AAUGS/Tools

1---2license: cc-by-4.03language:4- en5tags:6- function-calling7- tool-use8- bfcl9- agentic10pretty_name: Function-calling fine-tune data (xLAM + ToolACE + Java/JS)11configs:12- config_name: python-headline13  data_files: train.qwen.messages.jsonl14- config_name: javajs-augmented15  data_files: train.javajs.messages.jsonl16---17 18# Function-calling fine-tune data (xLAM + ToolACE + Java/JS)19 20Training data for fine-tuning small language models on function calling, built against the21BFCL non-live AST categories. OpenAI/ChatML `messages` schema; Pythonic `[func(arg=val)]`22labels.23 24## Configs25 26| Config | File | Examples | Focus |27|---|---|---|---|28| `python-headline` | `train.qwen.messages.jsonl` | 140,080 | xLAM Python backbone + ToolACE + synthetic irrelevance + parallel up-weight |29| `javajs-augmented` | `train.javajs.messages.jsonl` | 170,077 | the above **+ 15k Java + 15k JS** (synthesized) — adds Java/JS function calling |30 31## Schema32 33Each line:34`{"messages": [{"role":"system",...}, {"role":"user",...}, {"role":"assistant","content":"[func(arg=val), ...]"}]}`35 36Tools are injected into the system message as `List of tools: [...]`; the assistant label is a37Pythonic call list (or `[]` when no tool applies). No raw chat-template tokens (the trainer38applies the template); loss is on the assistant turn only.39 40## Sources & attribution41 42- **Salesforce xLAM / APIGen** (function-calling 60k) via the `argilla/apigen-function-calling`43  mirror — **CC-BY-4.0**. The backbone plus the synthesized irrelevance, the parallel up-weight,44  and the Java/JS transforms all derive from it. **Attribution: Salesforce / xLAM (APIGen).**45- **Team-ACE/ToolACE** — **Apache-2.0**. The AST-validated Pythonic calls and the46  refusal → `[]` (relevance-detection) slices.47- Deduplicated against the `gorilla-llm/Berkeley-Function-Calling-Leaderboard` test set48  (contamination guard).49 50## Construction51 52Built from the sources above by deterministic transformation (no LLM): conversion to Pythonic53call lists, hard-negative irrelevance synthesis (pairing queries with mismatched toolsets),54parallel-call up-weighting, and a Python → Java/JS idiom transformation (`Class.method` /55camelCase naming, language-specific type mapping, AST-validated). Seeded and deduplicated56against the BFCL test set, so the build is reproducible.57 58## License59 60`cc-by-4.0`, governed by the xLAM backbone (attribution required); ToolACE portions are61Apache-2.0. Please provide per-source attribution as above.62