Spaces:
Sleeping
Sleeping
update
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- .gitattributes +13 -0
- app.py +136 -509
- results-cot/CodeLlama-70b-Instruct-hf.pkl +0 -3
- results-cot/Mixtral-8x7B-Instruct-v0.1.csv +0 -3
- results-cot/Mixtral-8x7B-Instruct-v0.1.pkl +0 -3
- results-cot/Mixtral-8x7B-Instruct-v0.1.png +0 -3
- results-cot/Qwen1.5-72B-Chat.csv +0 -3
- results-cot/Qwen1.5-72B-Chat.jpg +0 -3
- results-cot/Qwen1.5-72B-Chat.pkl +0 -3
- results-cot/Qwen1.5-72B-Chat.png +0 -3
- results-cot/claude-3-sonnet-20240229.csv +0 -3
- results-cot/claude-3-sonnet-20240229.jpg +0 -3
- results-cot/claude-3-sonnet-20240229.pkl +0 -3
- results-cot/claude-3-sonnet-20240229.png +0 -3
- results-cot/dbrx-instruct.csv +0 -3
- results-cot/deepseek-llm-67b-chat.csv +0 -3
- results-cot/deepseek-llm-67b-chat.jpg +0 -3
- results-cot/deepseek-llm-67b-chat.pkl +0 -3
- results-cot/deepseek-llm-67b-chat.png +0 -3
- results-cot/gemini-pro.csv +0 -3
- results-cot/gemini-pro.jpg +0 -3
- results-cot/gemini-pro.pkl +0 -3
- results-cot/gemini-pro.png +0 -3
- results-cot/gemma-7b-it.csv +0 -3
- results-cot/gemma-7b-it.jpg +0 -3
- results-cot/gemma-7b-it.pkl +0 -3
- results-cot/gemma-7b-it.png +0 -3
- results-cot/gpt-3.5-turbo-0125.csv +0 -3
- results-cot/gpt-3.5-turbo-0125.jpg +0 -3
- results-cot/gpt-3.5-turbo-0125.pkl +0 -3
- results-cot/gpt-3.5-turbo-0125.png +0 -3
- results-cot/gpt-4-turbo-2024-04-09.csv +0 -3
- results-cot/gpt-4-turbo-2024-04-09.jpg +0 -3
- results-cot/gpt-4-turbo-2024-04-09.pkl +0 -3
- results-cot/gpt-4-turbo-2024-04-09.png +0 -3
- results-vision/claude-3-opus-20240229.csv +0 -3
- results-vision/claude-3-opus-20240229.jpg +0 -3
- results-vision/claude-3-opus-20240229.pkl +0 -3
- results-vision/claude-3-opus-20240229.png +0 -3
- results-vision/claude-3-opus-vision.jpg +0 -3
- results-vision/claude-3-opus-vision.pkl +0 -3
- results-vision/claude-3-opus-vision.png +0 -3
- results-vision/gemini-pro-vision.csv +0 -3
- results-vision/gemini-pro-vision.jpg +0 -3
- results-vision/gemini-pro-vision.pkl +0 -3
- results-vision/gemini-pro-vision.png +0 -3
- results-vision/gpt-4v.jpg +0 -3
- results-vision/gpt-4v.pkl +0 -3
- results-vision/gpt-4v.png +0 -3
- results/CodeLlama-70b-Instruct-hf.csv +0 -3
.gitattributes
CHANGED
@@ -271,3 +271,16 @@ results_qwen/Llama-3-70b-chat-hf.jpg filter=lfs diff=lfs merge=lfs -text
|
|
271 |
results_qwen/gpt-4.csv filter=lfs diff=lfs merge=lfs -text
|
272 |
results_qwen/gpt-4.jpg filter=lfs diff=lfs merge=lfs -text
|
273 |
results_qwen/Llama-3-70b-chat-hf.pkl filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
271 |
results_qwen/gpt-4.csv filter=lfs diff=lfs merge=lfs -text
|
272 |
results_qwen/gpt-4.jpg filter=lfs diff=lfs merge=lfs -text
|
273 |
results_qwen/Llama-3-70b-chat-hf.pkl filter=lfs diff=lfs merge=lfs -text
|
274 |
+
all_results.pkl filter=lfs diff=lfs merge=lfs -text
|
275 |
+
results/Llama-3-70b-chat-hf.png filter=lfs diff=lfs merge=lfs -text
|
276 |
+
results/dbrx-instruct.png filter=lfs diff=lfs merge=lfs -text
|
277 |
+
results/gpt-3.5-0613.png filter=lfs diff=lfs merge=lfs -text
|
278 |
+
results/gpt-4-1106.png filter=lfs diff=lfs merge=lfs -text
|
279 |
+
results/Llama-3-70b-chat-hf.jpg filter=lfs diff=lfs merge=lfs -text
|
280 |
+
results/dbrx-instruct.jpg filter=lfs diff=lfs merge=lfs -text
|
281 |
+
results/gpt-3.5-0613.jpg filter=lfs diff=lfs merge=lfs -text
|
282 |
+
results/gpt-4-1106.jpg filter=lfs diff=lfs merge=lfs -text
|
283 |
+
results/gpt-4-1106.pkl filter=lfs diff=lfs merge=lfs -text
|
284 |
+
results/Llama-3-70b-chat-hf.pkl filter=lfs diff=lfs merge=lfs -text
|
285 |
+
results/dbrx-instruct.pkl filter=lfs diff=lfs merge=lfs -text
|
286 |
+
results/gpt-3.5-0613.pkl filter=lfs diff=lfs merge=lfs -text
|
app.py
CHANGED
@@ -1,98 +1,49 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import pandas as pd
|
3 |
-
from glob import glob
|
4 |
-
import matplotlib.pyplot as plt
|
5 |
-
import seaborn as sns
|
6 |
-
from matplotlib.colors import ListedColormap, BoundaryNorm
|
7 |
-
from glob import glob
|
8 |
import os
|
|
|
9 |
|
10 |
-
|
11 |
import matplotlib.pyplot as plt
|
12 |
-
import seaborn as sns
|
13 |
-
from matplotlib.colors import ListedColormap, BoundaryNorm
|
14 |
import pandas as pd
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
noncot_results_qwen = glob("results_qwen/*.pkl")
|
20 |
-
# Load vision benchmark results
|
21 |
-
vision_results = glob("results-vision/*.pkl")
|
22 |
-
# Load CoT text benchmark results
|
23 |
-
cot_text_results = glob("results-cot/*.pkl")
|
24 |
-
# Load CoT vision benchmark results
|
25 |
-
# cot_vision_results = glob("results-vision-CoT/*.pkl")
|
26 |
-
|
27 |
-
|
28 |
-
# Function to load data, add model type and name
|
29 |
-
def load_data(files, model_type):
|
30 |
-
data = []
|
31 |
-
for file in files:
|
32 |
-
df = pd.read_pickle(file)
|
33 |
-
df["Model Type"] = model_type
|
34 |
-
df["Model Name"] = file.split("/")[-1].replace(".pkl", "")
|
35 |
-
data.append(df)
|
36 |
-
return pd.concat(data, ignore_index=True)
|
37 |
-
|
38 |
-
|
39 |
-
# Load and label all data
|
40 |
-
data = load_data(noncot_results, "Text Only")
|
41 |
-
data_qwen = load_data(noncot_results_qwen, "Text Only")
|
42 |
-
vision_data = load_data(vision_results, "Vision")
|
43 |
-
cot_text_data = load_data(cot_text_results, "CoT Text Only")
|
44 |
-
# cot_vision_data = load_data(cot_vision_results, "CoT Vision")
|
45 |
-
|
46 |
-
# Combine all data into a single DataFrame
|
47 |
-
all_data = pd.concat([data_qwen, vision_data, cot_text_data], ignore_index=True)
|
48 |
-
|
49 |
-
all_model_names = all_data["Model Name"].unique()
|
50 |
-
all_text_only_model_names = list(
|
51 |
-
all_data[all_data["Model Type"] == "Text Only"]["Model Name"].unique()
|
52 |
-
)
|
53 |
-
all_cot_text_only_models = list(
|
54 |
-
all_data[all_data["Model Type"] == "CoT Text Only"]["Model Name"].unique()
|
55 |
-
)
|
56 |
-
|
57 |
-
|
58 |
-
text_only_filtered_raw = None
|
59 |
-
text_only_filtered_raw_cot = None
|
60 |
-
|
61 |
-
## Continue with the cold code --
|
62 |
-
# TODO: Update me to read from all_data for later
|
63 |
-
|
64 |
-
|
65 |
-
# Load the csv files into a dict with keys being name of the file and values being the data
|
66 |
-
data = {file: pd.read_pickle(file) for file in noncot_results}
|
67 |
-
# Load the vision files into a dict
|
68 |
-
vision_data = {file: pd.read_pickle(file) for file in vision_results}
|
69 |
-
# Load the CoT text files into a dict
|
70 |
-
cot_text_data = {file: pd.read_pickle(file) for file in cot_text_results}
|
71 |
-
# Load the CoT vision files into a dict
|
72 |
-
# cot_vision_data = {file: pd.read_pickle(file) for file in cot_vision_results}
|
73 |
-
|
74 |
-
data_qwen = {file: pd.read_pickle(file) for file in noncot_results_qwen}
|
75 |
-
|
76 |
-
|
77 |
-
intersection_df = pd.read_pickle(
|
78 |
-
"./intersection_results/gpt-3.5-judge-by_Qwen_5times_intersection_subset_1.pkl"
|
79 |
-
)
|
80 |
-
# accuracy for each model
|
81 |
-
intersection_df_acc = (
|
82 |
-
intersection_df.groupby("model_name")["parsed_judge_response"].mean().reset_index()
|
83 |
-
)
|
84 |
-
intersection_df_acc["Accuracy"] = intersection_df_acc["parsed_judge_response"] * 100
|
85 |
-
intersection_df_acc.drop("parsed_judge_response", axis=1, inplace=True)
|
86 |
-
intersection_df_acc.sort_values("Accuracy", ascending=False, inplace=True)
|
87 |
-
|
88 |
-
|
89 |
-
def calculate_accuracy(df):
|
90 |
-
return df["parsed_judge_response"].mean() * 100
|
91 |
|
92 |
|
93 |
-
|
94 |
-
# 4 level accuracy
|
95 |
-
return (df.groupby("difficulty_level")["parsed_judge_response"].mean() * 100).values
|
96 |
|
97 |
|
98 |
# Define the column names with icons
|
@@ -114,450 +65,126 @@ column_names = [
|
|
114 |
"Level 4 Accuracy",
|
115 |
]
|
116 |
|
117 |
-
|
118 |
-
# Function to process data
|
119 |
-
def process_data(data):
|
120 |
-
data_for_df = []
|
121 |
-
for file, df in data.items():
|
122 |
-
overall_accuracy = round(calculate_accuracy(df), 2)
|
123 |
-
breakdown_accuracy = [round(acc, 2) for acc in accuracy_breakdown(df)]
|
124 |
-
model_name = file.split("/")[-1].replace(".pkl", "")
|
125 |
-
data_for_df.append([model_name, overall_accuracy] + breakdown_accuracy)
|
126 |
-
return data_for_df
|
127 |
-
|
128 |
-
|
129 |
-
# Process all data
|
130 |
-
text_data_for_df = process_data(data)
|
131 |
-
text_data_for_df_qwen = process_data(data_qwen)
|
132 |
-
|
133 |
-
vision_data_for_df = process_data(vision_data)
|
134 |
-
cot_text_data_for_df = process_data(cot_text_data)
|
135 |
-
# cot_vision_data_for_df = process_data(cot_vision_data)
|
136 |
-
|
137 |
-
# Create DataFrames
|
138 |
-
accuracy_df = pd.DataFrame(text_data_for_df, columns=column_names)
|
139 |
-
accuracy_df_qwen = pd.DataFrame(text_data_for_df_qwen, columns=column_names)
|
140 |
-
vision_accuracy_df = pd.DataFrame(vision_data_for_df, columns=column_names)
|
141 |
-
cot_text_accuracy_df = pd.DataFrame(cot_text_data_for_df, columns=column_names)
|
142 |
-
# cot_vision_accuracy_df = pd.DataFrame(cot_vision_data_for_df, columns=column_names)
|
143 |
-
|
144 |
-
|
145 |
-
# Function to finalize DataFrame
|
146 |
-
def finalize_df(df):
|
147 |
-
df = df.round(1) # Round to one decimal place
|
148 |
-
df = df.applymap(lambda x: f"{x:.1f}" if isinstance(x, (int, float)) else x)
|
149 |
-
df.columns = headers_with_icons
|
150 |
-
df.sort_values(by="⭐ Overall", ascending=False, inplace=True)
|
151 |
-
# add a new column with the order (index)
|
152 |
-
df["#"] = range(1, len(df) + 1)
|
153 |
-
# bring rank to the first column
|
154 |
-
cols = df.columns.tolist()
|
155 |
-
cols = cols[-1:] + cols[:-1]
|
156 |
-
df = df[cols]
|
157 |
-
|
158 |
-
return df
|
159 |
-
|
160 |
-
|
161 |
-
# Finalize all DataFrames
|
162 |
-
accuracy_df = finalize_df(accuracy_df)
|
163 |
-
accuracy_df_qwen = finalize_df(accuracy_df_qwen)
|
164 |
-
vision_accuracy_df = finalize_df(vision_accuracy_df)
|
165 |
-
cot_text_accuracy_df = finalize_df(cot_text_accuracy_df)
|
166 |
-
# cot_vision_accuracy_df = finalize_df(cot_vision_accuracy_df)
|
167 |
-
|
168 |
-
|
169 |
def load_heatmap(evt: gr.SelectData):
|
170 |
heatmap_image = gr.Image(f"results/{evt.value}.jpg")
|
171 |
return heatmap_image
|
172 |
|
173 |
|
174 |
-
def load_heatmap_qwen(evt: gr.SelectData):
|
175 |
-
heatmap_image = gr.Image(f"results_qwen/{evt.value}.jpg")
|
176 |
-
return heatmap_image
|
177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
|
179 |
-
def load_vision_heatmap(evt: gr.SelectData):
|
180 |
-
heatmap_image = gr.Image(f"results-vision/{evt.value}.jpg")
|
181 |
-
return heatmap_image
|
182 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
|
184 |
-
|
185 |
-
heatmap_image = gr.Image(f"results-cot/{evt.value}.jpg")
|
186 |
-
return heatmap_image
|
187 |
|
188 |
|
189 |
-
def
|
190 |
-
heatmap_image = gr.Image(f"results
|
191 |
return heatmap_image
|
192 |
|
193 |
|
194 |
-
def calculate_order_by_first_substring(selected_models):
|
195 |
-
global text_only_filtered_raw
|
196 |
-
first_columns = all_data[all_data["substring_index"] == 1]
|
197 |
-
query_ids_df = first_columns[first_columns["Model Type"] == "Text Only"]
|
198 |
-
query_ids_df = query_ids_df[query_ids_df["Model Name"].isin(selected_models)]
|
199 |
-
|
200 |
-
query_ids_df = query_ids_df.groupby("query_id").filter(
|
201 |
-
lambda x: x["parsed_judge_response"].eq(1).all()
|
202 |
-
)
|
203 |
-
|
204 |
-
fsm_ids = query_ids_df.fsm_id.unique()
|
205 |
-
|
206 |
-
text_only = all_data[all_data["Model Type"] == "Text Only"]
|
207 |
-
text_only_filtered = text_only[text_only["fsm_id"].isin(fsm_ids)]
|
208 |
-
text_only_filtered_raw = text_only_filtered.copy()
|
209 |
-
|
210 |
-
query_ids = text_only_filtered.query_id.unique()
|
211 |
-
text_only_filtered = (
|
212 |
-
text_only_filtered.groupby(["Model Name"])["parsed_judge_response"]
|
213 |
-
.mean()
|
214 |
-
.reset_index()
|
215 |
-
)
|
216 |
-
|
217 |
-
text_only_filtered["Accuracy"] = text_only_filtered["parsed_judge_response"] * 100
|
218 |
-
text_only_filtered.drop("parsed_judge_response", axis=1, inplace=True)
|
219 |
-
|
220 |
-
text_only_filtered["Accuracy"] = text_only_filtered["Accuracy"].apply(
|
221 |
-
lambda x: round(x, 2)
|
222 |
-
)
|
223 |
-
text_only_filtered.sort_values("Accuracy", ascending=False, inplace=True)
|
224 |
-
|
225 |
-
number_of_queries = len(query_ids)
|
226 |
-
number_of_fsms = len(fsm_ids)
|
227 |
-
|
228 |
-
return text_only_filtered, number_of_queries, number_of_fsms
|
229 |
-
|
230 |
-
|
231 |
-
def calculate_order_by_first_substring_cot(selected_models):
|
232 |
-
global text_only_filtered_raw_cot
|
233 |
-
first_columns = all_data[all_data["substring_index"] == 1]
|
234 |
-
query_ids_df = first_columns[first_columns["Model Type"] == "CoT Text Only"]
|
235 |
-
query_ids_df = query_ids_df[query_ids_df["Model Name"].isin(selected_models)]
|
236 |
-
|
237 |
-
query_ids_df = query_ids_df.groupby("query_id").filter(
|
238 |
-
lambda x: x["parsed_judge_response"].eq(1).all()
|
239 |
-
)
|
240 |
-
|
241 |
-
fsm_ids = query_ids_df.fsm_id.unique()
|
242 |
-
|
243 |
-
text_only = all_data[all_data["Model Type"] == "CoT Text Only"]
|
244 |
-
text_only_filtered = text_only[text_only["fsm_id"].isin(fsm_ids)]
|
245 |
-
text_only_filtered_raw_cot = text_only_filtered.copy()
|
246 |
-
|
247 |
-
query_ids = text_only_filtered.query_id.unique()
|
248 |
-
text_only_filtered = (
|
249 |
-
text_only_filtered.groupby(["Model Name"])["parsed_judge_response"]
|
250 |
-
.mean()
|
251 |
-
.reset_index()
|
252 |
-
)
|
253 |
-
|
254 |
-
text_only_filtered["Accuracy"] = text_only_filtered["parsed_judge_response"] * 100
|
255 |
-
text_only_filtered.drop("parsed_judge_response", axis=1, inplace=True)
|
256 |
-
|
257 |
-
text_only_filtered["Accuracy"] = text_only_filtered["Accuracy"].apply(
|
258 |
-
lambda x: round(x, 2)
|
259 |
-
)
|
260 |
-
text_only_filtered.sort_values("Accuracy", ascending=False, inplace=True)
|
261 |
-
|
262 |
-
number_of_queries = len(query_ids)
|
263 |
-
number_of_fsms = len(fsm_ids)
|
264 |
-
|
265 |
-
return text_only_filtered, number_of_queries, number_of_fsms
|
266 |
-
|
267 |
-
|
268 |
-
def generate_heatmap_for_specific_model(model_name):
|
269 |
-
global text_only_filtered_raw
|
270 |
-
|
271 |
-
cmap = ListedColormap(["lightblue", "red", "green"])
|
272 |
-
bounds = [-1.5, -0.5, 0.5, 1.5]
|
273 |
-
norm = BoundaryNorm(bounds, cmap.N)
|
274 |
-
|
275 |
-
model_df = text_only_filtered_raw[
|
276 |
-
text_only_filtered_raw["Model Name"] == model_name
|
277 |
-
]
|
278 |
-
model_df["fsm_info"] = model_df.apply(
|
279 |
-
lambda x: f"{x['num_states']} states, {x['num_alphabet']} alphabet", axis=1
|
280 |
-
)
|
281 |
-
model_df = model_df.sort_values(by=["num_states", "num_alphabet"])
|
282 |
-
|
283 |
-
pivot_df = (
|
284 |
-
model_df.pivot_table(
|
285 |
-
index="fsm_info",
|
286 |
-
columns="substring_index",
|
287 |
-
values="parsed_judge_response",
|
288 |
-
aggfunc="first",
|
289 |
-
)
|
290 |
-
.fillna(-1)
|
291 |
-
.astype(float)
|
292 |
-
)
|
293 |
-
|
294 |
-
# Dynamically adjust figure size
|
295 |
-
num_rows, num_cols = pivot_df.shape
|
296 |
-
fig_width = max(12, num_cols * 0.5) # Adjust width per column
|
297 |
-
fig_height = max(8, num_rows * 0.4) # Adjust height per row
|
298 |
-
|
299 |
-
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
|
300 |
-
sns.heatmap(
|
301 |
-
pivot_df,
|
302 |
-
cmap=cmap,
|
303 |
-
linewidths=1,
|
304 |
-
linecolor="black",
|
305 |
-
norm=norm,
|
306 |
-
cbar=False,
|
307 |
-
square=True,
|
308 |
-
ax=ax,
|
309 |
-
)
|
310 |
-
plt.title(f"Heatmap for Model: {model_name}", fontsize=12)
|
311 |
-
plt.xlabel("Substring Index")
|
312 |
-
plt.ylabel("FSM (States, Alphabet)")
|
313 |
-
plt.xticks(rotation=45)
|
314 |
-
|
315 |
-
sns.despine(ax=ax, top=True, right=True, left=True, bottom=True)
|
316 |
-
|
317 |
-
return fig
|
318 |
-
|
319 |
-
|
320 |
-
def generate_heatmap_for_specific_model_cot(model_name):
|
321 |
-
global text_only_filtered_raw_cot
|
322 |
-
|
323 |
-
cmap = ListedColormap(["lightblue", "red", "green"])
|
324 |
-
bounds = [-1.5, -0.5, 0.5, 1.5]
|
325 |
-
norm = BoundaryNorm(bounds, cmap.N)
|
326 |
-
|
327 |
-
model_df = text_only_filtered_raw_cot[
|
328 |
-
text_only_filtered_raw_cot["Model Name"] == model_name
|
329 |
-
]
|
330 |
-
model_df["fsm_info"] = model_df.apply(
|
331 |
-
lambda x: f"{x['num_states']} states, {x['num_alphabet']} alphabet", axis=1
|
332 |
-
)
|
333 |
-
model_df = model_df.sort_values(by=["num_states", "num_alphabet"])
|
334 |
-
|
335 |
-
pivot_df = (
|
336 |
-
model_df.pivot_table(
|
337 |
-
index="fsm_info",
|
338 |
-
columns="substring_index",
|
339 |
-
values="parsed_judge_response",
|
340 |
-
aggfunc="first",
|
341 |
-
)
|
342 |
-
.fillna(-1)
|
343 |
-
.astype(float)
|
344 |
-
)
|
345 |
-
|
346 |
-
# Dynamically adjust figure size
|
347 |
-
num_rows, num_cols = pivot_df.shape
|
348 |
-
fig_width = max(12, num_cols * 0.5) # Adjust width per column
|
349 |
-
fig_height = max(8, num_rows * 0.4) # Adjust height per row
|
350 |
-
|
351 |
-
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
|
352 |
-
sns.heatmap(
|
353 |
-
pivot_df,
|
354 |
-
cmap=cmap,
|
355 |
-
linewidths=1,
|
356 |
-
linecolor="black",
|
357 |
-
norm=norm,
|
358 |
-
cbar=False,
|
359 |
-
square=True,
|
360 |
-
ax=ax,
|
361 |
-
)
|
362 |
-
plt.title(f"Heatmap for Model: {model_name}", fontsize=12)
|
363 |
-
plt.xlabel("Substring Index")
|
364 |
-
plt.ylabel("FSM (States, Alphabet)")
|
365 |
-
plt.xticks(rotation=45)
|
366 |
-
|
367 |
-
sns.despine(ax=ax, top=True, right=True, left=True, bottom=True)
|
368 |
-
|
369 |
-
return fig
|
370 |
-
|
371 |
-
|
372 |
-
def generate_heatmap_for_intersection_model(model_name):
|
373 |
-
global intersection_df
|
374 |
-
|
375 |
-
cmap = ListedColormap(["lightblue", "red", "green"])
|
376 |
-
bounds = [-1.5, -0.5, 0.5, 1.5]
|
377 |
-
norm = BoundaryNorm(bounds, cmap.N)
|
378 |
-
|
379 |
-
# Filter for a specific model
|
380 |
-
model_df = intersection_df[intersection_df["model_name"] == model_name].copy()
|
381 |
-
|
382 |
-
if model_df.empty:
|
383 |
-
print(f"No data found for model {model_name}. Skipping heatmap generation.")
|
384 |
-
return None
|
385 |
-
|
386 |
-
model_df["fsm_info"] = model_df.apply(
|
387 |
-
lambda x: f"{x['num_states']} states, {x['num_alphabet']} alphabet", axis=1
|
388 |
-
)
|
389 |
-
model_df = model_df.sort_values(by=["num_states", "num_alphabet"])
|
390 |
-
|
391 |
-
pivot_df = (
|
392 |
-
model_df.pivot_table(
|
393 |
-
index="fsm_info",
|
394 |
-
columns="substring_index",
|
395 |
-
values="parsed_judge_response",
|
396 |
-
aggfunc="first",
|
397 |
-
)
|
398 |
-
.fillna(-1)
|
399 |
-
.astype(float)
|
400 |
-
)
|
401 |
-
|
402 |
-
# Dynamically adjust figure size
|
403 |
-
num_rows, num_cols = pivot_df.shape
|
404 |
-
fig_width = max(12, num_cols * 0.5)
|
405 |
-
fig_height = max(8, num_rows * 0.4)
|
406 |
-
|
407 |
-
fig, ax = plt.subplots(figsize=(fig_width, fig_height))
|
408 |
-
sns.heatmap(
|
409 |
-
pivot_df,
|
410 |
-
cmap=cmap,
|
411 |
-
linewidths=1,
|
412 |
-
linecolor="black",
|
413 |
-
norm=norm,
|
414 |
-
cbar=False,
|
415 |
-
square=True,
|
416 |
-
ax=ax,
|
417 |
-
)
|
418 |
-
plt.title(f"Heatmap for Model: {model_name}", fontsize=12)
|
419 |
-
plt.xlabel("Substring Index")
|
420 |
-
plt.ylabel("FSM (States, Alphabet)")
|
421 |
-
plt.xticks(rotation=45)
|
422 |
-
|
423 |
-
sns.despine(ax=ax, top=True, right=True, left=True, bottom=True)
|
424 |
-
|
425 |
-
plt.close(fig)
|
426 |
-
return fig
|
427 |
-
|
428 |
-
|
429 |
-
def show_constraint_heatmap(evt: gr.SelectData):
|
430 |
-
model_name = evt.value
|
431 |
-
return generate_heatmap_for_specific_model(model_name)
|
432 |
-
|
433 |
-
|
434 |
-
def show_constraint_heatmap_cot(evt: gr.SelectData):
|
435 |
-
model_name = evt.value
|
436 |
-
return generate_heatmap_for_specific_model_cot(model_name)
|
437 |
-
|
438 |
-
|
439 |
-
def show_intersection_heatmap(evt: gr.SelectData):
|
440 |
-
model_name = evt.value
|
441 |
-
return generate_heatmap_for_intersection_model(model_name)
|
442 |
-
|
443 |
-
|
444 |
with gr.Blocks() as demo:
|
445 |
gr.Markdown("# FSM Benchmark Leaderboard")
|
446 |
with gr.Tab("Text-only Benchmark"):
|
447 |
-
gr.
|
448 |
-
leader_board = gr.Dataframe(accuracy_df_qwen, headers=headers_with_icons)
|
449 |
gr.Markdown("## Heatmap")
|
450 |
heatmap_image_qwen = gr.Image(label="", show_label=False)
|
451 |
-
leader_board.select(fn=
|
452 |
|
453 |
-
with gr.Tab("Vision Benchmark", visible=False):
|
454 |
-
|
455 |
-
|
456 |
-
|
457 |
-
)
|
458 |
-
gr.Markdown("## Heatmap")
|
459 |
-
heatmap_image_vision = gr.Image(label="", show_label=False)
|
460 |
-
leader_board_vision.select(
|
461 |
-
fn=load_vision_heatmap, outputs=[heatmap_image_vision]
|
462 |
-
)
|
463 |
-
|
464 |
-
with gr.Tab("Text-only Benchmark (CoT)", visible=False):
|
465 |
-
gr.Markdown("# Text-only Leaderboard (CoT)")
|
466 |
-
cot_leader_board_text = gr.Dataframe(
|
467 |
-
cot_text_accuracy_df, headers=headers_with_icons
|
468 |
-
)
|
469 |
-
gr.Markdown("## Heatmap")
|
470 |
-
cot_heatmap_image_text = gr.Image(label="", show_label=False)
|
471 |
-
cot_leader_board_text.select(
|
472 |
-
fn=load_cot_heatmap, outputs=[cot_heatmap_image_text]
|
473 |
-
)
|
474 |
-
|
475 |
-
# with gr.Tab("Vision Benchmark (CoT)"):
|
476 |
-
# gr.Markdown("# Vision Benchmark Leaderboard (CoT)")
|
477 |
-
# cot_leader_board_vision = gr.Dataframe(
|
478 |
-
# cot_vision_accuracy_df, headers=headers_with_icons
|
479 |
# )
|
480 |
# gr.Markdown("## Heatmap")
|
481 |
-
#
|
482 |
-
#
|
483 |
-
# fn=
|
484 |
# )
|
485 |
|
486 |
-
with gr.Tab("
|
487 |
-
|
488 |
-
|
489 |
-
|
490 |
-
|
491 |
-
|
492 |
-
|
493 |
-
|
494 |
-
|
495 |
-
|
496 |
-
|
497 |
-
|
498 |
-
|
499 |
-
|
500 |
-
|
501 |
-
|
502 |
-
|
503 |
-
|
504 |
-
|
505 |
-
|
506 |
-
|
507 |
-
)
|
508 |
-
|
509 |
-
|
510 |
-
|
511 |
-
|
512 |
-
|
513 |
-
|
514 |
-
|
515 |
-
|
516 |
-
|
517 |
-
|
518 |
-
|
519 |
-
|
520 |
-
|
521 |
-
|
522 |
-
|
523 |
-
|
524 |
-
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
|
530 |
-
|
531 |
-
|
532 |
-
|
533 |
-
|
534 |
-
|
535 |
-
|
536 |
-
|
537 |
-
|
538 |
-
|
539 |
-
|
540 |
-
|
541 |
-
|
542 |
-
|
543 |
-
|
544 |
-
|
545 |
-
|
546 |
-
|
547 |
-
|
548 |
-
|
549 |
-
)
|
550 |
-
|
551 |
-
constrained_leader_board_text.select(
|
552 |
-
fn=show_constraint_heatmap, outputs=[constrained_leader_board_plot]
|
553 |
-
)
|
554 |
-
|
555 |
-
constrained_leader_board_text_cot.select(
|
556 |
-
fn=show_constraint_heatmap_cot, outputs=[constrained_leader_board_plot_cot]
|
557 |
-
)
|
558 |
-
|
559 |
-
intersection_leader_board.select(
|
560 |
-
fn=show_intersection_heatmap, outputs=[heatmap_image]
|
561 |
-
)
|
562 |
|
563 |
demo.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
+
from glob import glob
|
3 |
|
4 |
+
import gradio as gr
|
5 |
import matplotlib.pyplot as plt
|
|
|
|
|
6 |
import pandas as pd
|
7 |
+
import seaborn as sns
|
8 |
+
from matplotlib.colors import BoundaryNorm, ListedColormap
|
9 |
+
|
10 |
+
all_results = pd.read_pickle("all_results.pkl")
|
11 |
+
|
12 |
+
|
13 |
+
def get_accuracy_dataframe(df):
|
14 |
+
# Calculate overall model accuracy
|
15 |
+
df['parsed_judge_response'] = df['parsed_judge_response'].astype(float)
|
16 |
+
model_accuracy = df.groupby('model_name')['parsed_judge_response'].mean().reset_index()
|
17 |
+
|
18 |
+
# Calculate model accuracy per difficulty level
|
19 |
+
df['difficulty_level'] = df['difficulty_level'].astype(int)
|
20 |
+
model_accuracy_per_level = df.groupby(['model_name', 'difficulty_level'])['parsed_judge_response'].mean().reset_index()
|
21 |
+
model_accuracy_per_level_df = model_accuracy_per_level.pivot(index='model_name', columns='difficulty_level', values='parsed_judge_response')
|
22 |
+
|
23 |
+
# Merge overall accuracy and level-based accuracy into a single DataFrame
|
24 |
+
model_accuracy_df = model_accuracy.merge(model_accuracy_per_level_df, on='model_name')
|
25 |
+
model_accuracy_df.rename(columns={1: 'level_1', 2: 'level_2', 3: 'level_3', 4: 'level_4', 5: 'level_5'}, inplace=True)
|
26 |
+
model_accuracy_df.rename(columns={'parsed_judge_response': 'Accuracy'}, inplace=True)
|
27 |
+
|
28 |
+
# Multiply by 100 and format to one decimal point
|
29 |
+
model_accuracy_df = model_accuracy_df.applymap(lambda x: round(x * 100, 1) if isinstance(x, float) else x)
|
30 |
+
|
31 |
+
# Add headers with icons
|
32 |
+
model_accuracy_df.columns = [
|
33 |
+
"🤖 Model Name",
|
34 |
+
"⭐ Overall",
|
35 |
+
"📈 Level 1",
|
36 |
+
"🔍 Level 2",
|
37 |
+
"📘 Level 3",
|
38 |
+
"🔬 Level 4",
|
39 |
+
]
|
40 |
|
41 |
+
model_accuracy_df.sort_values(by="⭐ Overall", ascending=False, inplace=True)
|
42 |
+
|
43 |
+
return model_accuracy_df
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
|
46 |
+
accuracy_df = get_accuracy_dataframe(all_results)
|
|
|
|
|
47 |
|
48 |
|
49 |
# Define the column names with icons
|
|
|
65 |
"Level 4 Accuracy",
|
66 |
]
|
67 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
def load_heatmap(evt: gr.SelectData):
|
69 |
heatmap_image = gr.Image(f"results/{evt.value}.jpg")
|
70 |
return heatmap_image
|
71 |
|
72 |
|
|
|
|
|
|
|
73 |
|
74 |
+
# # Function to process data
|
75 |
+
# def process_data(data):
|
76 |
+
# data_for_df = []
|
77 |
+
# for file, df in data.items():
|
78 |
+
# overall_accuracy = round(calculate_accuracy(df), 2)
|
79 |
+
# breakdown_accuracy = [round(acc, 2) for acc in accuracy_breakdown(df)]
|
80 |
+
# model_name = file.split("/")[-1].replace(".pkl", "")
|
81 |
+
# data_for_df.append([model_name, overall_accuracy] + breakdown_accuracy)
|
82 |
+
# return data_for_df
|
83 |
|
|
|
|
|
|
|
84 |
|
85 |
+
# # Function to finalize DataFrame
|
86 |
+
# def finalize_df(df):
|
87 |
+
# df = df.round(1) # Round to one decimal place
|
88 |
+
# df = df.applymap(lambda x: f"{x:.1f}" if isinstance(x, (int, float)) else x)
|
89 |
+
# df.columns = headers_with_icons
|
90 |
+
# df.sort_values(by="⭐ Overall", ascending=False, inplace=True)
|
91 |
+
# # add a new column with the order (index)
|
92 |
+
# df["#"] = range(1, len(df) + 1)
|
93 |
+
# # bring rank to the first column
|
94 |
+
# cols = df.columns.tolist()
|
95 |
+
# cols = cols[-1:] + cols[:-1]
|
96 |
+
# df = df[cols]
|
97 |
|
98 |
+
# return df
|
|
|
|
|
99 |
|
100 |
|
101 |
+
def load_heatmap(evt: gr.SelectData):
|
102 |
+
heatmap_image = gr.Image(f"results/{evt.value}.jpg")
|
103 |
return heatmap_image
|
104 |
|
105 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
106 |
with gr.Blocks() as demo:
|
107 |
gr.Markdown("# FSM Benchmark Leaderboard")
|
108 |
with gr.Tab("Text-only Benchmark"):
|
109 |
+
leader_board = gr.Dataframe(accuracy_df, headers=headers_with_icons)
|
|
|
110 |
gr.Markdown("## Heatmap")
|
111 |
heatmap_image_qwen = gr.Image(label="", show_label=False)
|
112 |
+
leader_board.select(fn=load_heatmap, outputs=[heatmap_image_qwen])
|
113 |
|
114 |
+
# with gr.Tab("Vision Benchmark", visible=False):
|
115 |
+
# gr.Markdown("# Vision Benchmark Leaderboard")
|
116 |
+
# leader_board_vision = gr.Dataframe(
|
117 |
+
# vision_accuracy_df, headers=headers_with_icons
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
118 |
# )
|
119 |
# gr.Markdown("## Heatmap")
|
120 |
+
# heatmap_image_vision = gr.Image(label="", show_label=False)
|
121 |
+
# leader_board_vision.select(
|
122 |
+
# fn=load_vision_heatmap, outputs=[heatmap_image_vision]
|
123 |
# )
|
124 |
|
125 |
+
# with gr.Tab("Text-only Benchmark (CoT)", visible=False):
|
126 |
+
# gr.Markdown("# Text-only Leaderboard (CoT)")
|
127 |
+
# cot_leader_board_text = gr.Dataframe(
|
128 |
+
# cot_text_accuracy_df, headers=headers_with_icons
|
129 |
+
# )
|
130 |
+
# gr.Markdown("## Heatmap")
|
131 |
+
# cot_heatmap_image_text = gr.Image(label="", show_label=False)
|
132 |
+
# cot_leader_board_text.select(
|
133 |
+
# fn=load_cot_heatmap, outputs=[cot_heatmap_image_text]
|
134 |
+
# )
|
135 |
+
|
136 |
+
# with gr.Tab("Constraint Text-only Results (CoT)", visible=False):
|
137 |
+
# gr.Markdown("## Constraint Text-only Leaderboard by first substrin (CoT)")
|
138 |
+
# included_models_cot = gr.CheckboxGroup(
|
139 |
+
# label="Models to include",
|
140 |
+
# choices=all_cot_text_only_models,
|
141 |
+
# value=all_cot_text_only_models,
|
142 |
+
# interactive=True,
|
143 |
+
# )
|
144 |
+
# with gr.Row():
|
145 |
+
# number_of_queries_cot = gr.Textbox(label="Number of included queries")
|
146 |
+
# number_of_fsms_cot = gr.Textbox(label="Number of included FSMs")
|
147 |
+
|
148 |
+
# constrained_leader_board_text_cot = gr.Dataframe()
|
149 |
+
# constrained_leader_board_plot_cot = gr.Plot()
|
150 |
+
|
151 |
+
# with gr.Tab("Majority Vote (Subset 1)", visible=False):
|
152 |
+
# gr.Markdown("## Majority Vote (Subset 1)")
|
153 |
+
# intersection_leader_board = gr.Dataframe(
|
154 |
+
# intersection_df_acc, headers=headers_with_icons
|
155 |
+
# )
|
156 |
+
# heatmap_image = gr.Plot(label="Model Heatmap")
|
157 |
+
|
158 |
+
# with gr.Tab("Text-only Benchmark (deprecated)", visible=False):
|
159 |
+
# gr.Markdown("# Text-only Leaderboard")
|
160 |
+
# leader_board = gr.Dataframe(accuracy_df, headers=headers_with_icons)
|
161 |
+
# gr.Markdown("## Heatmap")
|
162 |
+
# heatmap_image = gr.Image(label="", show_label=False)
|
163 |
+
# leader_board.select(fn=load_heatmap, outputs=[heatmap_image])
|
164 |
+
|
165 |
+
# # ============ Callbacks ============
|
166 |
+
|
167 |
+
# included_models_cot.select(
|
168 |
+
# fn=calculate_order_by_first_substring_cot,
|
169 |
+
# inputs=[included_models_cot],
|
170 |
+
# outputs=[
|
171 |
+
# constrained_leader_board_text_cot,
|
172 |
+
# number_of_queries_cot,
|
173 |
+
# number_of_fsms_cot,
|
174 |
+
# ],
|
175 |
+
# queue=True,
|
176 |
+
# )
|
177 |
+
|
178 |
+
# constrained_leader_board_text.select(
|
179 |
+
# fn=show_constraint_heatmap, outputs=[constrained_leader_board_plot]
|
180 |
+
# )
|
181 |
+
|
182 |
+
# constrained_leader_board_text_cot.select(
|
183 |
+
# fn=show_constraint_heatmap_cot, outputs=[constrained_leader_board_plot_cot]
|
184 |
+
# )
|
185 |
+
|
186 |
+
# intersection_leader_board.select(
|
187 |
+
# fn=show_intersection_heatmap, outputs=[heatmap_image]
|
188 |
+
# )
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
189 |
|
190 |
demo.launch()
|
results-cot/CodeLlama-70b-Instruct-hf.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:a8d72952877e3e4023251396d037ebae8145e3e82e2d4a328ce132171ea70267
|
3 |
-
size 20756425
|
|
|
|
|
|
|
|
results-cot/Mixtral-8x7B-Instruct-v0.1.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:093e919d90609c3be8d6818cf56ca018214da3a42b78aeaf85f92581b72c5ad4
|
3 |
-
size 19494123
|
|
|
|
|
|
|
|
results-cot/Mixtral-8x7B-Instruct-v0.1.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:686692584c6ba027c454d699bbf585b95e5c99bfc426810ea74b327a975b9cf3
|
3 |
-
size 19489822
|
|
|
|
|
|
|
|
results-cot/Mixtral-8x7B-Instruct-v0.1.png
DELETED
Git LFS Details
|
results-cot/Qwen1.5-72B-Chat.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:32681449776facf1084405001e69ed7926b79c69f9717fb159e3eb064b333636
|
3 |
-
size 15795431
|
|
|
|
|
|
|
|
results-cot/Qwen1.5-72B-Chat.jpg
DELETED
Git LFS Details
|
results-cot/Qwen1.5-72B-Chat.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:1c20383298d4b6482ca7c30bf91822e24099dc67b71a3be10271005e25208c40
|
3 |
-
size 15778970
|
|
|
|
|
|
|
|
results-cot/Qwen1.5-72B-Chat.png
DELETED
Git LFS Details
|
results-cot/claude-3-sonnet-20240229.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:77c47e8698b8aa86de42bda82cb27c1efc0174c18b3ba306ff263cbd260de20e
|
3 |
-
size 13144187
|
|
|
|
|
|
|
|
results-cot/claude-3-sonnet-20240229.jpg
DELETED
Git LFS Details
|
results-cot/claude-3-sonnet-20240229.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:7e5a528d8fa3491adbc4e2e5a982e65a8a9c2fb5a96d321c864935dbe506453e
|
3 |
-
size 13047413
|
|
|
|
|
|
|
|
results-cot/claude-3-sonnet-20240229.png
DELETED
Git LFS Details
|
results-cot/dbrx-instruct.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:698626edcabf06c89b9b8fac2f929927e9b7351306f525cb87fc6e06ce1bc3e3
|
3 |
-
size 19267224
|
|
|
|
|
|
|
|
results-cot/deepseek-llm-67b-chat.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:191a06465559524615f8cca0c46ca2af417a289e9fbab2109e2d2a3c92432fe2
|
3 |
-
size 16692090
|
|
|
|
|
|
|
|
results-cot/deepseek-llm-67b-chat.jpg
DELETED
Git LFS Details
|
results-cot/deepseek-llm-67b-chat.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:005500788ecc004bb7a86054f8921e6331addd52c62d8e611a9d82b649ed4925
|
3 |
-
size 16712146
|
|
|
|
|
|
|
|
results-cot/deepseek-llm-67b-chat.png
DELETED
Git LFS Details
|
results-cot/gemini-pro.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:cd52b8cfe861dd9fc106dafbe11569f9f2bf5848482e9b42c0ad3e88ffc83035
|
3 |
-
size 14773471
|
|
|
|
|
|
|
|
results-cot/gemini-pro.jpg
DELETED
Git LFS Details
|
results-cot/gemini-pro.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:72caa2e3c4ef872b3d6065f55179aa85d3eaa7978f9abb6fc1e3f58db19d8a69
|
3 |
-
size 14759970
|
|
|
|
|
|
|
|
results-cot/gemini-pro.png
DELETED
Git LFS Details
|
results-cot/gemma-7b-it.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:f8535fa3f2ef5a94b1b552859930e0476ca0f3c77ec4c277893a9ab9ef45d6c3
|
3 |
-
size 16793758
|
|
|
|
|
|
|
|
results-cot/gemma-7b-it.jpg
DELETED
Git LFS Details
|
results-cot/gemma-7b-it.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:3c581027f8b78df5934117276cec3e53613f5ac953d045f71af4121b3ec2e1a4
|
3 |
-
size 16822239
|
|
|
|
|
|
|
|
results-cot/gemma-7b-it.png
DELETED
Git LFS Details
|
results-cot/gpt-3.5-turbo-0125.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:e9ea8fd5fd1e335397c14e3b312f60111614de4c20e50e5f3bc9d4fb6f049d70
|
3 |
-
size 14471960
|
|
|
|
|
|
|
|
results-cot/gpt-3.5-turbo-0125.jpg
DELETED
Git LFS Details
|
results-cot/gpt-3.5-turbo-0125.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:702c5dd6ffb22938a59815c65759ca5a201369b59633ee6c80c7030151e53634
|
3 |
-
size 14487477
|
|
|
|
|
|
|
|
results-cot/gpt-3.5-turbo-0125.png
DELETED
Git LFS Details
|
results-cot/gpt-4-turbo-2024-04-09.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:011eeea335d67be1a859d51ca2accfd6bc56a2222394db52e68c9f444fec1c9b
|
3 |
-
size 20522713
|
|
|
|
|
|
|
|
results-cot/gpt-4-turbo-2024-04-09.jpg
DELETED
Git LFS Details
|
results-cot/gpt-4-turbo-2024-04-09.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:e1383c823ccc8e09830dd4b96742c19fdf7515b56a715edd6ce131ff07fa029c
|
3 |
-
size 20513069
|
|
|
|
|
|
|
|
results-cot/gpt-4-turbo-2024-04-09.png
DELETED
Git LFS Details
|
results-vision/claude-3-opus-20240229.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:74ed9f652dc23c7408b307cb477e91692b561739bf23b33ba93319d9d073f294
|
3 |
-
size 10855333
|
|
|
|
|
|
|
|
results-vision/claude-3-opus-20240229.jpg
DELETED
Git LFS Details
|
results-vision/claude-3-opus-20240229.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:db495f6b9f264c3d9978c7d479b8206fbf8e67843c302c42416e0c6ee1aaa0a4
|
3 |
-
size 10789555
|
|
|
|
|
|
|
|
results-vision/claude-3-opus-20240229.png
DELETED
Git LFS Details
|
results-vision/claude-3-opus-vision.jpg
DELETED
Git LFS Details
|
results-vision/claude-3-opus-vision.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:db495f6b9f264c3d9978c7d479b8206fbf8e67843c302c42416e0c6ee1aaa0a4
|
3 |
-
size 10789555
|
|
|
|
|
|
|
|
results-vision/claude-3-opus-vision.png
DELETED
Git LFS Details
|
results-vision/gemini-pro-vision.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:a33999636396dc5a3dfc858de2b0a421a6fa48d78d51ad74838f2d6afc261999
|
3 |
-
size 6158469
|
|
|
|
|
|
|
|
results-vision/gemini-pro-vision.jpg
DELETED
Git LFS Details
|
results-vision/gemini-pro-vision.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:6dec98de51c833c57d8a322c9084391b0aeeffd67a6ee0edbed2f23061aeb1e0
|
3 |
-
size 6118424
|
|
|
|
|
|
|
|
results-vision/gemini-pro-vision.png
DELETED
Git LFS Details
|
results-vision/gpt-4v.jpg
DELETED
Git LFS Details
|
results-vision/gpt-4v.pkl
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:e98609b7b262836c49cdaf5d3dfc02b7037dc9fcc1f75a41d58a984015318759
|
3 |
-
size 6363780
|
|
|
|
|
|
|
|
results-vision/gpt-4v.png
DELETED
Git LFS Details
|
results/CodeLlama-70b-Instruct-hf.csv
DELETED
@@ -1,3 +0,0 @@
|
|
1 |
-
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:3726905a1656174f3c29edfced6f2eec63222f6be8965c0d970264901d8cfc75
|
3 |
-
size 16476347
|
|
|
|
|
|
|
|