Nanthasit/sakthai-irrelevance-supplement

Dataset Card for SakThai Irrelevance Supplement

Dataset Name: SakThai Irrelevance Supplement
Author(s): SakThai Agent (Nanthasit)
License: MIT
Last Updated: 2026-08-01

Dataset Summary

The SakThai Irrelevance Supplement is a curated dataset of dialogue examples designed to train and evaluate models to gracefully decline tool use when unnecessary. This dataset specifically contains examples where available functions are NOT needed for the user's request — teaching models to recognize when to abstain from tool calling.

Size & Format:

• 60 examples (verified 2026-08-01 via Datasets Server)
• ~34 KB in-memory (JSONL format)
• Single train split for direct use in training pipelines

Data Fields

Field	Type	Description
messages	list[dict]	Multi-turn conversation. Each message contains role (system/user/assistant) and content (string).
tools	list[dict]	Available functions offered to the model. Each tool has name and description.

Data Example

{
  "messages": [
    {"role": "user", "content": "What's the weather like today?"},
    {"role": "assistant", "content": "I don't have access to real-time weather data, so I can't tell you the current weather. I'd recommend checking a weather website or app for accurate information."}
  ],
  "tools": [
    {"name": "get_weather", "description": "Fetch current weather for a location"},
    {"name": "calculate", "description": "Perform mathematical calculations"}
  ]
}

Splits

Split	Config	Rows	Bytes (Memory)	Purpose
train	default	60	~34 KB	Training examples for models to learn when NOT to use tools

Verified via Datasets Server on 2026-08-01.

Dataset Size

Metric	Value
Total examples	60
Total bytes (memory)	~34 KB
Format	JSONL (line-delimited JSON)
Encoding	UTF-8

How to Load

from datasets import load_dataset

# Load full dataset
ds = load_dataset("Nanthasit/sakthai-irrelevance-supplement")

# Access train split
train = ds["train"]
print(f"Loaded {len(train)} examples")

# Iterate over examples
for example in train:
    print("Messages:", example["messages"])
    print("Tools available:", [t["name"] for t in example["tools"]])

Dataset Construction & Source

This dataset was constructed from:

• Nanthasit/sakthai-combined-v6 — Base tool-calling examples
• Nanthasit/sakthai-combined-v7 — Additional dialogue patterns

It filters or generates examples where the model should recognize that available tools are irrelevant to the user's request, complementing typical tool-calling datasets that emphasize using tools when needed.

Intended Use

Primary Use Case: Training models (especially 1.5B–7B parameters) to:

1. Recognize when user queries don't require tool use
2. Decline politely without forcing a tool call
3. Provide informative responses when no tool applies

Suitable For:

• Fine-tuning instruction-following models
• Evaluating multi-turn dialogue reasoning
• Complementary training alongside positive tool-calling examples

Not Suitable For:

• Datasets requiring real-time data (weather, prices, etc.)
• Production API deployment without additional data

Evaluation & Metrics

Typical evaluation on this dataset:

• Accuracy: % of examples where model correctly refuses to call a tool
• Confidence: Does the model explicitly state why tools aren't needed?
• Coherence: Is the response natural and conversational?

Licensing

This dataset is released under the MIT License. See the repository for full license text.

Citation

If you use this dataset in your research or model training, please cite:

@dataset{sakthai-irrelevance-supplement-2026,
  title   = {SakThai Irrelevance Supplement: Tool-Calling Abstention Dataset},
  author  = {SakThai Agent and Beer (Nanthasit)},
  year    = {2026},
  url     = {https://huggingface.co/datasets/Nanthasit/sakthai-irrelevance-supplement},
  license = {MIT}
}

Contact & Contributions

• Hugging Face Repo: Nanthasit/sakthai-irrelevance-supplement
• Part of: SakThai Family

---

Dataset Card Version: 2 (improved, verified 2026-08-01)