Datasets:
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- text-generation
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
- zh
|
| 8 |
+
tags:
|
| 9 |
+
- code
|
| 10 |
+
- python
|
| 11 |
+
- sft
|
| 12 |
+
- tool-use
|
| 13 |
+
- reasoning
|
| 14 |
+
size_categories:
|
| 15 |
+
- 1M<n<10M
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# Qwen3-Coder SFT Dataset
|
| 19 |
+
|
| 20 |
+
用于微调 Qwen3-Coder-Next 的 Python 编程 SFT 数据集。
|
| 21 |
+
|
| 22 |
+
## 构成
|
| 23 |
+
- **train.jsonl**: 2,361,724 条
|
| 24 |
+
- **val.jsonl**: 124,302 条
|
| 25 |
+
- 标准 `messages` 格式(system/user/assistant/tool 角色),部分含 `tools` 字段
|
| 26 |
+
- 含工具调用 ~7.1%,`<think>` 推理 ~26%,中文指令 ~13%
|
| 27 |
+
|
| 28 |
+
## 格式
|
| 29 |
+
```json
|
| 30 |
+
{"messages": [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
|
| 31 |
+
```
|
| 32 |
+
含工具调用的记录额外带 `tools` 字段(OpenAI function-calling schema)。
|
| 33 |
+
|
| 34 |
+
## 加载
|
| 35 |
+
```python
|
| 36 |
+
from datasets import load_dataset
|
| 37 |
+
ds = load_dataset("adasdadsd/Qwen3-Coder-SFT-2.4M", data_files={"train": "train.jsonl", "validation": "val.jsonl"})
|
| 38 |
+
```
|