Spaces:
Running
Running
Commit
·
f5894fd
1
Parent(s):
ecd298c
add
Browse files- app.py +1 -1
- results/DeepSeek-R1.json +1 -1
- results/Llama-3.1-70B-Instruct.json +1 -1
- results/Llama-3.1-8B-Instruct.json +1 -1
- results/Llama-3.3-70B-Instruct.json +1 -1
- results/Mistral-7B-Instruct-v0.3.json +1 -1
- results/Mistral-Large-Instruct-2411.json +1 -1
- results/Mistral-Small-Instruct-2409.json +1 -1
- results/QwQ-32B-Preview.json +1 -1
- results/Qwen2.5-32B-Instruct.json +1 -1
- results/Qwen2.5-72B-Instruct.json +1 -1
- results/Qwen2.5-7B-Instruct.json +1 -1
- src/leaderboard/read_evals.py +1 -1
app.py
CHANGED
@@ -64,7 +64,7 @@ def init_leaderboard(dataframe):
|
|
64 |
|
65 |
# pdb.set_trace()
|
66 |
num_cols = dataframe.select_dtypes(include=['number']).columns
|
67 |
-
styler = dataframe.style.format({col: "{:.
|
68 |
return gr.components.Dataframe(
|
69 |
value=styler,
|
70 |
headers=[c.name for c in fields(AutoEvalColumn)],
|
|
|
64 |
|
65 |
# pdb.set_trace()
|
66 |
num_cols = dataframe.select_dtypes(include=['number']).columns
|
67 |
+
styler = dataframe.style.format({col: "{:.1f}" for col in num_cols})
|
68 |
return gr.components.Dataframe(
|
69 |
value=styler,
|
70 |
headers=[c.name for c in fields(AutoEvalColumn)],
|
results/DeepSeek-R1.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "deepseek-ai/DeepSeek-R1",
|
4 |
"link": "https://huggingface.co/deepseek-ai/DeepSeek-R1",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "deepseek-ai/DeepSeek-R1",
|
4 |
"link": "https://huggingface.co/deepseek-ai/DeepSeek-R1",
|
5 |
+
"Params": 671
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Llama-3.1-70B-Instruct.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "meta-llama/Llama-3.1-70B-Instruct",
|
4 |
"link": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "meta-llama/Llama-3.1-70B-Instruct",
|
4 |
"link": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct",
|
5 |
+
"Params": 70
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Llama-3.1-8B-Instruct.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "meta-llama/Llama-3.1-8B-Instruct",
|
4 |
"link": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "meta-llama/Llama-3.1-8B-Instruct",
|
4 |
"link": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct",
|
5 |
+
"Params": 8
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Llama-3.3-70B-Instruct.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "meta-llama/Llama-3.3-70B-Instruct",
|
4 |
"link": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "meta-llama/Llama-3.3-70B-Instruct",
|
4 |
"link": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct",
|
5 |
+
"Params": 70
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Mistral-7B-Instruct-v0.3.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "mistralai/Mistral-7B-Instruct-v0.3",
|
4 |
"link": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "mistralai/Mistral-7B-Instruct-v0.3",
|
4 |
"link": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3",
|
5 |
+
"Params": 7
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Mistral-Large-Instruct-2411.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "mistralai/Mistral-Large-Instruct-2411",
|
4 |
"link": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "mistralai/Mistral-Large-Instruct-2411",
|
4 |
"link": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411",
|
5 |
+
"Params": 123
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Mistral-Small-Instruct-2409.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "mistralai/Mistral-Small-Instruct-2409",
|
4 |
"link": "https://huggingface.co/mistralai/Mistral-Small-Instruct-2409",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "mistralai/Mistral-Small-Instruct-2409",
|
4 |
"link": "https://huggingface.co/mistralai/Mistral-Small-Instruct-2409",
|
5 |
+
"Params": 22
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/QwQ-32B-Preview.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/QwQ-32B-Preview",
|
4 |
"link": "https://huggingface.co/Qwen/QwQ-32B-Preview",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/QwQ-32B-Preview",
|
4 |
"link": "https://huggingface.co/Qwen/QwQ-32B-Preview",
|
5 |
+
"Params": 32
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Qwen2.5-32B-Instruct.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/Qwen2.5-32B-Instruct",
|
4 |
"link": "https://huggingface.co/Qwen/Qwen2.5-32B-Instruct",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/Qwen2.5-32B-Instruct",
|
4 |
"link": "https://huggingface.co/Qwen/Qwen2.5-32B-Instruct",
|
5 |
+
"Params": 32
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Qwen2.5-72B-Instruct.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/Qwen2.5-72B-Instruct",
|
4 |
"link": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/Qwen2.5-72B-Instruct",
|
4 |
"link": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct",
|
5 |
+
"Params": 72
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
results/Qwen2.5-7B-Instruct.json
CHANGED
@@ -2,7 +2,7 @@
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/Qwen2.5-7B-Instruct",
|
4 |
"link": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct",
|
5 |
-
"Params":
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
|
|
2 |
"config": {
|
3 |
"model_name": "Qwen/Qwen2.5-7B-Instruct",
|
4 |
"link": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct",
|
5 |
+
"Params": 7
|
6 |
},
|
7 |
"results": {
|
8 |
"Overall": {
|
src/leaderboard/read_evals.py
CHANGED
@@ -28,7 +28,7 @@ class EvalResult:
|
|
28 |
# architecture: str = "Unknown"
|
29 |
# license: str = "?"
|
30 |
# likes: int = 0
|
31 |
-
num_params:
|
32 |
# date: str = "" # submission date of request file
|
33 |
# still_on_hub: bool = False
|
34 |
link: str = ''
|
|
|
28 |
# architecture: str = "Unknown"
|
29 |
# license: str = "?"
|
30 |
# likes: int = 0
|
31 |
+
num_params: int = None
|
32 |
# date: str = "" # submission date of request file
|
33 |
# still_on_hub: bool = False
|
34 |
link: str = ''
|