shasank0001/vizag-city-guide-cpt
Vizag City Guide — CPT Corpus Continued pre-training (CPT) text corpus for Visakhapatnam (Vizag), India. Plain .txt files, one document per line (or paragraph). Pairs with the supervised dataset shasank0001/vizag-city-guide-sft: pre-train on this corpus first, then fine-tune on the SFT set. Contents (~3.4 MB, ~850k tokens at ~4 chars/token) File Size What it is corpus.txt 228 KB Concatenated Wikipedia + government base documents (45 docs)… See the full description on the dataset page: https://huggingface.co/datasets/shasank0001/vizag-city-guide-cpt.
Vizag City Guide — CPT Corpus
Continued pre-training (CPT) text corpus for Visakhapatnam (Vizag), India. Plain .txt files, one document per line (or paragraph). Pairs with the supervised dataset `shasank0001/vizag-city-guide-sft`: pre-train on this corpus first, then fine-tune on the SFT set.
Contents (~3.4 MB, ~850k tokens at ~4 chars/token)
Note on Telugu: corpus_telugu.txt dominates by volume. For Vizag-focused training, either downsample it or keep it as a separate stream so city knowledge is not diluted. Temporal files (2026 events) are secondary-source summaries — verify against official notices before redistribution.
Usage
from datasets import load_dataset
# text files, one per corpus; example: base + hard government tables
ds = load_dataset("text", data_files=[
"corpus.txt", "corpus_chunks.txt", "corpus_hard.txt",
"corpus_bus.txt", "corpus_layout.txt",
])
tok = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM-360M-Instruct")
ds = ds.map(lambda x: tok(x["text"], truncation=True, max_length=2048),
batched=True, remove_columns=["text"])
# Trainer with standard causal-LM objective, then SFT on vizag-city-guide-sftSources & licensing
- Wikipedia + Wikivoyage extracts — CC BY-SA
- VMRDA / District Handbook / Smart Cities / AP Tourism / APSRTC timetables — Government Open Data License India (GODL) / public record, with attribution
- OpenStreetMap POIs — ODbL 1.0
- 2026 news items — fair-use summaries with source URLs (see harvest log in the build repo)
- Synthetic layout/proximity notes — original
Citation
Vizag City Guide CPT corpus (2026). Built from Wikipedia (CC BY-SA), Government of India / Andhra Pradesh open data (GODL), and OpenStreetMap (ODbL).
