Datasets:
|
Download README.md from thantzinphyo/Hey-Function-Burmese: direct link, hf CLI and curl.
- Browser
- Download file 14.2 kB
-
https://huggingface.co/datasets/thantzinphyo/Hey-Function-Burmese/resolve/main/README.md
- Command line
-
hf download hf://datasets/thantzinphyo/Hey-Function-Burmese/README.md
-
curl -L -o README.md https://huggingface.co/datasets/thantzinphyo/Hey-Function-Burmese/resolve/main/README.md
14.2 kB
metadata
license: cc-by-4.0
task_categories:
- text-generation
language:
- en
- my
tags:
- gemma
- functiongemma
- function-calling
- tool-use
- on-device
- mobile-actions
- burmese
- myanmar
- code-switching
size_categories:
- 10K<n<100K
configs:
- config_name: default
data_files:
- split: train
path: train.parquet
Hey-Function-Burmese
Burmese (Myanmar) & English Function-Calling Dataset for Lightweight / Edge LLMs
Model (Hey-Function-270M) • GitHub Codebase • English • မြန်မာဘာသာ
English
- Target Models: Compatible with FunctionGemma, Gemma, and on-device function-calling LLMs
Highlights
- Bilingual & Code-Switching: English, Burmese, and natural English-Burmese code-switching queries.
- 61 Real-World Tools: Phone controls, messaging, calendar, navigation, media, smart home, and vehicle IoT.
- Optimized for Edge LLMs: Formatted for lightweight on-device models such as FunctionGemma / Gemma.
- Parallel Function Calling: Over 15,000 instances with 2–3 tool invocations per prompt.
- Conversational Grounding: Clarifications, greetings, and refusals to prevent unprompted tool triggers.
- Schema Validation: Verified against tool schemas and the official FunctionGemma template.
Dataset Overview
| Field | Value |
|---|---|
| Total examples | 42,673 |
| Function-call examples | 40,410 (single: 24,962 · parallel: 15,448) |
| Conversational / no-call examples | 2,263 |
| Train / Eval split | 38,833 / 3,840 |
| Total distinct tools | 61 |
| Language registers | English · Burmese (Unicode) · Code-Switch |
Linguistic Composition
| Language Register | Example | Rows |
|---|---|---|
| English | "Send an email to Nan Khin about tomorrow's meeting" | 15,898 |
| Burmese | "မနန်းခင် ဆီ အီးမေးလ် ပို့ပေးပါ" | 13,096 |
| English–Burmese Code-Switch | "Stopwatch စဖွင့်ပေးပါ" | 13,679 |
| Total | 42,673 |
Task Coverage
| Category | Example | Target Action |
|---|---|---|
| Single call | "မနက် ၆ နာရီ နှိုးစက်ပေး" | set_alarm |
| Parallel calls | "Play lofi music and set volume to 20%" | play_music + set_volume_level |
| Clarification | "Alarm ပေးထား" | "Alarm ကို ဘယ်အချိန် set လုပ်ရမလဲ။" |
| Out-of-scope | "Grab ကား ခေါ်ပေး" | Polite refusal |
| Conversational | "Thank you ပါ" | Friendly text reply |
Dataset Structure
Each record follows the standard developer → user → assistant turn format with full function schemas:
{
"metadata": "train",
"tools": [{
"function": {
"name": "set_alarm",
"description": "Sets an alarm clock.",
"parameters": {
"type": "object",
"properties": { "time": { "type": "string", "description": "Alarm time" } },
"required": ["time"]
}
}
}],
"messages": [
{ "role": "developer", "content": "You are a model that can do function calling.
" },
{ "role": "user", "content": "မနက် ၆ နာရီ ၁၅ မိနစ် နှိုးစက်ပေး" },
{ "role": "assistant", "tool_calls": [{ "function": { "name": "set_alarm", "arguments": { "time": "06:15 AM" } } }] }
]
}
Tool Categories (61 Tools)
├── Hardware & System Controls (23)
├── Communication & Contacts (10)
├── Scheduling & Utilities (9)
├── Location & Navigation (5)
├── Connected Vehicle Telematics (7)
└── Media & Smart Home (7)
Data Quality & Usage Notes
- Schema Validation: All function calls adhere to declared parameter types, required fields, and the official FunctionGemma template.
- Privacy & Synthetic Data: All names, email addresses, and phone numbers in this dataset are synthetic data created strictly for testing purposes and do not contain real personal information. To protect user privacy, a trailing
00is appended to all phone numbers. - Unicode Standard: Burmese text strictly follows standard Myanmar Unicode with native colloquial phrasing.
- Dataset Scope: Single-turn exchanges covering 61 distinct tools across English, Burmese, and code-switching queries.
Notices for Dataset Users and Developers
- Synthetic phone number suffix (00): When fine-tuning a model on this dataset, consider stripping or normalizing the trailing 00 suffix as needed.
- Connected vehicle controls: Critical actions such as unlocking doors, opening trunks, or starting vehicle engines must require explicit user confirmation. Driving or steering controls are not included.
License
Released under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license.
Citation
@dataset{hey_function_burmese_2026,
title={Hey-Function-Burmese: A Burmese & English Function-Calling Dataset for Lightweight / Edge LLMs},
author={Thant Zin Phyo},
year={2026},
publisher={Hugging Face},
howpublished={\url{https://huggingface.co/datasets/thantzinphyo/Hey-Function-Burmese}}
}
Acknowledgements
This dataset follows the on-device function-calling schema and formatting conventions established for Google's FunctionGemma architecture.
မြန်မာဘာသာ
- အသုံးပြုနိုင်သော Model များ: FunctionGemma, Gemma အပါအဝင် On-Device Function-Calling LLM များတွင် အသုံးပြုနိုင်ပါသည်။
ထူးခြားချက်များ
- ဘာသာစကား လွှမ်းခြုံမှု: အင်္ဂလိပ်၊ မြန်မာ နှင့် မြန်မာ-အင်္ဂလိပ် ရောနှောစကားပြော (Code-Switching) ၃ မျိုးစလုံး ပါဝင်ပါသည်။
- Tool ၆၁ မျိုး ပါဝင်မှု: ဖုန်း Setting၊ ဆက်သွယ်ရေး၊ ပြက္ခဒိန်၊ လမ်းညွှန်၊ မီဒီယာ၊ Smart Home နှင့် မော်တော်ယာဉ် ထိန်းချုပ်မှုများ။
- Edge LLM များအတွက် အထူးပြုလုပ်ထားခြင်း: FunctionGemma / Gemma ကဲ့သို့ ပေါ့ပါးသည့် On-Device မော်ဒယ်များအတွက် ရည်ရွယ်ပါသည်။
- Parallel Function Calling: Prompt တစ်ခုတည်းမှ Tool ၂ ခု သို့မဟုတ် ၃ ခုကို တစ်ပြိုင်နက် ခေါ်ယူသော နမူနာ ၁၅,၀၀၀ ကျော် ပါဝင်ပါသည်။
- Function မခေါ်သင့်သည့် အခြေအနေများ: နှုတ်ဆက်စကား၊ မလုပ်ဆောင်နိုင်သော တောင်းဆိုမှု ငြင်းဆိုခြင်းနှင့် အချက်အလက် မေးမြန်းခြင်းများ ပါဝင်ပါသည်။
- Schema နှင့် Template ကိုက်ညီမှု: Tool Schema သတ်မှတ်ချက်များနှင့် FunctionGemma Chat Template စံနှုန်းအတိုင်း စနစ်တကျ စစ်ဆေးထားပါသည်။
အချက်အလက် အနှစ်ချုပ်
| ကဏ္ဍ | အသေးစိတ် |
|---|---|
| နမူနာ စုစုပေါင်း | ၄၂,၆၇၃ |
| Function Call နမူနာ | ၄၀,၄၁၀ (တစ်ခုတည်း: ၂၄,၉၆၂ · အများ: ၁၅,၄၄၈) |
| စကားပြော / Function မခေါ်သည့် နမူနာ | ၂,၂၆၃ |
| Train / Eval ခွဲဝေမှု | ၃၈,၈၃၃ / ၃,၈၄၀ |
| Tool စုစုပေါင်း | ၆၁ |
| ဘာသာစကား ပုံစံ | English · Burmese (Unicode) · Code-Switch |
ဘာသာစကား ပါဝင်ဖွဲ့စည်းမှု
| ဘာသာစကား ပုံစံ | ဥပမာ | အကြောင်းရေ |
|---|---|---|
| English | "Turn off the flashlight" | ၁၅,၈၉၈ |
| မြန်မာ | "ဖုန်းမီး ပိတ်ပေးပါ" | ၁၃,၀၉၆ |
| မြန်မာ-အင်္ဂလိပ် Code-Switch | "Screen auto-rotate off ပေးပါ" | ၁၃,၆၇၉ |
| စုစုပေါင်း | ၄၂,၆၇၃ |
လုပ်ဆောင်နိုင်စွမ်းများ
| အမျိုးအစား | ဥပမာ | သတ်မှတ် လုပ်ဆောင်ချက် |
|---|---|---|
| Function တစ်ခု | "မနက် ၆ နာရီ နှိုးစက်ပေး" | set_alarm |
| Function အများ | "သီချင်းဖွင့်ပြီး အသံ ၂၀% ထားပေး" | play_music + set_volume_level |
| ပြန်လည်မေးမြန်းခြင်း | "Alarm ပေးထား" | "Alarm ကို ဘယ်အချိန် set လုပ်ရမလဲ။" |
| မလုပ်ဆောင်နိုင်သော တောင်းဆိုမှု | "Grab ကား ခေါ်ပေး" | ယဉ်ကျေးစွာ ငြင်းဆိုခြင်း |
| နှုတ်ဆက်စကား | "ကျေးဇူးပဲ" | စာသားဖြင့် အဖြေပေးခြင်း |
ဒေတာ ဖွဲ့စည်းပုံ
ဒေတာ တစ်ကြောင်းစီသည် Tool Schema အပြည့်အစုံနှင့်အတူ developer → user → assistant စကားပြောခန်း တစ်ခု ဖြစ်ပြီး အသေးစိတ် ဖွဲ့စည်းပုံ နမူနာကို အထက်ပါ English အပိုင်းတွင် ကြည့်ရှုနိုင်ပါသည်။
ပါဝင်သော Tool ၆၁ မျိုး
├── ဖုန်းစနစ်နှင့် အသုံးအဆောင် (23)
├── ဆက်သွယ်ရေးနှင့် အဆက်အသွယ် (10)
├── အချိန်ဇယားနှင့် အသုံးဝင်ပစ္စည်းများ (9)
├── တည်နေရာနှင့် လမ်းပြစနစ် (5)
├── မော်တော်ယာဉ် IoT စနစ် (7)
└── မီဒီယာနှင့် အင်တာနက် (7)
ဒေတာ အရည်အသွေးနှင့် အသုံးပြုမှု မှတ်ချက်
- Schema ကိုက်ညီမှု: Function Call များကို Tool Schema သတ်မှတ်ချက်များနှင့် FunctionGemma Chat Template ပုံစံအတိုင်း စနစ်တကျ ပြုစုထားပါသည်။
- ကိုယ်ရေးလုံခြုံမှုနှင့် စမ်းသပ်ချက် အသိပေးချက်: ဤ Dataset တွင် ပါဝင်သော အမည်များ၊ အီးမေးလ်လိပ်စာများနှင့် ဖုန်းနံပါတ်များအားလုံးသည် စမ်းသပ်မှုအတွက် ဖန်တီးထားသော အချက်အလက်များဖြစ်ပြီး အမှန်တကယ် အသုံးပြုသူများ၏ ကိုယ်ရေးအချက်အလက်များ မပါဝင်ပါ။ ကိုယ်ရေးလုံခြုံမှုကို ကာကွယ်ရန် ဖုန်းနံပါတ်များ၏ အဆုံးတွင် 00 ကို ထည့်သွင်းထားပါသည်။
- စံ Unicode: မြန်မာစာအားလုံးကို Standard Myanmar Unicode ဖြင့် ရေးသားထားပါသည်။
- ဒေတာ နယ်ပယ်: Single-turn စကားပြောခန်းများဖြစ်ပြီး Tool ၆၁ မျိုးကို ဘာသာစကား ၃ မျိုးစလုံးဖြင့် လွှမ်းခြုံထားပါသည်။
Dataset အသုံးပြုသူ နှင့် Developer များ သတိပြုရန်
- ဖုန်းနံပါတ် 00 အပို: ဤ Dataset ဖြင့် Model တစ်ခုခုကို Fine-Tune လုပ်မည်ဆိုပါက နောက်ဆုံး 00 ကို ဖယ်ရှားပေးရန် လိုအပ်ပါသည်။
- မော်တော်ယာဉ် ထိန်းချုပ်မှု: ကားတံခါးဖွင့်ခြင်း၊ စက်နှိုးခြင်းများအတွက် အတည်ပြုချက် အဆင့် ထပ်မံခံထားပေးရန် လိုအပ်ပါသည်။ ကားမောင်းနှင်မှုဆိုင်ရာ (စတီယာရင်၊ လီဗာ၊ ဘရိတ်) အမိန့်များ မပါဝင်ပါ။
လိုင်စင်
ဤဒေတာဆက်ကို CC-BY-4.0 လိုင်စင်ဖြင့် ထုတ်ဝေထားပါသည်။
ကိုးကားရန် (BibTeX)
@dataset{hey_function_burmese_2026,
title={Hey-Function-Burmese: A Burmese & English Function-Calling Dataset for Lightweight / Edge LLMs},
author={Thant Zin Phyo},
year={2026},
publisher={Hugging Face},
howpublished={\url{https://huggingface.co/datasets/thantzinphyo/Hey-Function-Burmese}}
}