to the query. |
48
+ | `label` | float | A continuous relevance score indicating how well the passage addresses the query. Higher values indicate greater relevance. |
49
+
50
+ The continuous label formulation supports pointwise reranking objectives, where the model learns to predict a scalar relevance score for each query-passage pair independently.
51
+
52
+ ## Data Sources and Contributor Attribution
53
+
54
+ The OTel corpus was curated by 100+ domain experts from industry and academia. Each source class was contributed by a specific institutional partner:
55
+
56
+ | Source | Contributor | Raw samples |
57
+ |---|---|---:|
58
+ | arXiv telecom papers, 3GPP standards, telecom Wikipedia, telecom Common Crawl pages | Yale University | 681,172 |
59
+ | GSMA Permanent Reference Documents (PRDs), Discover portal, mixed telecom documents | GSMA | 158,006 |
60
+ | IETF RFC series | NetoAI | 100,751 |
61
+ | Industry whitepapers | Khalifa University | 62,000 |
62
+ | O-RAN specifications across working groups 1, 2, 4, 5, 6, 7, 8, 9, 10 | University of Leeds | 58,565 |
63
+ | O-RAN documents across working groups | The University of Texas at Dallas | 42,000 |
64
+ | **Total raw samples** | | **~1,102,494** |
65
+
66
+ After the four-stage cleaning pipeline (described below), the corpus was reduced to **326,767** higher-confidence examples across the OTel dataset family.
67
+
68
+ ## Data Collection and Processing
69
+
70
+ The raw corpus comprised approximately 1.1 million samples. A four-stage cleaning pipeline reduced this to 326K samples across the OTel dataset family:
71
+
72
+ 1. **Heuristic filtering**: Rule-based removal of malformed, duplicated, or low-quality entries.
73
+ 2. **Reranking filtering**: Relevance scoring with a cross-encoder reranker to discard weakly aligned query-passage pairs.
74
+ 3. **Embedding filtering**: Semantic quality assessment with a strong embedding model to remove outliers and near-duplicates in embedding space.
75
+ 4. **Deduplication**: Final deduplication pass to ensure sample uniqueness.
76
+
77
+ Relevance labels were derived from the data generation and filtering pipeline, producing continuous scores that capture graded relevance rather than binary judgments.
78
+
79
+ ## Intended Use
80
+
81
+ OTel-Reranker is intended for training cross-encoder reranker models for telecom-domain retrieval. Target applications include:
82
+
83
+ - Fine-tuning cross-encoder models for second-stage reranking in telecom RAG pipelines.
84
+ - Improving retrieval precision by rescoring candidate passages returned by a bi-encoder first-stage retriever (e.g., a model trained on OTel-Embedding).
85
+ - Research on domain-specific reranking and its effect on downstream generation quality in RAG systems.
86
+
87
+ ## Limitations
88
+
89
+ - **Domain scope**: The dataset covers telecommunications standards and specifications. Cross-encoder models trained on this data are optimized for telecom relevance scoring and may not generalize to other domains.
90
+ - **Language**: English only.
91
+ - **Label granularity**: Relevance scores are derived from the automated pipeline and may not perfectly correspond to expert human judgments in all cases.
92
+ - **Temporal coverage**: The source documents reflect standards and specifications available at the time of collection. Newer revisions may not be represented.
93
+
94
+ ## Citation
95
+
96
+ ```bibtex
97
+ @misc{otel_data_2026,
98
+ title = {{OTel}: Open Telco AI Datasets, Benchmarks, and Models},
99
+ author = {Tavakkoli, Farbod and others},
100
+ year = {2026},
101
+ note = {Open Telco (OTel) dataset release},
102
+ url = {https://huggingface.co/farbodtavakkoli}
103
+ }
104
+ ```
croissant.json ADDED
@@ -0,0 +1,190 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "@context": {
3
+ "@language": "en",
4
+ "@vocab": "https://schema.org/",
5
+ "arrayShape": "cr:arrayShape",
6
+ "citeAs": "cr:citeAs",
7
+ "column": "cr:column",
8
+ "conformsTo": "dct:conformsTo",
9
+ "containedIn": "cr:containedIn",
10
+ "cr": "http://mlcommons.org/croissant/",
11
+ "data": {
12
+ "@id": "cr:data",
13
+ "@type": "@json"
14
+ },
15
+ "dataBiases": "rai:dataBiases",
16
+ "dataCollection": "cr:dataCollection",
17
+ "dataType": {
18
+ "@id": "cr:dataType",
19
+ "@type": "@vocab"
20
+ },
21
+ "dct": "http://purl.org/dc/terms/",
22
+ "extract": "cr:extract",
23
+ "field": "cr:field",
24
+ "fileProperty": "cr:fileProperty",
25
+ "fileObject": "cr:fileObject",
26
+ "fileSet": "cr:fileSet",
27
+ "format": "cr:format",
28
+ "includes": "cr:includes",
29
+ "isArray": "cr:isArray",
30
+ "isLiveDataset": "cr:isLiveDataset",
31
+ "jsonPath": "cr:jsonPath",
32
+ "key": "cr:key",
33
+ "md5": "cr:md5",
34
+ "parentField": "cr:parentField",
35
+ "path": "cr:path",
36
+ "personalSensitiveInformation": "rai:personalSensitiveInformation",
37
+ "recordSet": "cr:recordSet",
38
+ "references": "cr:references",
39
+ "regex": "cr:regex",
40
+ "repeated": "cr:repeated",
41
+ "replace": "cr:replace",
42
+ "sc": "https://schema.org/",
43
+ "separator": "cr:separator",
44
+ "source": "cr:source",
45
+ "subField": "cr:subField",
46
+ "transform": "cr:transform",
47
+ "rai": "http://mlcommons.org/croissant/RAI/",
48
+ "dataLimitations": "rai:dataLimitations",
49
+ "dataUseCases": "rai:dataUseCases",
50
+ "dataSocialImpact": "rai:dataSocialImpact",
51
+ "hasSyntheticData": "rai:hasSyntheticData",
52
+ "prov": "http://www.w3.org/ns/prov#",
53
+ "wasGeneratedBy": "prov:wasGeneratedBy"
54
+ },
55
+ "@type": "sc:Dataset",
56
+ "name": "OTel-Reranker",
57
+ "description": "OTel-Reranker is a pointwise reranking dataset for training cross-encoder models that score the relevance of a passage to a given query in the telecommunications domain.",
58
+ "alternateName": [
59
+ "farbodtavakkoli/OTel-Reranker",
60
+ "OTel-Reranker"
61
+ ],
62
+ "conformsTo": "http://mlcommons.org/croissant/1.1",
63
+ "creator": {
64
+ "@type": "Person",
65
+ "name": "Farbod Tavakkoli",
66
+ "url": "https://huggingface.co/farbodtavakkoli"
67
+ },
68
+ "keywords": [
69
+ "text-classification",
70
+ "English",
71
+ "apache-2.0",
72
+ "100K - 1M",
73
+ "telecommunications",
74
+ "telecom",
75
+ "3gpp",
76
+ "oran",
77
+ "gsma",
78
+ "reranker",
79
+ "cross-encoder",
80
+ "retrieval",
81
+ "otel",
82
+ "Croissant"
83
+ ],
84
+ "license": "https://choosealicense.com/licenses/apache-2.0/",
85
+ "url": "https://huggingface.co/datasets/farbodtavakkoli/OTel-Reranker",
86
+ "distribution": [
87
+ {
88
+ "@type": "cr:FileObject",
89
+ "@id": "repo",
90
+ "name": "repo",
91
+ "description": "The Hugging Face git repository.",
92
+ "contentUrl": "https://huggingface.co/datasets/farbodtavakkoli/OTel-Reranker/tree/refs%2Fconvert%2Fparquet",
93
+ "encodingFormat": "git+https",
94
+ "sha256": "https://github.com/mlcommons/croissant/issues/80"
95
+ },
96
+ {
97
+ "@type": "cr:FileSet",
98
+ "@id": "parquet-files-for-config-default",
99
+ "containedIn": {
100
+ "@id": "repo"
101
+ },
102
+ "encodingFormat": "application/x-parquet",
103
+ "includes": "default/*/*.parquet"
104
+ }
105
+ ],
106
+ "recordSet": [
107
+ {
108
+ "@type": "cr:RecordSet",
109
+ "@id": "default",
110
+ "name": "default",
111
+ "field": [
112
+ {
113
+ "@type": "cr:Field",
114
+ "@id": "default/split",
115
+ "name": "split",
116
+ "description": "The dataset split (e.g., train, test, validation).",
117
+ "dataType": "sc:Text",
118
+ "source": {
119
+ "fileSet": {
120
+ "@id": "parquet-files-for-config-default"
121
+ },
122
+ "extract": {
123
+ "fileProperty": "fullpath"
124
+ },
125
+ "transform": {
126
+ "regex": "default/(?P<split>[^/]*)/.*"
127
+ }
128
+ }
129
+ },
130
+ {
131
+ "@type": "cr:Field",
132
+ "@id": "default/sentence_0",
133
+ "name": "sentence_0",
134
+ "description": "The query sentence.",
135
+ "dataType": "sc:Text",
136
+ "source": {
137
+ "fileSet": {
138
+ "@id": "parquet-files-for-config-default"
139
+ },
140
+ "extract": {
141
+ "column": "sentence_0"
142
+ }
143
+ }
144
+ },
145
+ {
146
+ "@type": "cr:Field",
147
+ "@id": "default/sentence_1",
148
+ "name": "sentence_1",
149
+ "description": "The passage sentence.",
150
+ "dataType": "sc:Text",
151
+ "source": {
152
+ "fileSet": {
153
+ "@id": "parquet-files-for-config-default"
154
+ },
155
+ "extract": {
156
+ "column": "sentence_1"
157
+ }
158
+ }
159
+ },
160
+ {
161
+ "@type": "cr:Field",
162
+ "@id": "default/label",
163
+ "name": "label",
164
+ "description": "The relevance score.",
165
+ "dataType": "sc:Float",
166
+ "source": {
167
+ "fileSet": {
168
+ "@id": "parquet-files-for-config-default"
169
+ },
170
+ "extract": {
171
+ "column": "label"
172
+ }
173
+ }
174
+ }
175
+ ]
176
+ }
177
+ ],
178
+ "rai:dataLimitations": "The dataset covers telecommunications standards and specifications only. It is English-only and reflects standards available at the time of collection. Performance of models trained on this data should not be generalized beyond the telecom domain.",
179
+ "rai:dataBiases": "The dataset may overrepresent certain telecom subdomains (e.g., 3GPP, O-RAN) relative to others due to source availability. Widely documented standards topics are better represented than niche or emerging telecom areas.",
180
+ "rai:personalSensitiveInformation": "The dataset does not contain personal or sensitive information. All source material is drawn from publicly available technical standards, specifications, and academic publications.",
181
+ "rai:dataUseCases": "Training, Fine Tuning",
182
+ "rai:dataSocialImpact": "The dataset supports open and reproducible telecom AI research, reducing barriers to domain-specific model development. The primary risk is potential misuse of domain-specialized models to generate misleading telecom documentation.",
183
+ "rai:hasSyntheticData": "The training data includes synthetically generated question-answer pairs derived from source documents through automated pipelines, including LLM-assisted answer recreation and fact-grounded verification.",
184
+ "rai:dataCollection": "Source datasets and collection methodology: the OTel corpus is derived from publicly available telecom sources, including 3GPP technical specifications, GSMA Permanent Reference Documents (PRDs) and Discover portal materials, O-RAN Alliance specifications across working groups 1, 2, 4, 5, 6, 7, 8, 9, and 10, IETF Request for Comments (RFC) series, telecom academic papers from arXiv, telecom Wikipedia articles, telecom-related Common Crawl pages, and industry whitepapers. The OTel datasets release only derived question-answer pairs and contrastive examples, not the raw source documents. Data was collected by 100+ domain experts across industry and academia and processed through a six-phase enrichment and four-stage cleaning pipeline.",
185
+ "prov:wasGeneratedBy": {
186
+ "@type": "prov:Activity",
187
+ "name": "OTel Data Curation Pipeline",
188
+ "description": "A six-phase enrichment and four-stage cleaning pipeline curated by 100+ domain experts from AT&T, AMD, GSMA, Microsoft, RelationalAI, MLCommons, Mantis NLP, Essential AI, Yale University, Khalifa University, University of Leeds, The University of Texas at Dallas, Purdue University, and NetoAI. The raw corpus of approximately 1.1M samples was cleaned to approximately 326K samples through heuristic filtering, cross-encoder reranking filtering, embedding-based semantic filtering, and deduplication."
189
+ }
190
+ }