File size: 1,512 Bytes
dd8a712
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
# 懂推荐

懂推荐任务的 baseline SID 对齐数据。

## 文件

| 文件 | 说明 |
| --- | --- |
| `baseline_caption_tag_lists.parquet` | 每条 baseline SFT 样本一行,保留原始 `messages`,以及从 `messages` 中抽取出的 SID token 序列和对应的 caption/tag 序列 |

## 数据规模

- 行数:19,204
- SID token 位置数:3,539,794
- 文件大小:约 696 MB

## 字段

| 字段 | 类型 | 说明 |
| --- | --- | --- |
| `record_id` | int64 | 样本行 ID,范围为 `0 - 19203` |
| `messages` | string | 原始 SFT messages JSON 字符串,通常包含 `system/user/assistant` 三条消息 |
| `sid_token_list` | list[string] | 从 `messages` 中按出现顺序抽取出的 SID token |
| `caption_list` | list[string] | 与 `sid_token_list` 位置对齐的 caption;未找到时为 null |
| `tag_list` | list[string] | 与 `sid_token_list` 位置对齐的 tag;未找到时为 null |

## 对齐规则

三个 list 字段按位置一一对应:

```text
sid_token_list[i] <-> caption_list[i] <-> tag_list[i]
```

例如第 `i` 个 SID token 的文本描述为 `caption_list[i]`,三级类目为 `tag_list[i]`。如果 caption 或 tag 没有匹配到,对应位置保留 null,不改变 list 长度和顺序。

## 覆盖率

| 指标 | 数量 | 占 SID 位置比例 |
| --- | ---: | ---: |
| caption hit | 3,478,100 | 98.26% |
| caption null | 61,694 | 1.74% |
| tag hit | 1,356,390 | 38.32% |
| tag null | 2,183,404 | 61.68% |

注意:goods 域 tag 为 100% null。