Nanthasit/sakthai-combined-v7

SakThai Combined v7

Curated, larger-scale instruction-tuning data for tool-calling, function-calling, and agent-style reasoning in the SakThai model family.

Dataset Summary

SakThai Combined v7 extends the v6 family with more multi-turn examples, broader tool coverage, and stronger <tool>/function-calling formatting. It is intended for fine-tuning models that should invoke tools naturally, then continue the conversation after tool results.

Data Fields

Field	Type	Description
messages	list[dict]	Chat turns in OpenAI-style format with roles system, user, assistant, and tool
tools	list[dict]	JSON schema definitions of callable tools available in each example
conversations	list[dict]	Multi-turn conversation annotations for examples with follow-up tool execution
source_split	string	Origin source/dataset tag when the example was derived from another split
fingerprint	string	Content hash for reproducibility and deduplication

Dataset Structure

• Config: default
• Train split: present
• Test split: present
• Format: JSONL
• Language: English, with multilingual assistant turns preserved from source conversations

Splits

Split	File	Examples	Approx. size
train	data/train.jsonl	2,000+	~4.7 MB
test	data/test.jsonl	235	~355 KB

Dataset Size

Metric	Value
Total repository size	~4.7 MB
Train file	data/train.jsonl
Test file	data/test.jsonl
Last modified	2026-08-01
Downloads	101

Verified Snapshot

Property	Value
Source	Nanthasit/sakthai-combined-v7
Files	README.md, data/train.jsonl, data/test.jsonl, load_dataset_demo.ipynb, notebooks/, scripts/
Train bytes	4,742,978
Test bytes	355,490
Verified by	SakThai / HF Hub metadata + file inspection
Last verified	2026-08-01

Loading Examples

from datasets import load_dataset

ds = load_dataset("Nanthasit/sakthai-combined-v7")
print(ds)

from datasets import load_dataset

ds = load_dataset("Nanthasit/sakthai-combined-v7", split="train")
example = ds[0]
print(example["messages"])
print(example["tools"][0]["function"]["name"])

from datasets import load_dataset

# inspect test split only
test = load_dataset("Nanthasit/sakthai-combined-v7", split="test")
print(f"Test examples: {len(test)}")
print(test[0]["messages"][-1]["content"])

Streaming / Partial Load

from datasets import load_dataset

ds = load_dataset("Nanthasit/sakthai-combined-v7", split="train", streaming=True)
for i, row in enumerate(ds):
    if i >= 1:
        break
    print(row["tools"][0]["function"]["name"])

Dataset Creation

• Curation: Combined from prior SakThai combined datasets with additional tool-calling and multi-turn coverage.
• Annotation: examples include expected tool-call formatting and assistant follow-up behavior.
• Scope: training and evaluation for tool-use instruction tuning.

Use Cases

• Fine-tuning instruction-following models for function-calling.
• Evaluating tool-calling format adherence on held-out tools.
• Agent-style multi-turn conversation training with tool results.

Citation

@misc{sakthai-combined-v7,
  title={SakThai Combined v7},
  author={Beer / Nanthasit},
  year={2026},
  url={https://huggingface.co/datasets/Nanthasit/sakthai-combined-v7
}

Evaluation & Community

Results are surfaced in the SakThai Leaderboard.

License

CC BY 4.0

Part of the SakThai Model Family.