nareshmodina/TeleSpec-Data

---
license: cc-by-nc-4.0
language:
- en
tags:
- telecom
- LLM
- ETSI
- 3GPP
- standards
- continual-pretraining
task_categories:
- text-generation
size_categories:
- 10M<n<100M
configs:
- config_name: default
data_files:
- split: train
path: data/*/*.parquet
- config_name: 3gpp-standard
data_files:
- split: train
path: data/3gpp-standard/*.parquet
- config_name: etsi-standard
data_files:
- split: train
path: data/etsi-standard/*.parquet
---

# TeleSpec-Data

## Dataset Summary

TeleSpec-Data is a dataset of telecommunications standards documents from two major bodies: ETSI and 3GPP. It is designed for continual pretraining of language models on telecom domain knowledge.

The dataset consists of two categories:

- **3gpp-standard**: 15,054 3GPP technical specifications and reports covering Release 8 through Release 19 (updated to April 2025), sourced from [TSpec-LLM](https://huggingface.co/datasets/rasoul-nikbakht/TSpec-LLM).
- **etsi-standard**: 23,248 ETSI documents spanning 15 working groups (TS, TR, EN, ES, GS, GR, EG, ETR, ETS, GTS, and others), covering years 2000–2024, extracted from the PDF corpus in [NetSpec-LLM](https://huggingface.co/datasets/rasoul-nikbakht/NetSpec-LLM).

**Total: 38,302 documents.**

## Dataset Structure

### Data Fields

- `id`: Unique identifier for each document.
- `category`: Category of the document — `3gpp-standard` or `etsi-standard`.
- `content`: Full text of the document. Sections are separated by ` \n `.
- `metadata`: JSON string with document-level metadata. Fields vary by category.

### Data Instances

An example from the 3GPP subset:

```
{
"id": "3gpp_24229_Rel-17_hc0",
"category": "3gpp-standard",
"content": "3GPP TS 24.229 Release 17 Vhc0 \n 1 Scope \n The present document specifies...",
"metadata": "{\"doc_number\": \"24229\", \"series\": \"24\", \"release\": \"17\", \"version\": \"hc0\", \"filename\": \"24229-hc0.md\", \"series_dir\": \"24_series\"}"
}
```

An example from the ETSI subset:

```
{
"id": "etsi_28737",
"category": "etsi-standard",
"content": "ETSI TS 132 371 V7.3.1 (2008-07) \n Security Management concept and requirements \n 1 Scope \n The present document specifies...",
"metadata": "{\"working_group\": \"TS\", \"year\": 2008, \"deliverable_type\": \"TS\", \"version\": \"7.3.1\", \"title\": \"Security Management concept and requirements\"}"
}
```

## Sample Code

```python
import json
from datasets import load_dataset

# Full dataset
ds = load_dataset("NareshModina/TeleSpec-Data")

# Single category
ds = load_dataset("NareshModina/TeleSpec-Data", name="etsi-standard")
ds = load_dataset("NareshModina/TeleSpec-Data", name="3gpp-standard")

sample = ds["train"][0]
print(f"ID: {sample['id']}\nCategory: {sample['category']}\nContent: {sample['content'][:200]}")
for key, value in json.loads(sample["metadata"]).items():
print(f"{key}: {value}")
```

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{modina2025telespecdata,
author = {Naresh Modina},
title = {TeleSpec-Data: A Telecommunications Standards Dataset for Language Model Pretraining},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/NareshModina/TeleSpec-Data}
}
```

Please also cite the upstream sources:

```bibtex
@misc{nikbakht2024tspecllm,
title = {TSpec-LLM: An Open-source Dataset for LLM Understanding of 3GPP Specifications},
author = {Rasoul Nikbakht and Mohamed Benzaghta and Giovanni Geraci},
year = {2024},
eprint = {2406.01768},
archivePrefix = {arXiv},
primaryClass = {cs.NI},
url = {https://arxiv.org/abs/2406.01768}
}
```

ETSI document corpus sourced from [rasoul-nikbakht/NetSpec-LLM](https://huggingface.co/datasets/rasoul-nikbakht/NetSpec-LLM) (CC BY-NC 4.0).