Hugging Face's logo

fdyrd
/
QwenMath-0.5B

Text Generation
Transformers
Safetensors
English
qwen2
text-generation-inference
conversational
Model card Files
xet
Community
2

Improve language tag

#2
by lbourdois - opened
base: refs/heads/main
←
from: refs/pr/2
Discussion Files changed
+142
-130
Files changed (1) hide show
  1. README.md +142 -130
README.md CHANGED
@@ -1,131 +1,143 @@
1
- ---
2
- language:
3
- - en
4
- license: mit
5
- datasets:
6
- - fdyrd/MATH
7
- base_model:
8
- - Qwen/Qwen2.5-0.5B
9
- library_name: transformers
10
- tags:
11
- - text-generation-inference
12
- metrics:
13
- - accuracy
14
- ---
15
-
16
- # QwenMath
17
-
18
- A generation LLM which can solve math problems.
19
-
20
- ## Training Statistics
21
- ```yaml
22
- training-method: lora
23
- training-time: "5:42"
24
- data-size: 500
25
- epoch: 3
26
- total_flos: "1372250GF"
27
- train_loss: 0.6441
28
- train_samples_per_second: 4.385
29
- train_steps_per_second: 0.544
30
- ```
31
-
32
- ## Validation Set Performance
33
- Dataset used: test split of [fdyrd/MATH](https://huggingface.co/datasets/fdyrd/MATH).
34
- Metric: accuracy
35
-
36
- <table>
37
- <tr>
38
- <th> Level </th>
39
- <th> Algebra </th>
40
- <th> Intermediate Algebra </th>
41
- <th> Prealgebra </th>
42
- <th> Precalculus </th>
43
- <th> Number Theory </th>
44
- <th> Geometry </th>
45
- <th> Counting & Probability </th>
46
- <th> Average </th>
47
- </tr>
48
- <tr>
49
- <td> Level 1 </td>
50
- <td> 0.541 : 135 </td>
51
- <td> 0.192 : 52 </td>
52
- <td> 0.477 : 86 </td>
53
- <td> 0.228 : 57 </td>
54
- <td> 0.467 : 30 </td>
55
- <td> 0.263 : 38 </td>
56
- <td> 0.359 : 39 </td>
57
- <td> 0.361 </td>
58
- </tr>
59
- <tr>
60
- <td> Level 2 </td>
61
- <td> 0.323 : 201 </td>
62
- <td> 0.109 : 128 </td>
63
- <td> 0.367 : 177 </td>
64
- <td> 0.044 : 113 </td>
65
- <td> 0.38 : 92 </td>
66
- <td> 0.134 : 82 </td>
67
- <td> 0.248 : 101 </td>
68
- <td> 0.229 </td>
69
- </tr>
70
- <tr>
71
- <td> Level 3 </td>
72
- <td> 0.291 : 261 </td>
73
- <td> 0.046 : 195 </td>
74
- <td> 0.308 : 224 </td>
75
- <td> 0.0 : 127 </td>
76
- <td> 0.262 : 122 </td>
77
- <td> 0.088 : 102 </td>
78
- <td> 0.16 : 100 </td>
79
- <td> 0.165 </td>
80
- </tr>
81
- <tr>
82
- <td> Level 4 </td>
83
- <td> 0.18 : 283 </td>
84
- <td> 0.024 : 248 </td>
85
- <td> 0.22 : 191 </td>
86
- <td> 0.009 : 114 </td>
87
- <td> 0.169 : 142 </td>
88
- <td> 0.064 : 125 </td>
89
- <td> 0.09 : 111 </td>
90
- <td> 0.108 </td>
91
- </tr>
92
- <tr>
93
- <td> Level 5 </td>
94
- <td> 0.088 : 307 </td>
95
- <td> 0.004 : 280 </td>
96
- <td> 0.104 : 193 </td>
97
- <td> 0.0 : 135 </td>
98
- <td> 0.136 : 154 </td>
99
- <td> 0.023 : 132 </td>
100
- <td> 0.065 : 123 </td>
101
- <td> 0.06 </td>
102
- </tr>
103
- <tr>
104
- <td> Average </td>
105
- <td> 0.285 </td>
106
- <td> 0.075 </td>
107
- <td> 0.295 </td>
108
- <td> 0.056 </td>
109
- <td> 0.283 </td>
110
- <td> 0.114 </td>
111
- <td> 0.184 </td>
112
- <td> 0.166 </td>
113
- </tr>
114
- </table>
115
-
116
- ## Test Set Performance
117
-
118
- ```json
119
- [
120
- {
121
- "dataset": "MATH500",
122
- "url": "https://huggingface.co/datasets/qq8933/MATH500",
123
- "accuracy": 0.286
124
- },
125
- {
126
- "dataset": "GSM8K",
127
- "url": "https://huggingface.co/datasets/openai/gsm8k",
128
- "accuracy": 0.382
129
- }
130
- ]
 
 
 
 
 
 
 
 
 
 
 
 
131
  ```
 
1
+ ---
2
+ language:
3
+ - zho
4
+ - eng
5
+ - fra
6
+ - spa
7
+ - por
8
+ - deu
9
+ - ita
10
+ - rus
11
+ - jpn
12
+ - kor
13
+ - vie
14
+ - tha
15
+ - ara
16
+ license: mit
17
+ datasets:
18
+ - fdyrd/MATH
19
+ base_model:
20
+ - Qwen/Qwen2.5-0.5B
21
+ library_name: transformers
22
+ tags:
23
+ - text-generation-inference
24
+ metrics:
25
+ - accuracy
26
+ ---
27
+
28
+ # QwenMath
29
+
30
+ A generation LLM which can solve math problems.
31
+
32
+ ## Training Statistics
33
+ ```yaml
34
+ training-method: lora
35
+ training-time: "5:42"
36
+ data-size: 500
37
+ epoch: 3
38
+ total_flos: "1372250GF"
39
+ train_loss: 0.6441
40
+ train_samples_per_second: 4.385
41
+ train_steps_per_second: 0.544
42
+ ```
43
+
44
+ ## Validation Set Performance
45
+ Dataset used: test split of [fdyrd/MATH](https://huggingface.co/datasets/fdyrd/MATH).
46
+ Metric: accuracy
47
+
48
+ <table>
49
+ <tr>
50
+ <th> Level </th>
51
+ <th> Algebra </th>
52
+ <th> Intermediate Algebra </th>
53
+ <th> Prealgebra </th>
54
+ <th> Precalculus </th>
55
+ <th> Number Theory </th>
56
+ <th> Geometry </th>
57
+ <th> Counting & Probability </th>
58
+ <th> Average </th>
59
+ </tr>
60
+ <tr>
61
+ <td> Level 1 </td>
62
+ <td> 0.541 : 135 </td>
63
+ <td> 0.192 : 52 </td>
64
+ <td> 0.477 : 86 </td>
65
+ <td> 0.228 : 57 </td>
66
+ <td> 0.467 : 30 </td>
67
+ <td> 0.263 : 38 </td>
68
+ <td> 0.359 : 39 </td>
69
+ <td> 0.361 </td>
70
+ </tr>
71
+ <tr>
72
+ <td> Level 2 </td>
73
+ <td> 0.323 : 201 </td>
74
+ <td> 0.109 : 128 </td>
75
+ <td> 0.367 : 177 </td>
76
+ <td> 0.044 : 113 </td>
77
+ <td> 0.38 : 92 </td>
78
+ <td> 0.134 : 82 </td>
79
+ <td> 0.248 : 101 </td>
80
+ <td> 0.229 </td>
81
+ </tr>
82
+ <tr>
83
+ <td> Level 3 </td>
84
+ <td> 0.291 : 261 </td>
85
+ <td> 0.046 : 195 </td>
86
+ <td> 0.308 : 224 </td>
87
+ <td> 0.0 : 127 </td>
88
+ <td> 0.262 : 122 </td>
89
+ <td> 0.088 : 102 </td>
90
+ <td> 0.16 : 100 </td>
91
+ <td> 0.165 </td>
92
+ </tr>
93
+ <tr>
94
+ <td> Level 4 </td>
95
+ <td> 0.18 : 283 </td>
96
+ <td> 0.024 : 248 </td>
97
+ <td> 0.22 : 191 </td>
98
+ <td> 0.009 : 114 </td>
99
+ <td> 0.169 : 142 </td>
100
+ <td> 0.064 : 125 </td>
101
+ <td> 0.09 : 111 </td>
102
+ <td> 0.108 </td>
103
+ </tr>
104
+ <tr>
105
+ <td> Level 5 </td>
106
+ <td> 0.088 : 307 </td>
107
+ <td> 0.004 : 280 </td>
108
+ <td> 0.104 : 193 </td>
109
+ <td> 0.0 : 135 </td>
110
+ <td> 0.136 : 154 </td>
111
+ <td> 0.023 : 132 </td>
112
+ <td> 0.065 : 123 </td>
113
+ <td> 0.06 </td>
114
+ </tr>
115
+ <tr>
116
+ <td> Average </td>
117
+ <td> 0.285 </td>
118
+ <td> 0.075 </td>
119
+ <td> 0.295 </td>
120
+ <td> 0.056 </td>
121
+ <td> 0.283 </td>
122
+ <td> 0.114 </td>
123
+ <td> 0.184 </td>
124
+ <td> 0.166 </td>
125
+ </tr>
126
+ </table>
127
+
128
+ ## Test Set Performance
129
+
130
+ ```json
131
+ [
132
+ {
133
+ "dataset": "MATH500",
134
+ "url": "https://huggingface.co/datasets/qq8933/MATH500",
135
+ "accuracy": 0.286
136
+ },
137
+ {
138
+ "dataset": "GSM8K",
139
+ "url": "https://huggingface.co/datasets/openai/gsm8k",
140
+ "accuracy": 0.382
141
+ }
142
+ ]
143
  ```