README.md

metadata

license: cc-by-sa-3.0
task_categories:
  - question-answering
  - text-generation
  - translation
  - summarization
language:
  - te
tags:
  - adaption-labs
  - autoscientist
  - telugu
  - low-resource-language
  - multi-task
  - reasoning-traces
  - hackathon
size_categories:
  - 10K<n<100K

Telugu MultiTask Instruct 77K — Adaption AutoScientist Challenge Dataset

Powered by Adaptive Data — Adaption Labs

Dataset Description

A large-scale, multi-task Telugu instruction-tuning dataset combining 77,653 rows from 7 open-source Telugu NLP collections. Covers diverse tasks including news summarization, QA, creative writing, translation, and general instruction following — all processed through the Adaption Labs AutoScientist platform for quality enhancement.

This dataset was created for the 2026 Adaption AutoScientist Challenge (Language Category).

Dataset Details

Field	Value
Size	77,653 rows
Format	JSONL
Language	Telugu (te)
License	CC-BY-SA-3.0

Source Data & Attribution

Source	Description	License
Telugu Alpaca Cleaned (telugu_ai_instructions)	General instruction following	CC-BY-4.0
Dolci SFT Telugu (telugu_instruction_pairs)	Instruction-response pairs	Apache-2.0
TyDi QA Telugu (telugu_qa_news_gen)	News-based QA	Apache-2.0
Aya Telugu Poems & News (telugu_news_and_poetry_gen)	Creative writing & journalism	Apache-2.0
Telugu News Summarization (telugu_news_summaries)	Article summarization	MIT
Samanantar EN-TE / Dolly 15K (english_to_telugu_translation)	Translation pairs	CC-0 / CC-BY-SA-3.0
Telugu QA Andhra Facts (telugu_qa_andhra_facts)	Regional knowledge QA	Mixed Open

Quality Enhancement via Adaption Labs

The dataset was processed through the Adaption Labs Adaptive Data Pipeline:

Metric	Before	After	Change
Grade	C	B	⬆️
Score	6.0	8.14	+35.6%
Percentile	—	18.23	—

AutoScientist features applied:

• ✅ Prompt Deduplication
• ✅ Prompt Rephrase
• ✅ Hallucination Mitigation
• ✅ Reasoning Traces

Associated Model

• HuggingFace: narendarcodes/adaption-gpt-oss-120b-telugu-instruct
• Base Model: togethercomputer/gpt-oss-120b-bf16 (117B)
• Win Rate: 66% adapted vs. 34% base

Citation

@misc{golla2026telugu_data,
  title={Telugu MultiTask Instruct 77K Dataset},
  author={Golla Narendar},
  year={2026},
  note={Processed via Adaption Labs AutoScientist. Powered by Adaptive Data.}
}

---

Powered by Adaptive Data — Adaption Labs