AmanPriyanshu/tool-reasoning-sft-1M-random-compilation

---
task_categories:
- text-generation
language:
- en
- zh
tags:
- reasoning
- tool-calling
- agentic
- multi-turn
license: apache-2.0
size_categories:
- 1M<n<10M
---

# Tool Reasoning SFT 1M Random Compilation

1M multi-turn tool-use and agentic reasoning trajectories compiled from 25 public datasets. Equitable water-fill sampling across categories and internal splits.

## Schema

| Column | Type | Description |
|---|---|---|
| `messages` | string | JSON list of `{role, content}` turns |
| `source_dataset` | string | Source dataset identifier |
| `source_category` | string | `TOOLS`, `CODING`, or `RESEARCH` |
| `source_split` | string | Split within the source dataset |
| `num_turns` | int32 | Total messages in the conversation |
| `num_tool_calls` | int32 | Number of tool call messages |
| `num_user_messages` | int32 | Number of user messages |

## Category Distribution

| Category | Samples |
|---|---|
| TOOLS | 375,000 |
| CODING | 375,000 |
| RESEARCH | 250,000 |
| **Total** | **1,000,000** |