Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Commit
·
80fb33a
1
Parent(s):
d3db3e5
change data structure
Browse files
.gitignore
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
.ruff_cache
|
{src/gen/data → data}/arena-hard-v0.1/question.jsonl
RENAMED
|
File without changes
|
{src/gen/data → data}/arena_hard_battles.jsonl
RENAMED
|
File without changes
|
{src/gen/data → data}/bootstrapping_results.jsonl
RENAMED
|
File without changes
|
src/gen/show_result.py
CHANGED
|
@@ -1,17 +1,15 @@
|
|
| 1 |
-
import pandas as pd
|
| 2 |
-
import numpy as np
|
| 3 |
-
import plotly.express as px
|
| 4 |
-
|
| 5 |
-
import datetime
|
| 6 |
import argparse
|
| 7 |
-
import
|
| 8 |
import math
|
| 9 |
-
|
|
|
|
| 10 |
from glob import glob
|
| 11 |
-
from tqdm import tqdm
|
| 12 |
|
|
|
|
|
|
|
|
|
|
| 13 |
from sklearn.linear_model import LogisticRegression
|
| 14 |
-
from
|
| 15 |
from utils import load_model_answers
|
| 16 |
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import argparse
|
| 2 |
+
import datetime
|
| 3 |
import math
|
| 4 |
+
import os
|
| 5 |
+
from collections import defaultdict
|
| 6 |
from glob import glob
|
|
|
|
| 7 |
|
| 8 |
+
import numpy as np
|
| 9 |
+
import pandas as pd
|
| 10 |
+
import plotly.express as px
|
| 11 |
from sklearn.linear_model import LogisticRegression
|
| 12 |
+
from tqdm import tqdm
|
| 13 |
from utils import load_model_answers
|
| 14 |
|
| 15 |
|
src/leaderboard/build_leaderboard.py
CHANGED
|
@@ -58,24 +58,26 @@ def build_leadearboard_df():
|
|
| 58 |
download_dataset("Vikhrmodels/openbench-eval", EVAL_RESULTS_PATH)
|
| 59 |
# print(subprocess.Popen('ls src'))
|
| 60 |
|
| 61 |
-
# copy the
|
| 62 |
subprocess.run(
|
| 63 |
[
|
| 64 |
"rsync",
|
| 65 |
"-azP",
|
| 66 |
"--ignore-existing",
|
| 67 |
-
f"{EVAL_RESULTS_PATH}/internal
|
| 68 |
-
"
|
| 69 |
],
|
| 70 |
check=False,
|
| 71 |
)
|
|
|
|
|
|
|
| 72 |
subprocess.run(
|
| 73 |
[
|
| 74 |
"rsync",
|
| 75 |
"-azP",
|
| 76 |
"--ignore-existing",
|
| 77 |
f"{EVAL_RESULTS_PATH}/model_judgment/*",
|
| 78 |
-
"
|
| 79 |
],
|
| 80 |
check=False,
|
| 81 |
)
|
|
|
|
| 58 |
download_dataset("Vikhrmodels/openbench-eval", EVAL_RESULTS_PATH)
|
| 59 |
# print(subprocess.Popen('ls src'))
|
| 60 |
|
| 61 |
+
# copy the trusted model answers to data
|
| 62 |
subprocess.run(
|
| 63 |
[
|
| 64 |
"rsync",
|
| 65 |
"-azP",
|
| 66 |
"--ignore-existing",
|
| 67 |
+
f"{EVAL_RESULTS_PATH}/internal/*",
|
| 68 |
+
"data/arena-hard-v0.1/model_answer/internal/",
|
| 69 |
],
|
| 70 |
check=False,
|
| 71 |
)
|
| 72 |
+
# copy the judgement pre generated
|
| 73 |
+
# Will be rewritten after we switch to new gen for each submit
|
| 74 |
subprocess.run(
|
| 75 |
[
|
| 76 |
"rsync",
|
| 77 |
"-azP",
|
| 78 |
"--ignore-existing",
|
| 79 |
f"{EVAL_RESULTS_PATH}/model_judgment/*",
|
| 80 |
+
"data/arena-hard-v0.1/model_judgement/",
|
| 81 |
],
|
| 82 |
check=False,
|
| 83 |
)
|