leonard-dls
commited on
Commit
·
d126096
1
Parent(s):
6136624
change model order
Browse files
app.py
CHANGED
@@ -11,8 +11,8 @@ with open("phi4_gsm8k_output.jsonl", "r") as file:
|
|
11 |
phi4_dict = [json.loads(line) for line in file]
|
12 |
|
13 |
models_data = {
|
|
|
14 |
"Qwen/Qwen2.5-14B" : qwen_dict,
|
15 |
-
"microsoft/phi-4" : phi4_dict
|
16 |
}
|
17 |
|
18 |
starting_index = 0
|
@@ -26,8 +26,8 @@ This space aims to partially reproduce this work.
|
|
26 |
|
27 |
I chose to look at the contamination of **Qwen/Qwen2.5-14B** and **microsoft/phi-4** by **GSM8K** dataset.
|
28 |
|
29 |
-
For **Qwen/Qwen2.5-14B** I found **729** GSM8K examples that had a least a 0.9 text similarity ratio between generated and original.
|
30 |
For **microsoft/phi-4** I found **172** GSM8K examples that had a least a 0.9 text similarity ratio between generated and original.
|
|
|
31 |
|
32 |
"""
|
33 |
|
|
|
11 |
phi4_dict = [json.loads(line) for line in file]
|
12 |
|
13 |
models_data = {
|
14 |
+
"microsoft/phi-4" : phi4_dict,
|
15 |
"Qwen/Qwen2.5-14B" : qwen_dict,
|
|
|
16 |
}
|
17 |
|
18 |
starting_index = 0
|
|
|
26 |
|
27 |
I chose to look at the contamination of **Qwen/Qwen2.5-14B** and **microsoft/phi-4** by **GSM8K** dataset.
|
28 |
|
|
|
29 |
For **microsoft/phi-4** I found **172** GSM8K examples that had a least a 0.9 text similarity ratio between generated and original.
|
30 |
+
For **Qwen/Qwen2.5-14B** I found **729** GSM8K examples that had a least a 0.9 text similarity ratio between generated and original.
|
31 |
|
32 |
"""
|
33 |
|