wcpsoft commited on
Commit
982f645
·
verified ·
1 Parent(s): e9228cc

Upload folder using huggingface_hub

Browse files
Files changed (4) hide show
  1. .ms_upload_cache +1 -0
  2. README.md +61 -0
  3. arithmetic_llm.parquet +3 -0
  4. mnist_data.parquet +3 -0
.ms_upload_cache ADDED
@@ -0,0 +1 @@
 
 
1
+ {"version": 3, "repo_id": "wcpsoft/arithmetic-llm", "files": {"./README.md|1786268159.3268275|1798": {"hash": "a3f3b1b828b5de53eb874358cbb4726109b46cd8dc2e2765ba8aa2eb1e1fb00d", "size": 1798, "status": "f", "error_type": "commit_file_invalid"}, "./mnist_data.parquet|1786265150.9011312|30658478": {"hash": "8b6f7fe2d5159014a0c0dce4be55b1169ccfc95b34c815a4255b05b85017b302", "size": 30658478, "status": "f", "error_type": "commit_file_invalid"}, "./arithmetic_llm.parquet|1786268159.261729|32210506": {"hash": "9aa050d7bdbf0521aec8d470e345da97bf36ac7e6ff666eb768126a2032c0935", "size": 32210506, "status": "f", "error_type": "commit_file_invalid"}, "README.md|1786268159.3268275|1798": {"hash": "a3f3b1b828b5de53eb874358cbb4726109b46cd8dc2e2765ba8aa2eb1e1fb00d", "size": 1798, "status": "c"}, "mnist_data.parquet|1786265150.9011312|30658478": {"hash": "8b6f7fe2d5159014a0c0dce4be55b1169ccfc95b34c815a4255b05b85017b302", "size": 30658478, "status": "c"}, "arithmetic_llm.parquet|1786268159.261729|32210506": {"hash": "9aa050d7bdbf0521aec8d470e345da97bf36ac7e6ff666eb768126a2032c0935", "size": 32210506, "status": "c"}}}
README.md ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - zh
4
+ - en
5
+ license: cc-by-sa-4.0
6
+ tags:
7
+ - mnist
8
+ - arithmetic
9
+ - byte-level
10
+ - image-to-text
11
+ task_categories:
12
+ - image-classification
13
+ ---
14
+
15
+ # arithmetic-llm
16
+
17
+ **MNIST 图像 → 字节级算术** 训练数据集。
18
+
19
+ 用于教学向的字节级算术大模型 (`byte-omni-model-zh`):给定一个显式数字 (digit 0-9) 与一张 MNIST 图像 (隐式数字),预测两者之和。
20
+
21
+ ```text
22
+ 输入: digit_byte + MNIST_image_bytes
23
+ 输出: result_bytes
24
+
25
+ 示例:
26
+ digit = "1" + image(数字 "2") → 预测 "3"
27
+ ```
28
+
29
+ ## Schema
30
+
31
+ | 列 | 类型 | 说明 |
32
+ |----|------|------|
33
+ | `id` | int | 序号 |
34
+ | `image_base64` | str | MNIST 图像 PNG (base64) |
35
+ | `pixels` | float[] | 28×28 展平像素 (0-1), 784 维 |
36
+ | `label` | int | 图像数字标签 (0-9) |
37
+ | `split` | str | `train` (59134) / `test` (9943) |
38
+ | `image_hash` | str | SHA-256 逐字节指纹 |
39
+ | `phash` | str | 16×16 感知哈希位串 (与 label 综合去重) |
40
+
41
+ ## 划分
42
+
43
+ | split | 行数 |
44
+ |-------|------|
45
+ | train | 59134 |
46
+ | test | 9943 |
47
+ | 合计 | 69077 |
48
+
49
+ ## 清洗说明 (去重)
50
+
51
+ - 原始 MNIST 无逐字节重复 (SHA-256 重复 = 0)。
52
+ - 但存在大量**视觉近重复** (同图数字, 仅差少量像素)。
53
+ - 采用 **16×16 感知哈希 + label 综合去重**: 键 = `(phash, label)`, 保留首现。
54
+ - 同图同数字 → 视为冗余删除; 同图异数字 → 作为难样本保留。
55
+ - 去重结果: train 60000 → 59134 (删 866), test 10000 → 9943 (删 57)。
56
+
57
+ ## 来源与许可
58
+
59
+ - 原始数据: Yann LeCun 的 MNIST 手写数字数据集 (cc-by-sa-4.0)。
60
+ - 本数据集为加工版本:补充 PNG base64、784 维像素数组、SHA-256 指纹与 16×16 感知哈希,按 train/test 划分并去重。
61
+ - 用法示例见 [byte-omni-model-zh](https://github.com/wcpsoft/byte-omni-model-zh)。
arithmetic_llm.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9aa050d7bdbf0521aec8d470e345da97bf36ac7e6ff666eb768126a2032c0935
3
+ size 32210506
mnist_data.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b6f7fe2d5159014a0c0dce4be55b1169ccfc95b34c815a4255b05b85017b302
3
+ size 30658478