Yusiko/azerbaijani-wiki-instruct-alpaca

An Azerbaijani instruction-following dataset in Alpaca format (instruction, input, output).
Useful for supervised fine-tuning (SFT) to improve instruction following and long-form, explanatory answers in Azerbaijani.

Quick facts

• Rows: 167,590
• Split: train only
• License: MIT
• Main file: azerbaijani_wiki_instruct.jsonl (~432 MB)
• Auto-converted Parquet: ~225 MB

Data schema

Each record contains:

• instruction (string): the task/prompt in Azerbaijani
• input (string): optional additional context (often empty)
• output (string): the target answer/response in Azerbaijani

Example JSONL line:

{"instruction":"Write about the Internet.","input":"","output":"..."}

Usage

Load with 🤗 Datasets

from datasets import load_dataset

ds = load_dataset("Yusiko/azerbaijani-wiki-instruct-alpaca")
print(ds)
print(ds["train"][0].keys())
print(ds["train"][0]["instruction"])
print(ds["train"][0]["input"])
print(ds["train"][0]["output"][:300])

Convert to pandas

from datasets import load_dataset

train = load_dataset("Yusiko/azerbaijani-wiki-instruct-alpaca", split="train")
df = train.to_pandas()
df.head()

Recommended training format

If you want a single prompt string for training, a common Alpaca-style template is:

### Instruction:
{instruction}

### Input:
{input}

### Response:
{output}

If input is empty, you can omit the input block.

Intended use

• Instruction tuning / SFT for Azerbaijani LLMs
• Building Azerbaijani assistants that produce structured explanatory answers
• As a base dataset to combine with domain-specific instruction data (math, code, dialogue, etc.)

Limitations & risks

• The dataset is geared toward encyclopedic/explanatory outputs; it may not produce strong conversational style without additional dialogue data.
• As with any large text dataset, it may include factual errors, outdated statements, formatting artifacts, repetitions, or bias.
• Not a substitute for domain-verified data in high-stakes settings (medical/legal/financial).

License

MIT License.

Citation

If you use this dataset in research or projects, please cite it as:

@dataset{yusiko_azerbaijani_wiki_instruct_alpaca,
  title        = {azerbaijani-wiki-instruct-alpaca},
  author       = {Yusiko},
  publisher    = {Hugging Face},
  howpublished = {Hugging Face Datasets},
  license      = {MIT}
}