AmanPriyanshu/reasoning-sft-3M-random-compilation

---
license: cc-by-4.0
---

# Reasoning SFT Compilation (3M)

Compiled and rebalanced reasoning SFT dataset. 3,000,000 samples across 283 domains from 20 source datasets.

## Schema

| Column | Type | Description |
|--------|------|-------------|
| `input` | `list[{role: str, content: str}]` | Conversation history with roles: `user`, `system`, `assistant` |
| `response` | `str` | Model response following strict think template |
| `domain` | `str` | Unified domain label |
| `source_dataset` | `str` | HuggingFace source dataset identifier |
| `dataset_license` | `str` | License of the source dataset |

## Response Template

Every response follows exactly:

```
<think>
{reasoning}
</think>
{answer}
```

- Exactly one `<think>` and one `</think>` per response
- Reasoning and answer are both stripped and non-empty
- No think tokens appear in assistant messages within `input`

## Rebalancing

Sampled from a 6.9M row compiled superset. Domains with large counts were capped; structured, tool-use, and code domains were prioritized:

- **Tools**: 84,924 (100% of available)
- **Structured**: 309,801 (100% of available)
- **Code**: 618,456 (capped at 105K per domain)
- **Other**: 1,986,819 (capped at ~100K per domain)

## Sources

| Source | License |
|--------|---------|
| AmanPriyanshu/reasoning-sft-CHIMERA | apache-2.0 |
| AmanPriyanshu/reasoning-sft-Edge-Agent-Reasoning-WebSearch-260K | mit |
| AmanPriyanshu/reasoning-sft-IF_multi_constraints_upto5 | odc-by |
| AmanPriyanshu/reasoning-sft-JSON-structuring-and-correcting | apache-2.0 |
| AmanPriyanshu/reasoning-sft-Nemotron-Cascade-SFT-SWE-210K | cc-by-4.0 |
| AmanPriyanshu/reasoning-sft-Nemotron-Instruction-Following-Chat-v1 | cc-by-4.0 |
| AmanPriyanshu/reasoning-sft-NextCoderDataset-100K | mit |
| AmanPriyanshu/reasoning-sft-One-Shot-CFT-Data-4.7K | apache-2.0 |
| AmanPriyanshu/reasoning-sft-OpenThoughts3-1.2M-450K | apache-2.0 |
| AmanPriyanshu/reasoning-sft-PleIAs-SYNTH-1M | cdla-permissive-2.0 |
| AmanPriyanshu/reasoning-sft-Superior-Reasoning-SFT-gpt-oss-120b-434K | cc-by-4.0 |
| AmanPriyanshu/reasoning-sft-dolci-think-sft-32b-1M | odc-by |
| AmanPriyanshu/reasoning-sft-extract-0 | apache-2.0 |
| AmanPriyanshu/reasoning-sft-github-codereview | mit |
| AmanPriyanshu/reasoning-sft-interstellarninja-json-mode-reasoning-160K | apache-2.0 |
| AmanPriyanshu/reasoning-sft-minimax-microsoft-orca-agentinstruct-1M-v1 | cdla-permissive-2.0 |
| AmanPriyanshu/reasoning-sft-minimax-stratified-kmeans-diverse-reasoning-842K-only | cc-by-4.0 |
| AmanPriyanshu/reasoning-sft-poor-quality-reasoning-sample-mix | apache-2.0 |
| AmanPriyanshu/reasoning-sft-stem-reasoning-complex-FineProofs-126K | apache-2.0 |
| AmanPriyanshu/reasoning-sft-synthetic_text_to_sql-128K | apache-2.0 |

## Files

20 parquet files (`part_00.parquet` through `part_19.parquet`), 150K rows each, ~20GB total. Rows are shuffled randomly (seed=42).