Nanthasit/sakthai-combined-v11

SakThai Combined v11

Part of the SakThai Model Family.

Dataset Summary

SakThai Combined v11 is a conversation/tool-use dataset in the SakThai family. It contains multi-turn agent-style dialogues with structured tool definitions, formatted for instruction tuning of function-calling assistants.

Dataset Structure

• Config: default
• Split: train
• Rows: 2,965
• Columns: messages, tools
• Format: Parquet
• Size: ~5.32 MB
• License: Apache 2.0
• Language: English

Data Fields

Field	Type	Description
messages	list[dict]	Multi-turn dialogue. Each dict has role (system/user/assistant/tool) and content. Assistant turns may include tool_calls.
tools	list[dict]	Tool definitions available in the conversation. Each dict includes type, function.name, function.description, and function.parameters.
tool_calls	list[dict] (inside assistant messages)	Structured calls with id, type, and function.name / function.arguments.
tool_call_id	str (inside tool messages)	ID linking a tool result to its preceding tool_calls entry.

Splits

Split	Rows	Format	Notes
train	2,965	Parquet	Single combined split for instruction tuning.

Dataset Size & Stats

Metric	Value
Total rows	2,965
Total columns	2
File size (Parquet)	~5.32 MB
Estimated memory (in-memory)	~5.32 MB
Average message turns per example	~6.2
Language	English

Verified Snapshot

Property	Value
Source	Nanthasit/sakthai-combined-v11
Verified by	SakThai / HF Hub metadata + file inspection
Last verified	2026-08-01

Loading Examples

from datasets import load_dataset

# Load the dataset
ds = load_dataset("Nanthasit/sakthai-combined-v11")
train = ds["train"]

# Inspect one example
example = train[0]
print(example["messages"])
print(example["tools"])

from datasets import load_dataset

# Streaming load for large workloads
ds_stream = load_dataset("Nanthasit/sakthai-combined-v11", split="train", streaming=True)
for row in ds_stream.take(3):
    print(row["messages"][0]["role"])

import pandas as pd

# Load directly as a DataFrame
df = pd.read_json("hf://datasets/Nanthasit/sakthai-combined-v11/data/train.parquet", lines=True)
print(df.shape)
print(df[["messages", "tools"]].head())

from datasets import load_dataset

# Filter examples by tool count
train = load_dataset("Nanthasit/sakthai-combined-v11", split="train")
multi_tool = train.filter(lambda x: len(x["tools"]) > 1)
print(multi_tool)

# Save to disk as Arrow/Parquet
train.save_to_disk("sakthai-combined-v11-arrow")

Dataset Construction

This dataset combines examples from multiple SakThai-family datasets and is optimized for instruction-tuning small models to handle function-calling and agentic tasks.

Use Cases

• Instruction tuning: Fine-tune small models on tool-calling behavior.
• Function-calling evaluation: Benchmark models' ability to select and invoke tools.
• Agentic training: Train models for multi-step reasoning with tool access.
• Prompt engineering: Use as reference examples for few-shot prompting.

Citation

@dataset{sakthai_combined_v11,
  title     = {SakThai Combined v11},
  author    = {Beer},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Nanthasit/sakthai-combined-v11}
}

License

Apache License 2.0 — Free to use, modify, and distribute with attribution.

See LICENSE for full terms.

---

Links

• Dataset on Hub
• SakThai Model Family Collection
• GitHub Repository

---

Built for SakThai — Beer's AI companion family.