astra-augment 0.6.0

SFT data augmentation via conversation truncation

astra-augment

SFT data augmentation via conversation truncation for tool-calling trajectories.

Part of the Astra ecosystem — a data factory for high-quality multi-turn tool-calling conversation trajectories.

Install

pip install astra-augment

Usage

# Expand last 20% of tool calls into separate training samples
astra-augment expand input.jsonl -o output.jsonl --ratio 0.2 --mode tool_call

# Expand last 40% of assistant responses
astra-augment expand input.jsonl -o output.jsonl --ratio 0.4 --mode response

Modes

• tool_call: Truncates conversations at assistant tool-call positions (skips those followed by failed responses)
• response: Truncates at assistant response positions (excludes the final one to avoid duplicating the original)

Parameters

Parameter	Description
--ratio	Fraction of tail positions to expand (0, 1]
--mode	Truncation mode: tool_call or response
--format	Dataset format (default: qwen3)
-o	Output JSONL path

Development

# Clone and set up
git clone https://github.com/zhangdw156/astra-augment.git
cd astra-augment
uv sync --all-groups

# Run tests
uv run pytest

# Lint
uv run ruff check .

License

MIT