You can deploy this model and start earning money today!

Kaun9MyatHt3t/
mt5-base-burmese-sports-qa

Name: mt5-base-burmese-sports-qa
Availability: InStock

safetensors

mt5

Views

No views yet

markdown

---
language:
- my
license: mit
library_name: transformers
pipeline_tag: text2text-generation
tags:
- Burmese
- Myanmar
- sports
- question-answer-generation
- mT5
- text2text-generation
---

# mT5-base Burmese Sports QA Generation

A fine-tuned `google/mt5-base` model for generating question-answer pairs from Burmese sports news articles.

## Model Description

This model was fine-tuned for a university NLP project using a dataset of Burmese sports news articles and generated question-answer pairs.

The task is **text-to-text question-answer generation**, not conversational question answering or chatbot generation.

### Input

A Burmese sports news article.

### Output

A generated question-answer pair in Burmese.

Example format:

```text
မေးခွန်း: ...
အဖြေ: ...

Training Dataset

The training dataset contains:

5,644 sports news articles
26,279 question-answer pairs
Average 4.66 QA pairs per article
Minimum 3 QA pairs per article
Maximum 10 QA pairs per article

The dataset was split at the article level to reduce data leakage between training, validation, and test sets.

Split	Articles	QA Pairs
Training	4,515	21,049
Validation	564	2,614
Test	565	2,616
Total	5,644	26,279

The raw dataset is not publicly distributed with this model because the source field contains article text collected from external news sources.

Training Configuration

Parameter	Value
Base model	`google/mt5-base`
Epochs	10
Learning rate	`3e-5`
Training batch size	2
Evaluation batch size	2
Gradient accumulation	8
Effective batch size	16
Weight decay	0.01
Warmup ratio	0.1
Label smoothing	0.1
Optimizer	Adafactor
FP16	Disabled
Gradient checkpointing	Enabled

Usage

Install Transformers:

pip install transformers sentencepiece torch

Load the model:

python

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_name = "Kaun9MyatHt3t/mt5-base-burmese-sports-qa"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

article = "မြန်မာ့လက်ရွေးစင်အသင်းသည် ယနေ့ ဘောလုံးပွဲတစ်ပွဲကို ယှဉ်ပြိုင်ကစားခဲ့သည်။"

inputs = tokenizer(
    article,
    return_tensors="pt",
    truncation=True,
    max_length=256
)

outputs = model.generate(
    **inputs,
    max_new_tokens=128,
    num_beams=4
)

result = tokenizer.decode(
    outputs[0],
    skip_special_tokens=True
)

print(result)

Intended Use

This model is intended for:

Burmese sports article QA generation
NLP research and experimentation
Educational projects
Demonstrations of Burmese text-to-text generation

Limitations

The model has several limitations:

It is specialized for Burmese sports news.
Generated answers may contain factual or grammatical errors.
Output quality depends on the input article.
The model may not perform well on domains outside sports news.
It should not be treated as a general-purpose Burmese language model.
Generated QA pairs should be reviewed before being used in applications.

Project

This model was developed as part of a university team project.

GitHub repository:

https://github.com/Kaun9MyatHt3t/sport-qa-generation

Base Model

This model is based on:

google/mt5-base

License

MIT