Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Commit
•
311dc3a
1
Parent(s):
c237b5a
init
Browse files- .gitignore +1 -0
- app.py +114 -0
- dev.ipynb +1340 -0
- release_date_mapping.json +457 -0
- requirements.txt +4 -0
.gitignore
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
.venv/
|
app.py
ADDED
@@ -0,0 +1,114 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pickle
|
2 |
+
|
3 |
+
import pandas as pd
|
4 |
+
import gradio as gr
|
5 |
+
from huggingface_hub import HfFileSystem, hf_hub_download
|
6 |
+
|
7 |
+
if gr.NO_RELOAD:
|
8 |
+
###################
|
9 |
+
### Load Data
|
10 |
+
###################
|
11 |
+
|
12 |
+
key_to_category_name = {
|
13 |
+
"full": "Overall",
|
14 |
+
"coding": "Coding",
|
15 |
+
"long_user": "Longer Query",
|
16 |
+
"english": "English",
|
17 |
+
"chinese": "Chinese",
|
18 |
+
"french": "French",
|
19 |
+
"no_tie": "Exclude Ties",
|
20 |
+
"no_short": "Exclude Short Query (< 5 tokens)",
|
21 |
+
"no_refusal": "Exclude Refusal",
|
22 |
+
}
|
23 |
+
cat_name_to_explanation = {
|
24 |
+
"Overall": "Overall Questions",
|
25 |
+
"Coding": "Coding: whether conversation contains code snippets",
|
26 |
+
"Longer Query": "Longer Query (>= 500 tokens)",
|
27 |
+
"English": "English Prompts",
|
28 |
+
"Chinese": "Chinese Prompts",
|
29 |
+
"French": "French Prompts",
|
30 |
+
"Exclude Ties": "Exclude Ties and Bothbad",
|
31 |
+
"Exclude Short Query (< 5 tokens)": "Exclude Short User Query (< 5 tokens)",
|
32 |
+
"Exclude Refusal": 'Exclude model responses with refusal (e.g., "I cannot answer")',
|
33 |
+
}
|
34 |
+
|
35 |
+
fs = HfFileSystem()
|
36 |
+
|
37 |
+
def extract_date(filename):
|
38 |
+
return filename.split("/")[-1].split(".")[0].split("_")[-1]
|
39 |
+
|
40 |
+
# gather ELO data
|
41 |
+
ELO_DATA_FILES = "spaces/lmsys/chatbot-arena-leaderboard/*.pkl"
|
42 |
+
elo_files = fs.glob(ELO_DATA_FILES)
|
43 |
+
latest_elo_file = sorted(elo_files, key=extract_date, reverse=True)[0]
|
44 |
+
|
45 |
+
latest_elo_file_local = hf_hub_download(
|
46 |
+
repo_id="lmsys/chatbot-arena-leaderboard",
|
47 |
+
filename=latest_elo_file.split("/")[-1],
|
48 |
+
repo_type="space",
|
49 |
+
)
|
50 |
+
|
51 |
+
with open(latest_elo_file_local, "rb") as fin:
|
52 |
+
elo_results = pickle.load(fin)
|
53 |
+
|
54 |
+
arena_dfs = {}
|
55 |
+
for k in key_to_category_name.keys():
|
56 |
+
if k not in elo_results:
|
57 |
+
continue
|
58 |
+
arena_dfs[key_to_category_name[k]] = elo_results[k]["leaderboard_table_df"]
|
59 |
+
|
60 |
+
# gather open llm leaderboard data
|
61 |
+
LEADERBOARD_DATA_FILES = "spaces/lmsys/chatbot-arena-leaderboard/*.csv"
|
62 |
+
leaderboard_files = fs.glob(LEADERBOARD_DATA_FILES)
|
63 |
+
latest_leaderboard_file = sorted(leaderboard_files, key=extract_date, reverse=True)[
|
64 |
+
0
|
65 |
+
]
|
66 |
+
|
67 |
+
latest_leaderboard_file_local = hf_hub_download(
|
68 |
+
repo_id="lmsys/chatbot-arena-leaderboard",
|
69 |
+
filename=latest_leaderboard_file.split("/")[-1],
|
70 |
+
repo_type="space",
|
71 |
+
)
|
72 |
+
leaderboard_df = pd.read_csv(latest_leaderboard_file_local)
|
73 |
+
|
74 |
+
###################
|
75 |
+
### Prepare Data
|
76 |
+
###################
|
77 |
+
|
78 |
+
# merge leaderboard data with ELO data
|
79 |
+
merged_dfs = {}
|
80 |
+
for k, v in arena_dfs.items():
|
81 |
+
merged_dfs[k] = (
|
82 |
+
pd.merge(arena_dfs[k], leaderboard_df, left_index=True, right_on="key")
|
83 |
+
.sort_values("rating", ascending=False)
|
84 |
+
.reset_index(drop=True)
|
85 |
+
)
|
86 |
+
|
87 |
+
# add release dates into the merged data
|
88 |
+
release_date_mapping = pd.read_json("release_date_mapping.json", orient="records")
|
89 |
+
for k, v in merged_dfs.items():
|
90 |
+
merged_dfs[k] = pd.merge(
|
91 |
+
merged_dfs[k], release_date_mapping[["key", "Release Date"]], on="key"
|
92 |
+
)
|
93 |
+
df = merged_dfs["Overall"]
|
94 |
+
y_min = df["rating"].min()
|
95 |
+
y_max = df["rating"].max()
|
96 |
+
y_buffer = (y_max - y_min) * 0.1
|
97 |
+
|
98 |
+
with gr.Blocks() as demo:
|
99 |
+
gr.Markdown("# Chatbot Arena Leaderboard")
|
100 |
+
with gr.Row():
|
101 |
+
gr.ScatterPlot(
|
102 |
+
df,
|
103 |
+
title="hello",
|
104 |
+
x="Release Date",
|
105 |
+
y="rating",
|
106 |
+
tooltip=["Model", "rating", "num_battles", "Organization", "License"],
|
107 |
+
width=1000,
|
108 |
+
height=700,
|
109 |
+
x_label_angle=-45,
|
110 |
+
y_lim=[y_min - y_buffer, y_max + y_buffer],
|
111 |
+
)
|
112 |
+
|
113 |
+
if __name__ == "__main__":
|
114 |
+
demo.launch()
|
dev.ipynb
ADDED
@@ -0,0 +1,1340 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cells": [
|
3 |
+
{
|
4 |
+
"cell_type": "code",
|
5 |
+
"execution_count": 47,
|
6 |
+
"metadata": {},
|
7 |
+
"outputs": [],
|
8 |
+
"source": [
|
9 |
+
"import os\n",
|
10 |
+
"import pickle\n",
|
11 |
+
"\n",
|
12 |
+
"import pandas as pd\n",
|
13 |
+
"from huggingface_hub import HfFileSystem, hf_hub_download"
|
14 |
+
]
|
15 |
+
},
|
16 |
+
{
|
17 |
+
"cell_type": "markdown",
|
18 |
+
"metadata": {},
|
19 |
+
"source": [
|
20 |
+
"## Prepare data"
|
21 |
+
]
|
22 |
+
},
|
23 |
+
{
|
24 |
+
"cell_type": "code",
|
25 |
+
"execution_count": 72,
|
26 |
+
"metadata": {},
|
27 |
+
"outputs": [],
|
28 |
+
"source": [
|
29 |
+
"fs = HfFileSystem()\n",
|
30 |
+
"\n",
|
31 |
+
"\n",
|
32 |
+
"def extract_date(filename):\n",
|
33 |
+
" return filename.split(\"/\")[-1].split(\".\")[0].split(\"_\")[-1]\n",
|
34 |
+
"\n",
|
35 |
+
"\n",
|
36 |
+
"ELO_DATA_FILES = \"spaces/lmsys/chatbot-arena-leaderboard/*.pkl\"\n",
|
37 |
+
"elo_files = fs.glob(ELO_DATA_FILES)\n",
|
38 |
+
"latest_elo_file = sorted(elo_files, key=extract_date, reverse=True)[0]\n",
|
39 |
+
"\n",
|
40 |
+
"LEADERBOARD_DATA_FILES = \"spaces/lmsys/chatbot-arena-leaderboard/*.csv\"\n",
|
41 |
+
"leaderboard_files = fs.glob(LEADERBOARD_DATA_FILES)\n",
|
42 |
+
"latest_leaderboard_file = sorted(leaderboard_files, key=extract_date, reverse=True)[0]"
|
43 |
+
]
|
44 |
+
},
|
45 |
+
{
|
46 |
+
"cell_type": "code",
|
47 |
+
"execution_count": 73,
|
48 |
+
"metadata": {},
|
49 |
+
"outputs": [
|
50 |
+
{
|
51 |
+
"data": {
|
52 |
+
"text/plain": [
|
53 |
+
"('leaderboard_table_20240426.csv', 'elo_results_20240426.pkl')"
|
54 |
+
]
|
55 |
+
},
|
56 |
+
"execution_count": 73,
|
57 |
+
"metadata": {},
|
58 |
+
"output_type": "execute_result"
|
59 |
+
}
|
60 |
+
],
|
61 |
+
"source": [
|
62 |
+
"latest_leaderboard_file.split(\"/\")[-1], latest_elo_file.split(\"/\")[-1]"
|
63 |
+
]
|
64 |
+
},
|
65 |
+
{
|
66 |
+
"cell_type": "code",
|
67 |
+
"execution_count": 74,
|
68 |
+
"metadata": {},
|
69 |
+
"outputs": [],
|
70 |
+
"source": [
|
71 |
+
"latest_elo_file_local = hf_hub_download(\n",
|
72 |
+
" repo_id=\"lmsys/chatbot-arena-leaderboard\",\n",
|
73 |
+
" filename=latest_elo_file.split(\"/\")[-1],\n",
|
74 |
+
" repo_type=\"space\",\n",
|
75 |
+
")\n",
|
76 |
+
"latest_leaderboard_file_local = hf_hub_download(\n",
|
77 |
+
" repo_id=\"lmsys/chatbot-arena-leaderboard\",\n",
|
78 |
+
" filename=latest_leaderboard_file.split(\"/\")[-1],\n",
|
79 |
+
" repo_type=\"space\",\n",
|
80 |
+
")"
|
81 |
+
]
|
82 |
+
},
|
83 |
+
{
|
84 |
+
"cell_type": "code",
|
85 |
+
"execution_count": 76,
|
86 |
+
"metadata": {},
|
87 |
+
"outputs": [],
|
88 |
+
"source": [
|
89 |
+
"# load and prepare ELO data\n",
|
90 |
+
"key_to_category_name = {\n",
|
91 |
+
" \"full\": \"Overall\",\n",
|
92 |
+
" \"coding\": \"Coding\",\n",
|
93 |
+
" \"long_user\": \"Longer Query\",\n",
|
94 |
+
" \"english\": \"English\",\n",
|
95 |
+
" \"chinese\": \"Chinese\",\n",
|
96 |
+
" \"french\": \"French\",\n",
|
97 |
+
" \"no_tie\": \"Exclude Ties\",\n",
|
98 |
+
" \"no_short\": \"Exclude Short Query (< 5 tokens)\",\n",
|
99 |
+
" \"no_refusal\": \"Exclude Refusal\",\n",
|
100 |
+
"}\n",
|
101 |
+
"cat_name_to_explanation = {\n",
|
102 |
+
" \"Overall\": \"Overall Questions\",\n",
|
103 |
+
" \"Coding\": \"Coding: whether conversation contains code snippets\",\n",
|
104 |
+
" \"Longer Query\": \"Longer Query (>= 500 tokens)\",\n",
|
105 |
+
" \"English\": \"English Prompts\",\n",
|
106 |
+
" \"Chinese\": \"Chinese Prompts\",\n",
|
107 |
+
" \"French\": \"French Prompts\",\n",
|
108 |
+
" \"Exclude Ties\": \"Exclude Ties and Bothbad\",\n",
|
109 |
+
" \"Exclude Short Query (< 5 tokens)\": \"Exclude Short User Query (< 5 tokens)\",\n",
|
110 |
+
" \"Exclude Refusal\": 'Exclude model responses with refusal (e.g., \"I cannot answer\")',\n",
|
111 |
+
"}\n",
|
112 |
+
"\n",
|
113 |
+
"with open(latest_elo_file_local, \"rb\") as fin:\n",
|
114 |
+
" elo_results = pickle.load(fin)\n",
|
115 |
+
"\n",
|
116 |
+
"arena_dfs = {}\n",
|
117 |
+
"for k in key_to_category_name.keys():\n",
|
118 |
+
" if k not in elo_results:\n",
|
119 |
+
" continue\n",
|
120 |
+
" arena_dfs[key_to_category_name[k]] = elo_results[k][\"leaderboard_table_df\"]"
|
121 |
+
]
|
122 |
+
},
|
123 |
+
{
|
124 |
+
"cell_type": "code",
|
125 |
+
"execution_count": 77,
|
126 |
+
"metadata": {},
|
127 |
+
"outputs": [
|
128 |
+
{
|
129 |
+
"data": {
|
130 |
+
"text/plain": [
|
131 |
+
"dict_keys(['Overall', 'Coding', 'Longer Query', 'English', 'Chinese', 'French', 'Exclude Ties', 'Exclude Short Query (< 5 tokens)', 'Exclude Refusal'])"
|
132 |
+
]
|
133 |
+
},
|
134 |
+
"execution_count": 77,
|
135 |
+
"metadata": {},
|
136 |
+
"output_type": "execute_result"
|
137 |
+
}
|
138 |
+
],
|
139 |
+
"source": [
|
140 |
+
"arena_dfs.keys()"
|
141 |
+
]
|
142 |
+
},
|
143 |
+
{
|
144 |
+
"cell_type": "code",
|
145 |
+
"execution_count": 78,
|
146 |
+
"metadata": {},
|
147 |
+
"outputs": [
|
148 |
+
{
|
149 |
+
"data": {
|
150 |
+
"text/html": [
|
151 |
+
"<div>\n",
|
152 |
+
"<style scoped>\n",
|
153 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
154 |
+
" vertical-align: middle;\n",
|
155 |
+
" }\n",
|
156 |
+
"\n",
|
157 |
+
" .dataframe tbody tr th {\n",
|
158 |
+
" vertical-align: top;\n",
|
159 |
+
" }\n",
|
160 |
+
"\n",
|
161 |
+
" .dataframe thead th {\n",
|
162 |
+
" text-align: right;\n",
|
163 |
+
" }\n",
|
164 |
+
"</style>\n",
|
165 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
166 |
+
" <thead>\n",
|
167 |
+
" <tr style=\"text-align: right;\">\n",
|
168 |
+
" <th></th>\n",
|
169 |
+
" <th>rating</th>\n",
|
170 |
+
" <th>variance</th>\n",
|
171 |
+
" <th>rating_q975</th>\n",
|
172 |
+
" <th>rating_q025</th>\n",
|
173 |
+
" <th>num_battles</th>\n",
|
174 |
+
" <th>final_ranking</th>\n",
|
175 |
+
" </tr>\n",
|
176 |
+
" </thead>\n",
|
177 |
+
" <tbody>\n",
|
178 |
+
" <tr>\n",
|
179 |
+
" <th>RWKV-4-Raven-14B</th>\n",
|
180 |
+
" <td>927.710294</td>\n",
|
181 |
+
" <td>27.143015</td>\n",
|
182 |
+
" <td>935.717850</td>\n",
|
183 |
+
" <td>916.546369</td>\n",
|
184 |
+
" <td>5129</td>\n",
|
185 |
+
" <td>81</td>\n",
|
186 |
+
" </tr>\n",
|
187 |
+
" <tr>\n",
|
188 |
+
" <th>alpaca-13b</th>\n",
|
189 |
+
" <td>907.324482</td>\n",
|
190 |
+
" <td>20.736682</td>\n",
|
191 |
+
" <td>915.536856</td>\n",
|
192 |
+
" <td>899.330070</td>\n",
|
193 |
+
" <td>6111</td>\n",
|
194 |
+
" <td>85</td>\n",
|
195 |
+
" </tr>\n",
|
196 |
+
" <tr>\n",
|
197 |
+
" <th>bard-jan-24-gemini-pro</th>\n",
|
198 |
+
" <td>1208.505408</td>\n",
|
199 |
+
" <td>6.679087</td>\n",
|
200 |
+
" <td>1213.291358</td>\n",
|
201 |
+
" <td>1203.926901</td>\n",
|
202 |
+
" <td>12388</td>\n",
|
203 |
+
" <td>6</td>\n",
|
204 |
+
" </tr>\n",
|
205 |
+
" <tr>\n",
|
206 |
+
" <th>chatglm-6b</th>\n",
|
207 |
+
" <td>886.107553</td>\n",
|
208 |
+
" <td>17.110417</td>\n",
|
209 |
+
" <td>894.034333</td>\n",
|
210 |
+
" <td>878.094776</td>\n",
|
211 |
+
" <td>5195</td>\n",
|
212 |
+
" <td>86</td>\n",
|
213 |
+
" </tr>\n",
|
214 |
+
" <tr>\n",
|
215 |
+
" <th>chatglm2-6b</th>\n",
|
216 |
+
" <td>932.678460</td>\n",
|
217 |
+
" <td>33.530570</td>\n",
|
218 |
+
" <td>943.455598</td>\n",
|
219 |
+
" <td>921.346322</td>\n",
|
220 |
+
" <td>2880</td>\n",
|
221 |
+
" <td>81</td>\n",
|
222 |
+
" </tr>\n",
|
223 |
+
" <tr>\n",
|
224 |
+
" <th>...</th>\n",
|
225 |
+
" <td>...</td>\n",
|
226 |
+
" <td>...</td>\n",
|
227 |
+
" <td>...</td>\n",
|
228 |
+
" <td>...</td>\n",
|
229 |
+
" <td>...</td>\n",
|
230 |
+
" <td>...</td>\n",
|
231 |
+
" </tr>\n",
|
232 |
+
" <tr>\n",
|
233 |
+
" <th>wizardlm-70b</th>\n",
|
234 |
+
" <td>1107.992552</td>\n",
|
235 |
+
" <td>9.385887</td>\n",
|
236 |
+
" <td>1114.218223</td>\n",
|
237 |
+
" <td>1102.655575</td>\n",
|
238 |
+
" <td>8868</td>\n",
|
239 |
+
" <td>29</td>\n",
|
240 |
+
" </tr>\n",
|
241 |
+
" <tr>\n",
|
242 |
+
" <th>yi-34b-chat</th>\n",
|
243 |
+
" <td>1109.722447</td>\n",
|
244 |
+
" <td>8.596908</td>\n",
|
245 |
+
" <td>1115.182579</td>\n",
|
246 |
+
" <td>1103.991095</td>\n",
|
247 |
+
" <td>12252</td>\n",
|
248 |
+
" <td>29</td>\n",
|
249 |
+
" </tr>\n",
|
250 |
+
" <tr>\n",
|
251 |
+
" <th>zephyr-7b-alpha</th>\n",
|
252 |
+
" <td>1042.108710</td>\n",
|
253 |
+
" <td>43.900714</td>\n",
|
254 |
+
" <td>1052.991768</td>\n",
|
255 |
+
" <td>1027.160917</td>\n",
|
256 |
+
" <td>1901</td>\n",
|
257 |
+
" <td>58</td>\n",
|
258 |
+
" </tr>\n",
|
259 |
+
" <tr>\n",
|
260 |
+
" <th>zephyr-7b-beta</th>\n",
|
261 |
+
" <td>1053.655680</td>\n",
|
262 |
+
" <td>10.297607</td>\n",
|
263 |
+
" <td>1059.923254</td>\n",
|
264 |
+
" <td>1047.601629</td>\n",
|
265 |
+
" <td>11924</td>\n",
|
266 |
+
" <td>54</td>\n",
|
267 |
+
" </tr>\n",
|
268 |
+
" <tr>\n",
|
269 |
+
" <th>zephyr-orpo-141b-A35b-v0.1</th>\n",
|
270 |
+
" <td>1124.677515</td>\n",
|
271 |
+
" <td>22.288515</td>\n",
|
272 |
+
" <td>1132.728887</td>\n",
|
273 |
+
" <td>1113.848432</td>\n",
|
274 |
+
" <td>4276</td>\n",
|
275 |
+
" <td>22</td>\n",
|
276 |
+
" </tr>\n",
|
277 |
+
" </tbody>\n",
|
278 |
+
"</table>\n",
|
279 |
+
"<p>91 rows × 6 columns</p>\n",
|
280 |
+
"</div>"
|
281 |
+
],
|
282 |
+
"text/plain": [
|
283 |
+
" rating variance rating_q975 rating_q025 \\\n",
|
284 |
+
"RWKV-4-Raven-14B 927.710294 27.143015 935.717850 916.546369 \n",
|
285 |
+
"alpaca-13b 907.324482 20.736682 915.536856 899.330070 \n",
|
286 |
+
"bard-jan-24-gemini-pro 1208.505408 6.679087 1213.291358 1203.926901 \n",
|
287 |
+
"chatglm-6b 886.107553 17.110417 894.034333 878.094776 \n",
|
288 |
+
"chatglm2-6b 932.678460 33.530570 943.455598 921.346322 \n",
|
289 |
+
"... ... ... ... ... \n",
|
290 |
+
"wizardlm-70b 1107.992552 9.385887 1114.218223 1102.655575 \n",
|
291 |
+
"yi-34b-chat 1109.722447 8.596908 1115.182579 1103.991095 \n",
|
292 |
+
"zephyr-7b-alpha 1042.108710 43.900714 1052.991768 1027.160917 \n",
|
293 |
+
"zephyr-7b-beta 1053.655680 10.297607 1059.923254 1047.601629 \n",
|
294 |
+
"zephyr-orpo-141b-A35b-v0.1 1124.677515 22.288515 1132.728887 1113.848432 \n",
|
295 |
+
"\n",
|
296 |
+
" num_battles final_ranking \n",
|
297 |
+
"RWKV-4-Raven-14B 5129 81 \n",
|
298 |
+
"alpaca-13b 6111 85 \n",
|
299 |
+
"bard-jan-24-gemini-pro 12388 6 \n",
|
300 |
+
"chatglm-6b 5195 86 \n",
|
301 |
+
"chatglm2-6b 2880 81 \n",
|
302 |
+
"... ... ... \n",
|
303 |
+
"wizardlm-70b 8868 29 \n",
|
304 |
+
"yi-34b-chat 12252 29 \n",
|
305 |
+
"zephyr-7b-alpha 1901 58 \n",
|
306 |
+
"zephyr-7b-beta 11924 54 \n",
|
307 |
+
"zephyr-orpo-141b-A35b-v0.1 4276 22 \n",
|
308 |
+
"\n",
|
309 |
+
"[91 rows x 6 columns]"
|
310 |
+
]
|
311 |
+
},
|
312 |
+
"execution_count": 78,
|
313 |
+
"metadata": {},
|
314 |
+
"output_type": "execute_result"
|
315 |
+
}
|
316 |
+
],
|
317 |
+
"source": [
|
318 |
+
"arena_dfs[\"Overall\"]"
|
319 |
+
]
|
320 |
+
},
|
321 |
+
{
|
322 |
+
"cell_type": "code",
|
323 |
+
"execution_count": 79,
|
324 |
+
"metadata": {},
|
325 |
+
"outputs": [],
|
326 |
+
"source": [
|
327 |
+
"# load and prepare Leaderboard data\n",
|
328 |
+
"leaderboard_df = pd.read_csv(latest_leaderboard_file_local)"
|
329 |
+
]
|
330 |
+
},
|
331 |
+
{
|
332 |
+
"cell_type": "code",
|
333 |
+
"execution_count": 80,
|
334 |
+
"metadata": {},
|
335 |
+
"outputs": [
|
336 |
+
{
|
337 |
+
"data": {
|
338 |
+
"text/html": [
|
339 |
+
"<div>\n",
|
340 |
+
"<style scoped>\n",
|
341 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
342 |
+
" vertical-align: middle;\n",
|
343 |
+
" }\n",
|
344 |
+
"\n",
|
345 |
+
" .dataframe tbody tr th {\n",
|
346 |
+
" vertical-align: top;\n",
|
347 |
+
" }\n",
|
348 |
+
"\n",
|
349 |
+
" .dataframe thead th {\n",
|
350 |
+
" text-align: right;\n",
|
351 |
+
" }\n",
|
352 |
+
"</style>\n",
|
353 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
354 |
+
" <thead>\n",
|
355 |
+
" <tr style=\"text-align: right;\">\n",
|
356 |
+
" <th></th>\n",
|
357 |
+
" <th>key</th>\n",
|
358 |
+
" <th>Model</th>\n",
|
359 |
+
" <th>MT-bench (score)</th>\n",
|
360 |
+
" <th>MMLU</th>\n",
|
361 |
+
" <th>Knowledge cutoff date</th>\n",
|
362 |
+
" <th>License</th>\n",
|
363 |
+
" <th>Organization</th>\n",
|
364 |
+
" <th>Link</th>\n",
|
365 |
+
" </tr>\n",
|
366 |
+
" </thead>\n",
|
367 |
+
" <tbody>\n",
|
368 |
+
" <tr>\n",
|
369 |
+
" <th>0</th>\n",
|
370 |
+
" <td>wizardlm-30b</td>\n",
|
371 |
+
" <td>WizardLM-30B</td>\n",
|
372 |
+
" <td>7.01</td>\n",
|
373 |
+
" <td>0.587</td>\n",
|
374 |
+
" <td>2023/6</td>\n",
|
375 |
+
" <td>Non-commercial</td>\n",
|
376 |
+
" <td>Microsoft</td>\n",
|
377 |
+
" <td>https://huggingface.co/WizardLM/WizardLM-30B-V1.0</td>\n",
|
378 |
+
" </tr>\n",
|
379 |
+
" <tr>\n",
|
380 |
+
" <th>1</th>\n",
|
381 |
+
" <td>vicuna-13b-16k</td>\n",
|
382 |
+
" <td>Vicuna-13B-16k</td>\n",
|
383 |
+
" <td>6.92</td>\n",
|
384 |
+
" <td>0.545</td>\n",
|
385 |
+
" <td>2023/7</td>\n",
|
386 |
+
" <td>Llama 2 Community</td>\n",
|
387 |
+
" <td>LMSYS</td>\n",
|
388 |
+
" <td>https://huggingface.co/lmsys/vicuna-13b-v1.5-16k</td>\n",
|
389 |
+
" </tr>\n",
|
390 |
+
" <tr>\n",
|
391 |
+
" <th>2</th>\n",
|
392 |
+
" <td>wizardlm-13b-v1.1</td>\n",
|
393 |
+
" <td>WizardLM-13B-v1.1</td>\n",
|
394 |
+
" <td>6.76</td>\n",
|
395 |
+
" <td>0.500</td>\n",
|
396 |
+
" <td>2023/7</td>\n",
|
397 |
+
" <td>Non-commercial</td>\n",
|
398 |
+
" <td>Microsoft</td>\n",
|
399 |
+
" <td>https://huggingface.co/WizardLM/WizardLM-13B-V1.1</td>\n",
|
400 |
+
" </tr>\n",
|
401 |
+
" <tr>\n",
|
402 |
+
" <th>3</th>\n",
|
403 |
+
" <td>tulu-30b</td>\n",
|
404 |
+
" <td>Tulu-30B</td>\n",
|
405 |
+
" <td>6.43</td>\n",
|
406 |
+
" <td>0.581</td>\n",
|
407 |
+
" <td>2023/6</td>\n",
|
408 |
+
" <td>Non-commercial</td>\n",
|
409 |
+
" <td>AllenAI/UW</td>\n",
|
410 |
+
" <td>https://huggingface.co/allenai/tulu-30b</td>\n",
|
411 |
+
" </tr>\n",
|
412 |
+
" <tr>\n",
|
413 |
+
" <th>4</th>\n",
|
414 |
+
" <td>guanaco-65b</td>\n",
|
415 |
+
" <td>Guanaco-65B</td>\n",
|
416 |
+
" <td>6.41</td>\n",
|
417 |
+
" <td>0.621</td>\n",
|
418 |
+
" <td>2023/5</td>\n",
|
419 |
+
" <td>Non-commercial</td>\n",
|
420 |
+
" <td>UW</td>\n",
|
421 |
+
" <td>https://huggingface.co/timdettmers/guanaco-65b...</td>\n",
|
422 |
+
" </tr>\n",
|
423 |
+
" <tr>\n",
|
424 |
+
" <th>...</th>\n",
|
425 |
+
" <td>...</td>\n",
|
426 |
+
" <td>...</td>\n",
|
427 |
+
" <td>...</td>\n",
|
428 |
+
" <td>...</td>\n",
|
429 |
+
" <td>...</td>\n",
|
430 |
+
" <td>...</td>\n",
|
431 |
+
" <td>...</td>\n",
|
432 |
+
" <td>...</td>\n",
|
433 |
+
" </tr>\n",
|
434 |
+
" <tr>\n",
|
435 |
+
" <th>100</th>\n",
|
436 |
+
" <td>mixtral-8x22b-instruct-v0.1</td>\n",
|
437 |
+
" <td>Mixtral-8x22b-Instruct-v0.1</td>\n",
|
438 |
+
" <td>-</td>\n",
|
439 |
+
" <td>0.778</td>\n",
|
440 |
+
" <td>2024/4</td>\n",
|
441 |
+
" <td>Apache 2.0</td>\n",
|
442 |
+
" <td>Mistral</td>\n",
|
443 |
+
" <td>https://mistral.ai/news/mixtral-8x22b/</td>\n",
|
444 |
+
" </tr>\n",
|
445 |
+
" <tr>\n",
|
446 |
+
" <th>101</th>\n",
|
447 |
+
" <td>llama-3-70b-instruct</td>\n",
|
448 |
+
" <td>Llama-3-70b-Instruct</td>\n",
|
449 |
+
" <td>-</td>\n",
|
450 |
+
" <td>0.820</td>\n",
|
451 |
+
" <td>2023/12</td>\n",
|
452 |
+
" <td>Llama 3 Community</td>\n",
|
453 |
+
" <td>Meta</td>\n",
|
454 |
+
" <td>https://llama.meta.com/llama3/</td>\n",
|
455 |
+
" </tr>\n",
|
456 |
+
" <tr>\n",
|
457 |
+
" <th>102</th>\n",
|
458 |
+
" <td>llama-3-8b-instruct</td>\n",
|
459 |
+
" <td>Llama-3-8b-Instruct</td>\n",
|
460 |
+
" <td>-</td>\n",
|
461 |
+
" <td>0.684</td>\n",
|
462 |
+
" <td>2023/3</td>\n",
|
463 |
+
" <td>Llama 3 Community</td>\n",
|
464 |
+
" <td>Meta</td>\n",
|
465 |
+
" <td>https://llama.meta.com/llama3/</td>\n",
|
466 |
+
" </tr>\n",
|
467 |
+
" <tr>\n",
|
468 |
+
" <th>103</th>\n",
|
469 |
+
" <td>gemini-1.5-pro-api-0409-preview</td>\n",
|
470 |
+
" <td>Gemini 1.5 Pro API-0409-Preview</td>\n",
|
471 |
+
" <td>-</td>\n",
|
472 |
+
" <td>0.819</td>\n",
|
473 |
+
" <td>2023/11</td>\n",
|
474 |
+
" <td>Proprietary</td>\n",
|
475 |
+
" <td>Google</td>\n",
|
476 |
+
" <td>https://blog.google/technology/ai/google-gemin...</td>\n",
|
477 |
+
" </tr>\n",
|
478 |
+
" <tr>\n",
|
479 |
+
" <th>104</th>\n",
|
480 |
+
" <td>phi-3-mini-128k-instruct</td>\n",
|
481 |
+
" <td>Phi-3-Mini-128k-Instruct</td>\n",
|
482 |
+
" <td>-</td>\n",
|
483 |
+
" <td>0.681</td>\n",
|
484 |
+
" <td>2023/10</td>\n",
|
485 |
+
" <td>MIT</td>\n",
|
486 |
+
" <td>Microsoft</td>\n",
|
487 |
+
" <td>https://azure.microsoft.com/en-us/blog/introdu...</td>\n",
|
488 |
+
" </tr>\n",
|
489 |
+
" </tbody>\n",
|
490 |
+
"</table>\n",
|
491 |
+
"<p>105 rows × 8 columns</p>\n",
|
492 |
+
"</div>"
|
493 |
+
],
|
494 |
+
"text/plain": [
|
495 |
+
" key Model \\\n",
|
496 |
+
"0 wizardlm-30b WizardLM-30B \n",
|
497 |
+
"1 vicuna-13b-16k Vicuna-13B-16k \n",
|
498 |
+
"2 wizardlm-13b-v1.1 WizardLM-13B-v1.1 \n",
|
499 |
+
"3 tulu-30b Tulu-30B \n",
|
500 |
+
"4 guanaco-65b Guanaco-65B \n",
|
501 |
+
".. ... ... \n",
|
502 |
+
"100 mixtral-8x22b-instruct-v0.1 Mixtral-8x22b-Instruct-v0.1 \n",
|
503 |
+
"101 llama-3-70b-instruct Llama-3-70b-Instruct \n",
|
504 |
+
"102 llama-3-8b-instruct Llama-3-8b-Instruct \n",
|
505 |
+
"103 gemini-1.5-pro-api-0409-preview Gemini 1.5 Pro API-0409-Preview \n",
|
506 |
+
"104 phi-3-mini-128k-instruct Phi-3-Mini-128k-Instruct \n",
|
507 |
+
"\n",
|
508 |
+
" MT-bench (score) MMLU Knowledge cutoff date License \\\n",
|
509 |
+
"0 7.01 0.587 2023/6 Non-commercial \n",
|
510 |
+
"1 6.92 0.545 2023/7 Llama 2 Community \n",
|
511 |
+
"2 6.76 0.500 2023/7 Non-commercial \n",
|
512 |
+
"3 6.43 0.581 2023/6 Non-commercial \n",
|
513 |
+
"4 6.41 0.621 2023/5 Non-commercial \n",
|
514 |
+
".. ... ... ... ... \n",
|
515 |
+
"100 - 0.778 2024/4 Apache 2.0 \n",
|
516 |
+
"101 - 0.820 2023/12 Llama 3 Community \n",
|
517 |
+
"102 - 0.684 2023/3 Llama 3 Community \n",
|
518 |
+
"103 - 0.819 2023/11 Proprietary \n",
|
519 |
+
"104 - 0.681 2023/10 MIT \n",
|
520 |
+
"\n",
|
521 |
+
" Organization Link \n",
|
522 |
+
"0 Microsoft https://huggingface.co/WizardLM/WizardLM-30B-V1.0 \n",
|
523 |
+
"1 LMSYS https://huggingface.co/lmsys/vicuna-13b-v1.5-16k \n",
|
524 |
+
"2 Microsoft https://huggingface.co/WizardLM/WizardLM-13B-V1.1 \n",
|
525 |
+
"3 AllenAI/UW https://huggingface.co/allenai/tulu-30b \n",
|
526 |
+
"4 UW https://huggingface.co/timdettmers/guanaco-65b... \n",
|
527 |
+
".. ... ... \n",
|
528 |
+
"100 Mistral https://mistral.ai/news/mixtral-8x22b/ \n",
|
529 |
+
"101 Meta https://llama.meta.com/llama3/ \n",
|
530 |
+
"102 Meta https://llama.meta.com/llama3/ \n",
|
531 |
+
"103 Google https://blog.google/technology/ai/google-gemin... \n",
|
532 |
+
"104 Microsoft https://azure.microsoft.com/en-us/blog/introdu... \n",
|
533 |
+
"\n",
|
534 |
+
"[105 rows x 8 columns]"
|
535 |
+
]
|
536 |
+
},
|
537 |
+
"execution_count": 80,
|
538 |
+
"metadata": {},
|
539 |
+
"output_type": "execute_result"
|
540 |
+
}
|
541 |
+
],
|
542 |
+
"source": [
|
543 |
+
"leaderboard_df"
|
544 |
+
]
|
545 |
+
},
|
546 |
+
{
|
547 |
+
"cell_type": "code",
|
548 |
+
"execution_count": 82,
|
549 |
+
"metadata": {},
|
550 |
+
"outputs": [
|
551 |
+
{
|
552 |
+
"data": {
|
553 |
+
"text/plain": [
|
554 |
+
"dict_keys(['Overall', 'Coding', 'Longer Query', 'English', 'Chinese', 'French', 'Exclude Ties', 'Exclude Short Query (< 5 tokens)', 'Exclude Refusal'])"
|
555 |
+
]
|
556 |
+
},
|
557 |
+
"execution_count": 82,
|
558 |
+
"metadata": {},
|
559 |
+
"output_type": "execute_result"
|
560 |
+
}
|
561 |
+
],
|
562 |
+
"source": [
|
563 |
+
"arena_dfs.keys()"
|
564 |
+
]
|
565 |
+
},
|
566 |
+
{
|
567 |
+
"cell_type": "code",
|
568 |
+
"execution_count": 86,
|
569 |
+
"metadata": {},
|
570 |
+
"outputs": [],
|
571 |
+
"source": [
|
572 |
+
"# merge ELO and Leaderboard data\n",
|
573 |
+
"merged_dfs = {}\n",
|
574 |
+
"for k, v in arena_dfs.items():\n",
|
575 |
+
" merged_dfs[k] = (\n",
|
576 |
+
" pd.merge(arena_dfs[k], leaderboard_df, left_index=True, right_on=\"key\")\n",
|
577 |
+
" .sort_values(\"rating\", ascending=False)\n",
|
578 |
+
" .reset_index(drop=True)\n",
|
579 |
+
" )"
|
580 |
+
]
|
581 |
+
},
|
582 |
+
{
|
583 |
+
"cell_type": "code",
|
584 |
+
"execution_count": 101,
|
585 |
+
"metadata": {},
|
586 |
+
"outputs": [
|
587 |
+
{
|
588 |
+
"data": {
|
589 |
+
"text/html": [
|
590 |
+
"<div>\n",
|
591 |
+
"<style scoped>\n",
|
592 |
+
" .dataframe tbody tr th:only-of-type {\n",
|
593 |
+
" vertical-align: middle;\n",
|
594 |
+
" }\n",
|
595 |
+
"\n",
|
596 |
+
" .dataframe tbody tr th {\n",
|
597 |
+
" vertical-align: top;\n",
|
598 |
+
" }\n",
|
599 |
+
"\n",
|
600 |
+
" .dataframe thead th {\n",
|
601 |
+
" text-align: right;\n",
|
602 |
+
" }\n",
|
603 |
+
"</style>\n",
|
604 |
+
"<table border=\"1\" class=\"dataframe\">\n",
|
605 |
+
" <thead>\n",
|
606 |
+
" <tr style=\"text-align: right;\">\n",
|
607 |
+
" <th></th>\n",
|
608 |
+
" <th>rating</th>\n",
|
609 |
+
" <th>variance</th>\n",
|
610 |
+
" <th>rating_q975</th>\n",
|
611 |
+
" <th>rating_q025</th>\n",
|
612 |
+
" <th>num_battles</th>\n",
|
613 |
+
" <th>final_ranking</th>\n",
|
614 |
+
" <th>key</th>\n",
|
615 |
+
" <th>Model</th>\n",
|
616 |
+
" <th>MT-bench (score)</th>\n",
|
617 |
+
" <th>MMLU</th>\n",
|
618 |
+
" <th>Knowledge cutoff date</th>\n",
|
619 |
+
" <th>License</th>\n",
|
620 |
+
" <th>Organization</th>\n",
|
621 |
+
" <th>Link</th>\n",
|
622 |
+
" </tr>\n",
|
623 |
+
" </thead>\n",
|
624 |
+
" <tbody>\n",
|
625 |
+
" <tr>\n",
|
626 |
+
" <th>0</th>\n",
|
627 |
+
" <td>1257.399407</td>\n",
|
628 |
+
" <td>4.283316</td>\n",
|
629 |
+
" <td>1261.676224</td>\n",
|
630 |
+
" <td>1254.003626</td>\n",
|
631 |
+
" <td>30562</td>\n",
|
632 |
+
" <td>1</td>\n",
|
633 |
+
" <td>gpt-4-turbo-2024-04-09</td>\n",
|
634 |
+
" <td>GPT-4-Turbo-2024-04-09</td>\n",
|
635 |
+
" <td>-</td>\n",
|
636 |
+
" <td>-</td>\n",
|
637 |
+
" <td>2023/12</td>\n",
|
638 |
+
" <td>Proprietary</td>\n",
|
639 |
+
" <td>OpenAI</td>\n",
|
640 |
+
" <td>https://platform.openai.com/docs/models/gpt-4-...</td>\n",
|
641 |
+
" </tr>\n",
|
642 |
+
" <tr>\n",
|
643 |
+
" <th>1</th>\n",
|
644 |
+
" <td>1253.025095</td>\n",
|
645 |
+
" <td>2.069534</td>\n",
|
646 |
+
" <td>1256.111392</td>\n",
|
647 |
+
" <td>1250.435207</td>\n",
|
648 |
+
" <td>69871</td>\n",
|
649 |
+
" <td>1</td>\n",
|
650 |
+
" <td>gpt-4-1106-preview</td>\n",
|
651 |
+
" <td>GPT-4-1106-preview</td>\n",
|
652 |
+
" <td>9.32</td>\n",
|
653 |
+
" <td>-</td>\n",
|
654 |
+
" <td>2023/4</td>\n",
|
655 |
+
" <td>Proprietary</td>\n",
|
656 |
+
" <td>OpenAI</td>\n",
|
657 |
+
" <td>https://openai.com/blog/new-models-and-develop...</td>\n",
|
658 |
+
" </tr>\n",
|
659 |
+
" <tr>\n",
|
660 |
+
" <th>2</th>\n",
|
661 |
+
" <td>1251.114220</td>\n",
|
662 |
+
" <td>1.862842</td>\n",
|
663 |
+
" <td>1253.629093</td>\n",
|
664 |
+
" <td>1248.362042</td>\n",
|
665 |
+
" <td>75684</td>\n",
|
666 |
+
" <td>2</td>\n",
|
667 |
+
" <td>claude-3-opus-20240229</td>\n",
|
668 |
+
" <td>Claude 3 Opus</td>\n",
|
669 |
+
" <td>-</td>\n",
|
670 |
+
" <td>0.868</td>\n",
|
671 |
+
" <td>2023/8</td>\n",
|
672 |
+
" <td>Proprietary</td>\n",
|
673 |
+
" <td>Anthropic</td>\n",
|
674 |
+
" <td>https://www.anthropic.com/news/claude-3-family</td>\n",
|
675 |
+
" </tr>\n",
|
676 |
+
" <tr>\n",
|
677 |
+
" <th>3</th>\n",
|
678 |
+
" <td>1247.662508</td>\n",
|
679 |
+
" <td>3.263747</td>\n",
|
680 |
+
" <td>1251.582645</td>\n",
|
681 |
+
" <td>1244.380454</td>\n",
|
682 |
+
" <td>33723</td>\n",
|
683 |
+
" <td>2</td>\n",
|
684 |
+
" <td>gemini-1.5-pro-api-0409-preview</td>\n",
|
685 |
+
" <td>Gemini 1.5 Pro API-0409-Preview</td>\n",
|
686 |
+
" <td>-</td>\n",
|
687 |
+
" <td>0.819</td>\n",
|
688 |
+
" <td>2023/11</td>\n",
|
689 |
+
" <td>Proprietary</td>\n",
|
690 |
+
" <td>Google</td>\n",
|
691 |
+
" <td>https://blog.google/technology/ai/google-gemin...</td>\n",
|
692 |
+
" </tr>\n",
|
693 |
+
" <tr>\n",
|
694 |
+
" <th>4</th>\n",
|
695 |
+
" <td>1247.277052</td>\n",
|
696 |
+
" <td>1.923014</td>\n",
|
697 |
+
" <td>1249.489411</td>\n",
|
698 |
+
" <td>1244.340257</td>\n",
|
699 |
+
" <td>61924</td>\n",
|
700 |
+
" <td>3</td>\n",
|
701 |
+
" <td>gpt-4-0125-preview</td>\n",
|
702 |
+
" <td>GPT-4-0125-preview</td>\n",
|
703 |
+
" <td>-</td>\n",
|
704 |
+
" <td>-</td>\n",
|
705 |
+
" <td>2023/12</td>\n",
|
706 |
+
" <td>Proprietary</td>\n",
|
707 |
+
" <td>OpenAI</td>\n",
|
708 |
+
" <td>https://openai.com/blog/new-models-and-develop...</td>\n",
|
709 |
+
" </tr>\n",
|
710 |
+
" <tr>\n",
|
711 |
+
" <th>5</th>\n",
|
712 |
+
" <td>1208.505408</td>\n",
|
713 |
+
" <td>6.679087</td>\n",
|
714 |
+
" <td>1213.291358</td>\n",
|
715 |
+
" <td>1203.926901</td>\n",
|
716 |
+
" <td>12388</td>\n",
|
717 |
+
" <td>6</td>\n",
|
718 |
+
" <td>bard-jan-24-gemini-pro</td>\n",
|
719 |
+
" <td>Bard (Gemini Pro)</td>\n",
|
720 |
+
" <td>-</td>\n",
|
721 |
+
" <td>-</td>\n",
|
722 |
+
" <td>Online</td>\n",
|
723 |
+
" <td>Proprietary</td>\n",
|
724 |
+
" <td>Google</td>\n",
|
725 |
+
" <td>https://bard.google.com/</td>\n",
|
726 |
+
" </tr>\n",
|
727 |
+
" <tr>\n",
|
728 |
+
" <th>6</th>\n",
|
729 |
+
" <td>1207.497541</td>\n",
|
730 |
+
" <td>4.109466</td>\n",
|
731 |
+
" <td>1211.720734</td>\n",
|
732 |
+
" <td>1203.322762</td>\n",
|
733 |
+
" <td>27298</td>\n",
|
734 |
+
" <td>6</td>\n",
|
735 |
+
" <td>llama-3-70b-instruct</td>\n",
|
736 |
+
" <td>Llama-3-70b-Instruct</td>\n",
|
737 |
+
" <td>-</td>\n",
|
738 |
+
" <td>0.820</td>\n",
|
739 |
+
" <td>2023/12</td>\n",
|
740 |
+
" <td>Llama 3 Community</td>\n",
|
741 |
+
" <td>Meta</td>\n",
|
742 |
+
" <td>https://llama.meta.com/llama3/</td>\n",
|
743 |
+
" </tr>\n",
|
744 |
+
" <tr>\n",
|
745 |
+
" <th>7</th>\n",
|
746 |
+
" <td>1201.671254</td>\n",
|
747 |
+
" <td>2.525563</td>\n",
|
748 |
+
" <td>1204.862512</td>\n",
|
749 |
+
" <td>1198.658822</td>\n",
|
750 |
+
" <td>75418</td>\n",
|
751 |
+
" <td>6</td>\n",
|
752 |
+
" <td>claude-3-sonnet-20240229</td>\n",
|
753 |
+
" <td>Claude 3 Sonnet</td>\n",
|
754 |
+
" <td>-</td>\n",
|
755 |
+
" <td>0.790</td>\n",
|
756 |
+
" <td>2023/8</td>\n",
|
757 |
+
" <td>Proprietary</td>\n",
|
758 |
+
" <td>Anthropic</td>\n",
|
759 |
+
" <td>https://www.anthropic.com/news/claude-3-family</td>\n",
|
760 |
+
" </tr>\n",
|
761 |
+
" <tr>\n",
|
762 |
+
" <th>8</th>\n",
|
763 |
+
" <td>1191.684542</td>\n",
|
764 |
+
" <td>3.459717</td>\n",
|
765 |
+
" <td>1195.080256</td>\n",
|
766 |
+
" <td>1188.222382</td>\n",
|
767 |
+
" <td>41262</td>\n",
|
768 |
+
" <td>9</td>\n",
|
769 |
+
" <td>command-r-plus</td>\n",
|
770 |
+
" <td>Command R+</td>\n",
|
771 |
+
" <td>-</td>\n",
|
772 |
+
" <td>-</td>\n",
|
773 |
+
" <td>2024/3</td>\n",
|
774 |
+
" <td>CC-BY-NC-4.0</td>\n",
|
775 |
+
" <td>Cohere</td>\n",
|
776 |
+
" <td>https://txt.cohere.com/command-r-plus-microsof...</td>\n",
|
777 |
+
" </tr>\n",
|
778 |
+
" <tr>\n",
|
779 |
+
" <th>9</th>\n",
|
780 |
+
" <td>1188.987389</td>\n",
|
781 |
+
" <td>3.124792</td>\n",
|
782 |
+
" <td>1193.335535</td>\n",
|
783 |
+
" <td>1185.935928</td>\n",
|
784 |
+
" <td>48390</td>\n",
|
785 |
+
" <td>9</td>\n",
|
786 |
+
" <td>gpt-4-0314</td>\n",
|
787 |
+
" <td>GPT-4-0314</td>\n",
|
788 |
+
" <td>8.96</td>\n",
|
789 |
+
" <td>0.864</td>\n",
|
790 |
+
" <td>2021/9</td>\n",
|
791 |
+
" <td>Proprietary</td>\n",
|
792 |
+
" <td>OpenAI</td>\n",
|
793 |
+
" <td>https://openai.com/research/gpt-4</td>\n",
|
794 |
+
" </tr>\n",
|
795 |
+
" <tr>\n",
|
796 |
+
" <th>10</th>\n",
|
797 |
+
" <td>1180.606870</td>\n",
|
798 |
+
" <td>3.097542</td>\n",
|
799 |
+
" <td>1183.825403</td>\n",
|
800 |
+
" <td>1177.255203</td>\n",
|
801 |
+
" <td>66065</td>\n",
|
802 |
+
" <td>11</td>\n",
|
803 |
+
" <td>claude-3-haiku-20240307</td>\n",
|
804 |
+
" <td>Claude 3 Haiku</td>\n",
|
805 |
+
" <td>-</td>\n",
|
806 |
+
" <td>0.752</td>\n",
|
807 |
+
" <td>2023/8</td>\n",
|
808 |
+
" <td>Proprietary</td>\n",
|
809 |
+
" <td>Anthropic</td>\n",
|
810 |
+
" <td>https://www.anthropic.com/news/claude-3-family</td>\n",
|
811 |
+
" </tr>\n",
|
812 |
+
" <tr>\n",
|
813 |
+
" <th>11</th>\n",
|
814 |
+
" <td>1164.896561</td>\n",
|
815 |
+
" <td>2.585577</td>\n",
|
816 |
+
" <td>1167.595696</td>\n",
|
817 |
+
" <td>1161.727454</td>\n",
|
818 |
+
" <td>67038</td>\n",
|
819 |
+
" <td>12</td>\n",
|
820 |
+
" <td>gpt-4-0613</td>\n",
|
821 |
+
" <td>GPT-4-0613</td>\n",
|
822 |
+
" <td>9.18</td>\n",
|
823 |
+
" <td>-</td>\n",
|
824 |
+
" <td>2021/9</td>\n",
|
825 |
+
" <td>Proprietary</td>\n",
|
826 |
+
" <td>OpenAI</td>\n",
|
827 |
+
" <td>https://platform.openai.com/docs/models/gpt-4-...</td>\n",
|
828 |
+
" </tr>\n",
|
829 |
+
" <tr>\n",
|
830 |
+
" <th>12</th>\n",
|
831 |
+
" <td>1157.638992</td>\n",
|
832 |
+
" <td>2.541320</td>\n",
|
833 |
+
" <td>1160.496116</td>\n",
|
834 |
+
" <td>1154.927748</td>\n",
|
835 |
+
" <td>44120</td>\n",
|
836 |
+
" <td>13</td>\n",
|
837 |
+
" <td>mistral-large-2402</td>\n",
|
838 |
+
" <td>Mistral-Large-2402</td>\n",
|
839 |
+
" <td>-</td>\n",
|
840 |
+
" <td>0.812</td>\n",
|
841 |
+
" <td>-</td>\n",
|
842 |
+
" <td>Proprietary</td>\n",
|
843 |
+
" <td>Mistral</td>\n",
|
844 |
+
" <td>https://mistral.ai/news/mistral-large/</td>\n",
|
845 |
+
" </tr>\n",
|
846 |
+
" <tr>\n",
|
847 |
+
" <th>13</th>\n",
|
848 |
+
" <td>1153.464280</td>\n",
|
849 |
+
" <td>3.631512</td>\n",
|
850 |
+
" <td>1157.068850</td>\n",
|
851 |
+
" <td>1150.178903</td>\n",
|
852 |
+
" <td>32999</td>\n",
|
853 |
+
" <td>13</td>\n",
|
854 |
+
" <td>qwen1.5-72b-chat</td>\n",
|
855 |
+
" <td>Qwen1.5-72B-Chat</td>\n",
|
856 |
+
" <td>8.61</td>\n",
|
857 |
+
" <td>0.775</td>\n",
|
858 |
+
" <td>2024/2</td>\n",
|
859 |
+
" <td>Qianwen LICENSE</td>\n",
|
860 |
+
" <td>Alibaba</td>\n",
|
861 |
+
" <td>https://qwenlm.github.io/blog/qwen1.5/</td>\n",
|
862 |
+
" </tr>\n",
|
863 |
+
" <tr>\n",
|
864 |
+
" <th>14</th>\n",
|
865 |
+
" <td>1150.918473</td>\n",
|
866 |
+
" <td>9.062217</td>\n",
|
867 |
+
" <td>1155.969721</td>\n",
|
868 |
+
" <td>1145.229885</td>\n",
|
869 |
+
" <td>8622</td>\n",
|
870 |
+
" <td>13</td>\n",
|
871 |
+
" <td>reka-flash-21b-20240226-online</td>\n",
|
872 |
+
" <td>Reka-Flash-21B-online</td>\n",
|
873 |
+
" <td>-</td>\n",
|
874 |
+
" <td>-</td>\n",
|
875 |
+
" <td>Online</td>\n",
|
876 |
+
" <td>Proprietary</td>\n",
|
877 |
+
" <td>Reka AI</td>\n",
|
878 |
+
" <td>https://docs.reka.ai/http-api.html#generation</td>\n",
|
879 |
+
" </tr>\n",
|
880 |
+
" <tr>\n",
|
881 |
+
" <th>15</th>\n",
|
882 |
+
" <td>1150.244313</td>\n",
|
883 |
+
" <td>5.551373</td>\n",
|
884 |
+
" <td>1154.745214</td>\n",
|
885 |
+
" <td>1145.496466</td>\n",
|
886 |
+
" <td>21768</td>\n",
|
887 |
+
" <td>14</td>\n",
|
888 |
+
" <td>claude-1</td>\n",
|
889 |
+
" <td>Claude-1</td>\n",
|
890 |
+
" <td>7.90</td>\n",
|
891 |
+
" <td>0.770</td>\n",
|
892 |
+
" <td>-</td>\n",
|
893 |
+
" <td>Proprietary</td>\n",
|
894 |
+
" <td>Anthropic</td>\n",
|
895 |
+
" <td>https://www.anthropic.com/index/introducing-cl...</td>\n",
|
896 |
+
" </tr>\n",
|
897 |
+
" <tr>\n",
|
898 |
+
" <th>16</th>\n",
|
899 |
+
" <td>1149.267578</td>\n",
|
900 |
+
" <td>11.452272</td>\n",
|
901 |
+
" <td>1154.290155</td>\n",
|
902 |
+
" <td>1141.931621</td>\n",
|
903 |
+
" <td>9059</td>\n",
|
904 |
+
" <td>14</td>\n",
|
905 |
+
" <td>reka-flash-21b-20240226</td>\n",
|
906 |
+
" <td>Reka-Flash-21B</td>\n",
|
907 |
+
" <td>-</td>\n",
|
908 |
+
" <td>0.735</td>\n",
|
909 |
+
" <td>2023/11</td>\n",
|
910 |
+
" <td>Proprietary</td>\n",
|
911 |
+
" <td>Reka AI</td>\n",
|
912 |
+
" <td>https://www.reka.ai/news/reka-flash-efficient-...</td>\n",
|
913 |
+
" </tr>\n",
|
914 |
+
" <tr>\n",
|
915 |
+
" <th>17</th>\n",
|
916 |
+
" <td>1148.072155</td>\n",
|
917 |
+
" <td>3.071222</td>\n",
|
918 |
+
" <td>1151.980865</td>\n",
|
919 |
+
" <td>1144.992044</td>\n",
|
920 |
+
" <td>37413</td>\n",
|
921 |
+
" <td>14</td>\n",
|
922 |
+
" <td>command-r</td>\n",
|
923 |
+
" <td>Command R</td>\n",
|
924 |
+
" <td>-</td>\n",
|
925 |
+
" <td>-</td>\n",
|
926 |
+
" <td>2024/3</td>\n",
|
927 |
+
" <td>CC-BY-NC-4.0</td>\n",
|
928 |
+
" <td>Cohere</td>\n",
|
929 |
+
" <td>https://txt.cohere.com/command-r</td>\n",
|
930 |
+
" </tr>\n",
|
931 |
+
" <tr>\n",
|
932 |
+
" <th>18</th>\n",
|
933 |
+
" <td>1147.668325</td>\n",
|
934 |
+
" <td>3.542229</td>\n",
|
935 |
+
" <td>1150.726489</td>\n",
|
936 |
+
" <td>1143.868385</td>\n",
|
937 |
+
" <td>32738</td>\n",
|
938 |
+
" <td>14</td>\n",
|
939 |
+
" <td>mistral-medium</td>\n",
|
940 |
+
" <td>Mistral Medium</td>\n",
|
941 |
+
" <td>8.61</td>\n",
|
942 |
+
" <td>0.753</td>\n",
|
943 |
+
" <td>-</td>\n",
|
944 |
+
" <td>Proprietary</td>\n",
|
945 |
+
" <td>Mistral</td>\n",
|
946 |
+
" <td>https://mistral.ai/news/la-plateforme/</td>\n",
|
947 |
+
" </tr>\n",
|
948 |
+
" <tr>\n",
|
949 |
+
" <th>19</th>\n",
|
950 |
+
" <td>1147.473989</td>\n",
|
951 |
+
" <td>5.789710</td>\n",
|
952 |
+
" <td>1151.989352</td>\n",
|
953 |
+
" <td>1143.322918</td>\n",
|
954 |
+
" <td>17214</td>\n",
|
955 |
+
" <td>14</td>\n",
|
956 |
+
" <td>mixtral-8x22b-instruct-v0.1</td>\n",
|
957 |
+
" <td>Mixtral-8x22b-Instruct-v0.1</td>\n",
|
958 |
+
" <td>-</td>\n",
|
959 |
+
" <td>0.778</td>\n",
|
960 |
+
" <td>2024/4</td>\n",
|
961 |
+
" <td>Apache 2.0</td>\n",
|
962 |
+
" <td>Mistral</td>\n",
|
963 |
+
" <td>https://mistral.ai/news/mixtral-8x22b/</td>\n",
|
964 |
+
" </tr>\n",
|
965 |
+
" </tbody>\n",
|
966 |
+
"</table>\n",
|
967 |
+
"</div>"
|
968 |
+
],
|
969 |
+
"text/plain": [
|
970 |
+
" rating variance rating_q975 rating_q025 num_battles \\\n",
|
971 |
+
"0 1257.399407 4.283316 1261.676224 1254.003626 30562 \n",
|
972 |
+
"1 1253.025095 2.069534 1256.111392 1250.435207 69871 \n",
|
973 |
+
"2 1251.114220 1.862842 1253.629093 1248.362042 75684 \n",
|
974 |
+
"3 1247.662508 3.263747 1251.582645 1244.380454 33723 \n",
|
975 |
+
"4 1247.277052 1.923014 1249.489411 1244.340257 61924 \n",
|
976 |
+
"5 1208.505408 6.679087 1213.291358 1203.926901 12388 \n",
|
977 |
+
"6 1207.497541 4.109466 1211.720734 1203.322762 27298 \n",
|
978 |
+
"7 1201.671254 2.525563 1204.862512 1198.658822 75418 \n",
|
979 |
+
"8 1191.684542 3.459717 1195.080256 1188.222382 41262 \n",
|
980 |
+
"9 1188.987389 3.124792 1193.335535 1185.935928 48390 \n",
|
981 |
+
"10 1180.606870 3.097542 1183.825403 1177.255203 66065 \n",
|
982 |
+
"11 1164.896561 2.585577 1167.595696 1161.727454 67038 \n",
|
983 |
+
"12 1157.638992 2.541320 1160.496116 1154.927748 44120 \n",
|
984 |
+
"13 1153.464280 3.631512 1157.068850 1150.178903 32999 \n",
|
985 |
+
"14 1150.918473 9.062217 1155.969721 1145.229885 8622 \n",
|
986 |
+
"15 1150.244313 5.551373 1154.745214 1145.496466 21768 \n",
|
987 |
+
"16 1149.267578 11.452272 1154.290155 1141.931621 9059 \n",
|
988 |
+
"17 1148.072155 3.071222 1151.980865 1144.992044 37413 \n",
|
989 |
+
"18 1147.668325 3.542229 1150.726489 1143.868385 32738 \n",
|
990 |
+
"19 1147.473989 5.789710 1151.989352 1143.322918 17214 \n",
|
991 |
+
"\n",
|
992 |
+
" final_ranking key \\\n",
|
993 |
+
"0 1 gpt-4-turbo-2024-04-09 \n",
|
994 |
+
"1 1 gpt-4-1106-preview \n",
|
995 |
+
"2 2 claude-3-opus-20240229 \n",
|
996 |
+
"3 2 gemini-1.5-pro-api-0409-preview \n",
|
997 |
+
"4 3 gpt-4-0125-preview \n",
|
998 |
+
"5 6 bard-jan-24-gemini-pro \n",
|
999 |
+
"6 6 llama-3-70b-instruct \n",
|
1000 |
+
"7 6 claude-3-sonnet-20240229 \n",
|
1001 |
+
"8 9 command-r-plus \n",
|
1002 |
+
"9 9 gpt-4-0314 \n",
|
1003 |
+
"10 11 claude-3-haiku-20240307 \n",
|
1004 |
+
"11 12 gpt-4-0613 \n",
|
1005 |
+
"12 13 mistral-large-2402 \n",
|
1006 |
+
"13 13 qwen1.5-72b-chat \n",
|
1007 |
+
"14 13 reka-flash-21b-20240226-online \n",
|
1008 |
+
"15 14 claude-1 \n",
|
1009 |
+
"16 14 reka-flash-21b-20240226 \n",
|
1010 |
+
"17 14 command-r \n",
|
1011 |
+
"18 14 mistral-medium \n",
|
1012 |
+
"19 14 mixtral-8x22b-instruct-v0.1 \n",
|
1013 |
+
"\n",
|
1014 |
+
" Model MT-bench (score) MMLU \\\n",
|
1015 |
+
"0 GPT-4-Turbo-2024-04-09 - - \n",
|
1016 |
+
"1 GPT-4-1106-preview 9.32 - \n",
|
1017 |
+
"2 Claude 3 Opus - 0.868 \n",
|
1018 |
+
"3 Gemini 1.5 Pro API-0409-Preview - 0.819 \n",
|
1019 |
+
"4 GPT-4-0125-preview - - \n",
|
1020 |
+
"5 Bard (Gemini Pro) - - \n",
|
1021 |
+
"6 Llama-3-70b-Instruct - 0.820 \n",
|
1022 |
+
"7 Claude 3 Sonnet - 0.790 \n",
|
1023 |
+
"8 Command R+ - - \n",
|
1024 |
+
"9 GPT-4-0314 8.96 0.864 \n",
|
1025 |
+
"10 Claude 3 Haiku - 0.752 \n",
|
1026 |
+
"11 GPT-4-0613 9.18 - \n",
|
1027 |
+
"12 Mistral-Large-2402 - 0.812 \n",
|
1028 |
+
"13 Qwen1.5-72B-Chat 8.61 0.775 \n",
|
1029 |
+
"14 Reka-Flash-21B-online - - \n",
|
1030 |
+
"15 Claude-1 7.90 0.770 \n",
|
1031 |
+
"16 Reka-Flash-21B - 0.735 \n",
|
1032 |
+
"17 Command R - - \n",
|
1033 |
+
"18 Mistral Medium 8.61 0.753 \n",
|
1034 |
+
"19 Mixtral-8x22b-Instruct-v0.1 - 0.778 \n",
|
1035 |
+
"\n",
|
1036 |
+
" Knowledge cutoff date License Organization \\\n",
|
1037 |
+
"0 2023/12 Proprietary OpenAI \n",
|
1038 |
+
"1 2023/4 Proprietary OpenAI \n",
|
1039 |
+
"2 2023/8 Proprietary Anthropic \n",
|
1040 |
+
"3 2023/11 Proprietary Google \n",
|
1041 |
+
"4 2023/12 Proprietary OpenAI \n",
|
1042 |
+
"5 Online Proprietary Google \n",
|
1043 |
+
"6 2023/12 Llama 3 Community Meta \n",
|
1044 |
+
"7 2023/8 Proprietary Anthropic \n",
|
1045 |
+
"8 2024/3 CC-BY-NC-4.0 Cohere \n",
|
1046 |
+
"9 2021/9 Proprietary OpenAI \n",
|
1047 |
+
"10 2023/8 Proprietary Anthropic \n",
|
1048 |
+
"11 2021/9 Proprietary OpenAI \n",
|
1049 |
+
"12 - Proprietary Mistral \n",
|
1050 |
+
"13 2024/2 Qianwen LICENSE Alibaba \n",
|
1051 |
+
"14 Online Proprietary Reka AI \n",
|
1052 |
+
"15 - Proprietary Anthropic \n",
|
1053 |
+
"16 2023/11 Proprietary Reka AI \n",
|
1054 |
+
"17 2024/3 CC-BY-NC-4.0 Cohere \n",
|
1055 |
+
"18 - Proprietary Mistral \n",
|
1056 |
+
"19 2024/4 Apache 2.0 Mistral \n",
|
1057 |
+
"\n",
|
1058 |
+
" Link \n",
|
1059 |
+
"0 https://platform.openai.com/docs/models/gpt-4-... \n",
|
1060 |
+
"1 https://openai.com/blog/new-models-and-develop... \n",
|
1061 |
+
"2 https://www.anthropic.com/news/claude-3-family \n",
|
1062 |
+
"3 https://blog.google/technology/ai/google-gemin... \n",
|
1063 |
+
"4 https://openai.com/blog/new-models-and-develop... \n",
|
1064 |
+
"5 https://bard.google.com/ \n",
|
1065 |
+
"6 https://llama.meta.com/llama3/ \n",
|
1066 |
+
"7 https://www.anthropic.com/news/claude-3-family \n",
|
1067 |
+
"8 https://txt.cohere.com/command-r-plus-microsof... \n",
|
1068 |
+
"9 https://openai.com/research/gpt-4 \n",
|
1069 |
+
"10 https://www.anthropic.com/news/claude-3-family \n",
|
1070 |
+
"11 https://platform.openai.com/docs/models/gpt-4-... \n",
|
1071 |
+
"12 https://mistral.ai/news/mistral-large/ \n",
|
1072 |
+
"13 https://qwenlm.github.io/blog/qwen1.5/ \n",
|
1073 |
+
"14 https://docs.reka.ai/http-api.html#generation \n",
|
1074 |
+
"15 https://www.anthropic.com/index/introducing-cl... \n",
|
1075 |
+
"16 https://www.reka.ai/news/reka-flash-efficient-... \n",
|
1076 |
+
"17 https://txt.cohere.com/command-r \n",
|
1077 |
+
"18 https://mistral.ai/news/la-plateforme/ \n",
|
1078 |
+
"19 https://mistral.ai/news/mixtral-8x22b/ "
|
1079 |
+
]
|
1080 |
+
},
|
1081 |
+
"execution_count": 101,
|
1082 |
+
"metadata": {},
|
1083 |
+
"output_type": "execute_result"
|
1084 |
+
}
|
1085 |
+
],
|
1086 |
+
"source": [
|
1087 |
+
"merged_dfs[\"Overall\"][:20]"
|
1088 |
+
]
|
1089 |
+
},
|
1090 |
+
{
|
1091 |
+
"cell_type": "markdown",
|
1092 |
+
"metadata": {},
|
1093 |
+
"source": [
|
1094 |
+
"### Manually map release dates - MEH."
|
1095 |
+
]
|
1096 |
+
},
|
1097 |
+
{
|
1098 |
+
"cell_type": "code",
|
1099 |
+
"execution_count": 113,
|
1100 |
+
"metadata": {},
|
1101 |
+
"outputs": [],
|
1102 |
+
"source": [
|
1103 |
+
"t = merged_dfs[\"Overall\"].loc[:, [\"key\", \"Model\"]]\n",
|
1104 |
+
"t[\"Release Date\"] = \"\""
|
1105 |
+
]
|
1106 |
+
},
|
1107 |
+
{
|
1108 |
+
"cell_type": "code",
|
1109 |
+
"execution_count": 120,
|
1110 |
+
"metadata": {},
|
1111 |
+
"outputs": [],
|
1112 |
+
"source": [
|
1113 |
+
"t.to_json(\"release_date_mapping.json\", orient=\"records\", lines=True)"
|
1114 |
+
]
|
1115 |
+
},
|
1116 |
+
{
|
1117 |
+
"cell_type": "code",
|
1118 |
+
"execution_count": 119,
|
1119 |
+
"metadata": {},
|
1120 |
+
"outputs": [
|
1121 |
+
{
|
1122 |
+
"data": {
|
1123 |
+
"text/plain": [
|
1124 |
+
"[{'key': 'gpt-4-turbo-2024-04-09',\n",
|
1125 |
+
" 'Model': 'GPT-4-Turbo-2024-04-09',\n",
|
1126 |
+
" 'Release Date': ''},\n",
|
1127 |
+
" {'key': 'gpt-4-1106-preview',\n",
|
1128 |
+
" 'Model': 'GPT-4-1106-preview',\n",
|
1129 |
+
" 'Release Date': ''},\n",
|
1130 |
+
" {'key': 'claude-3-opus-20240229',\n",
|
1131 |
+
" 'Model': 'Claude 3 Opus',\n",
|
1132 |
+
" 'Release Date': ''},\n",
|
1133 |
+
" {'key': 'gemini-1.5-pro-api-0409-preview',\n",
|
1134 |
+
" 'Model': 'Gemini 1.5 Pro API-0409-Preview',\n",
|
1135 |
+
" 'Release Date': ''},\n",
|
1136 |
+
" {'key': 'gpt-4-0125-preview',\n",
|
1137 |
+
" 'Model': 'GPT-4-0125-preview',\n",
|
1138 |
+
" 'Release Date': ''},\n",
|
1139 |
+
" {'key': 'bard-jan-24-gemini-pro',\n",
|
1140 |
+
" 'Model': 'Bard (Gemini Pro)',\n",
|
1141 |
+
" 'Release Date': ''},\n",
|
1142 |
+
" {'key': 'llama-3-70b-instruct',\n",
|
1143 |
+
" 'Model': 'Llama-3-70b-Instruct',\n",
|
1144 |
+
" 'Release Date': ''},\n",
|
1145 |
+
" {'key': 'claude-3-sonnet-20240229',\n",
|
1146 |
+
" 'Model': 'Claude 3 Sonnet',\n",
|
1147 |
+
" 'Release Date': ''},\n",
|
1148 |
+
" {'key': 'command-r-plus', 'Model': 'Command R+', 'Release Date': ''},\n",
|
1149 |
+
" {'key': 'gpt-4-0314', 'Model': 'GPT-4-0314', 'Release Date': ''},\n",
|
1150 |
+
" {'key': 'claude-3-haiku-20240307',\n",
|
1151 |
+
" 'Model': 'Claude 3 Haiku',\n",
|
1152 |
+
" 'Release Date': ''},\n",
|
1153 |
+
" {'key': 'gpt-4-0613', 'Model': 'GPT-4-0613', 'Release Date': ''},\n",
|
1154 |
+
" {'key': 'mistral-large-2402',\n",
|
1155 |
+
" 'Model': 'Mistral-Large-2402',\n",
|
1156 |
+
" 'Release Date': ''},\n",
|
1157 |
+
" {'key': 'qwen1.5-72b-chat', 'Model': 'Qwen1.5-72B-Chat', 'Release Date': ''},\n",
|
1158 |
+
" {'key': 'reka-flash-21b-20240226-online',\n",
|
1159 |
+
" 'Model': 'Reka-Flash-21B-online',\n",
|
1160 |
+
" 'Release Date': ''},\n",
|
1161 |
+
" {'key': 'claude-1', 'Model': 'Claude-1', 'Release Date': ''},\n",
|
1162 |
+
" {'key': 'reka-flash-21b-20240226',\n",
|
1163 |
+
" 'Model': 'Reka-Flash-21B',\n",
|
1164 |
+
" 'Release Date': ''},\n",
|
1165 |
+
" {'key': 'command-r', 'Model': 'Command R', 'Release Date': ''},\n",
|
1166 |
+
" {'key': 'mistral-medium', 'Model': 'Mistral Medium', 'Release Date': ''},\n",
|
1167 |
+
" {'key': 'mixtral-8x22b-instruct-v0.1',\n",
|
1168 |
+
" 'Model': 'Mixtral-8x22b-Instruct-v0.1',\n",
|
1169 |
+
" 'Release Date': ''},\n",
|
1170 |
+
" {'key': 'llama-3-8b-instruct',\n",
|
1171 |
+
" 'Model': 'Llama-3-8b-Instruct',\n",
|
1172 |
+
" 'Release Date': ''},\n",
|
1173 |
+
" {'key': 'gemini-pro-dev-api',\n",
|
1174 |
+
" 'Model': 'Gemini Pro (Dev API)',\n",
|
1175 |
+
" 'Release Date': ''},\n",
|
1176 |
+
" {'key': 'qwen1.5-32b-chat', 'Model': 'Qwen1.5-32B-Chat', 'Release Date': ''},\n",
|
1177 |
+
" {'key': 'claude-2.0', 'Model': 'Claude-2.0', 'Release Date': ''},\n",
|
1178 |
+
" {'key': 'mistral-next', 'Model': 'Mistral-Next', 'Release Date': ''},\n",
|
1179 |
+
" {'key': 'zephyr-orpo-141b-A35b-v0.1',\n",
|
1180 |
+
" 'Model': 'Zephyr-ORPO-141b-A35b-v0.1',\n",
|
1181 |
+
" 'Release Date': ''},\n",
|
1182 |
+
" {'key': 'gpt-3.5-turbo-0613',\n",
|
1183 |
+
" 'Model': 'GPT-3.5-Turbo-0613',\n",
|
1184 |
+
" 'Release Date': ''},\n",
|
1185 |
+
" {'key': 'claude-2.1', 'Model': 'Claude-2.1', 'Release Date': ''},\n",
|
1186 |
+
" {'key': 'qwen1.5-14b-chat', 'Model': 'Qwen1.5-14B-Chat', 'Release Date': ''},\n",
|
1187 |
+
" {'key': 'starling-lm-7b-beta',\n",
|
1188 |
+
" 'Model': 'Starling-LM-7B-beta',\n",
|
1189 |
+
" 'Release Date': ''},\n",
|
1190 |
+
" {'key': 'gemini-pro', 'Model': 'Gemini Pro', 'Release Date': ''},\n",
|
1191 |
+
" {'key': 'mixtral-8x7b-instruct-v0.1',\n",
|
1192 |
+
" 'Model': 'Mixtral-8x7b-Instruct-v0.1',\n",
|
1193 |
+
" 'Release Date': ''},\n",
|
1194 |
+
" {'key': 'claude-instant-1', 'Model': 'Claude-Instant-1', 'Release Date': ''},\n",
|
1195 |
+
" {'key': 'yi-34b-chat', 'Model': 'Yi-34B-Chat', 'Release Date': ''},\n",
|
1196 |
+
" {'key': 'gpt-3.5-turbo-0314',\n",
|
1197 |
+
" 'Model': 'GPT-3.5-Turbo-0314',\n",
|
1198 |
+
" 'Release Date': ''},\n",
|
1199 |
+
" {'key': 'wizardlm-70b', 'Model': 'WizardLM-70B-v1.0', 'Release Date': ''},\n",
|
1200 |
+
" {'key': 'gpt-3.5-turbo-0125',\n",
|
1201 |
+
" 'Model': 'GPT-3.5-Turbo-0125',\n",
|
1202 |
+
" 'Release Date': ''},\n",
|
1203 |
+
" {'key': 'tulu-2-dpo-70b', 'Model': 'Tulu-2-DPO-70B', 'Release Date': ''},\n",
|
1204 |
+
" {'key': 'dbrx-instruct-preview',\n",
|
1205 |
+
" 'Model': 'DBRX-Instruct-Preview',\n",
|
1206 |
+
" 'Release Date': ''},\n",
|
1207 |
+
" {'key': 'openchat-3.5-0106',\n",
|
1208 |
+
" 'Model': 'OpenChat-3.5-0106',\n",
|
1209 |
+
" 'Release Date': ''},\n",
|
1210 |
+
" {'key': 'vicuna-33b', 'Model': 'Vicuna-33B', 'Release Date': ''},\n",
|
1211 |
+
" {'key': 'starling-lm-7b-alpha',\n",
|
1212 |
+
" 'Model': 'Starling-LM-7B-alpha',\n",
|
1213 |
+
" 'Release Date': ''},\n",
|
1214 |
+
" {'key': 'llama-2-70b-chat', 'Model': 'Llama-2-70b-chat', 'Release Date': ''},\n",
|
1215 |
+
" {'key': 'nous-hermes-2-mixtral-8x7b-dpo',\n",
|
1216 |
+
" 'Model': 'Nous-Hermes-2-Mixtral-8x7B-DPO',\n",
|
1217 |
+
" 'Release Date': ''},\n",
|
1218 |
+
" {'key': 'gemma-1.1-7b-it', 'Model': 'Gemma-1.1-7B-it', 'Release Date': ''},\n",
|
1219 |
+
" {'key': 'llama2-70b-steerlm-chat',\n",
|
1220 |
+
" 'Model': 'NV-Llama2-70B-SteerLM-Chat',\n",
|
1221 |
+
" 'Release Date': ''},\n",
|
1222 |
+
" {'key': 'deepseek-llm-67b-chat',\n",
|
1223 |
+
" 'Model': 'DeepSeek-LLM-67B-Chat',\n",
|
1224 |
+
" 'Release Date': ''},\n",
|
1225 |
+
" {'key': 'openhermes-2.5-mistral-7b',\n",
|
1226 |
+
" 'Model': 'OpenHermes-2.5-Mistral-7b',\n",
|
1227 |
+
" 'Release Date': ''},\n",
|
1228 |
+
" {'key': 'openchat-3.5', 'Model': 'OpenChat-3.5', 'Release Date': ''},\n",
|
1229 |
+
" {'key': 'pplx-70b-online', 'Model': 'pplx-70b-online', 'Release Date': ''},\n",
|
1230 |
+
" {'key': 'mistral-7b-instruct-v0.2',\n",
|
1231 |
+
" 'Model': 'Mistral-7B-Instruct-v0.2',\n",
|
1232 |
+
" 'Release Date': ''},\n",
|
1233 |
+
" {'key': 'qwen1.5-7b-chat', 'Model': 'Qwen1.5-7B-Chat', 'Release Date': ''},\n",
|
1234 |
+
" {'key': 'gpt-3.5-turbo-1106',\n",
|
1235 |
+
" 'Model': 'GPT-3.5-Turbo-1106',\n",
|
1236 |
+
" 'Release Date': ''},\n",
|
1237 |
+
" {'key': 'dolphin-2.2.1-mistral-7b',\n",
|
1238 |
+
" 'Model': 'Dolphin-2.2.1-Mistral-7B',\n",
|
1239 |
+
" 'Release Date': ''},\n",
|
1240 |
+
" {'key': 'solar-10.7b-instruct-v1.0',\n",
|
1241 |
+
" 'Model': 'SOLAR-10.7B-Instruct-v1.0',\n",
|
1242 |
+
" 'Release Date': ''},\n",
|
1243 |
+
" {'key': 'phi-3-mini-128k-instruct',\n",
|
1244 |
+
" 'Model': 'Phi-3-Mini-128k-Instruct',\n",
|
1245 |
+
" 'Release Date': ''},\n",
|
1246 |
+
" {'key': 'wizardlm-13b', 'Model': 'WizardLM-13b-v1.2', 'Release Date': ''},\n",
|
1247 |
+
" {'key': 'llama-2-13b-chat', 'Model': 'Llama-2-13b-chat', 'Release Date': ''},\n",
|
1248 |
+
" {'key': 'zephyr-7b-beta', 'Model': 'Zephyr-7b-beta', 'Release Date': ''},\n",
|
1249 |
+
" {'key': 'codellama-70b-instruct',\n",
|
1250 |
+
" 'Model': 'CodeLlama-70B-instruct',\n",
|
1251 |
+
" 'Release Date': ''},\n",
|
1252 |
+
" {'key': 'mpt-30b-chat', 'Model': 'MPT-30B-chat', 'Release Date': ''},\n",
|
1253 |
+
" {'key': 'vicuna-13b', 'Model': 'Vicuna-13B', 'Release Date': ''},\n",
|
1254 |
+
" {'key': 'codellama-34b-instruct',\n",
|
1255 |
+
" 'Model': 'CodeLlama-34B-instruct',\n",
|
1256 |
+
" 'Release Date': ''},\n",
|
1257 |
+
" {'key': 'gemma-7b-it', 'Model': 'Gemma-7B-it', 'Release Date': ''},\n",
|
1258 |
+
" {'key': 'pplx-7b-online', 'Model': 'pplx-7b-online', 'Release Date': ''},\n",
|
1259 |
+
" {'key': 'zephyr-7b-alpha', 'Model': 'Zephyr-7b-alpha', 'Release Date': ''},\n",
|
1260 |
+
" {'key': 'llama-2-7b-chat', 'Model': 'Llama-2-7b-chat', 'Release Date': ''},\n",
|
1261 |
+
" {'key': 'qwen-14b-chat', 'Model': 'Qwen-14B-Chat', 'Release Date': ''},\n",
|
1262 |
+
" {'key': 'falcon-180b-chat', 'Model': 'falcon-180b-chat', 'Release Date': ''},\n",
|
1263 |
+
" {'key': 'guanaco-33b', 'Model': 'Guanaco-33B', 'Release Date': ''},\n",
|
1264 |
+
" {'key': 'stripedhyena-nous-7b',\n",
|
1265 |
+
" 'Model': 'StripedHyena-Nous-7B',\n",
|
1266 |
+
" 'Release Date': ''},\n",
|
1267 |
+
" {'key': 'olmo-7b-instruct', 'Model': 'OLMo-7B-instruct', 'Release Date': ''},\n",
|
1268 |
+
" {'key': 'gemma-1.1-2b-it', 'Model': 'Gemma-1.1-2B-it', 'Release Date': ''},\n",
|
1269 |
+
" {'key': 'mistral-7b-instruct',\n",
|
1270 |
+
" 'Model': 'Mistral-7B-Instruct-v0.1',\n",
|
1271 |
+
" 'Release Date': ''},\n",
|
1272 |
+
" {'key': 'palm-2', 'Model': 'PaLM-Chat-Bison-001', 'Release Date': ''},\n",
|
1273 |
+
" {'key': 'vicuna-7b', 'Model': 'Vicuna-7B', 'Release Date': ''},\n",
|
1274 |
+
" {'key': 'qwen1.5-4b-chat', 'Model': 'Qwen1.5-4B-Chat', 'Release Date': ''},\n",
|
1275 |
+
" {'key': 'gemma-2b-it', 'Model': 'Gemma-2B-it', 'Release Date': ''},\n",
|
1276 |
+
" {'key': 'koala-13b', 'Model': 'Koala-13B', 'Release Date': ''},\n",
|
1277 |
+
" {'key': 'chatglm3-6b', 'Model': 'ChatGLM3-6B', 'Release Date': ''},\n",
|
1278 |
+
" {'key': 'gpt4all-13b-snoozy',\n",
|
1279 |
+
" 'Model': 'GPT4All-13B-Snoozy',\n",
|
1280 |
+
" 'Release Date': ''},\n",
|
1281 |
+
" {'key': 'chatglm2-6b', 'Model': 'ChatGLM2-6B', 'Release Date': ''},\n",
|
1282 |
+
" {'key': 'mpt-7b-chat', 'Model': 'MPT-7B-Chat', 'Release Date': ''},\n",
|
1283 |
+
" {'key': 'RWKV-4-Raven-14B', 'Model': 'RWKV-4-Raven-14B', 'Release Date': ''},\n",
|
1284 |
+
" {'key': 'alpaca-13b', 'Model': 'Alpaca-13B', 'Release Date': ''},\n",
|
1285 |
+
" {'key': 'oasst-pythia-12b',\n",
|
1286 |
+
" 'Model': 'OpenAssistant-Pythia-12B',\n",
|
1287 |
+
" 'Release Date': ''},\n",
|
1288 |
+
" {'key': 'chatglm-6b', 'Model': 'ChatGLM-6B', 'Release Date': ''},\n",
|
1289 |
+
" {'key': 'fastchat-t5-3b', 'Model': 'FastChat-T5-3B', 'Release Date': ''},\n",
|
1290 |
+
" {'key': 'stablelm-tuned-alpha-7b',\n",
|
1291 |
+
" 'Model': 'StableLM-Tuned-Alpha-7B',\n",
|
1292 |
+
" 'Release Date': ''},\n",
|
1293 |
+
" {'key': 'dolly-v2-12b', 'Model': 'Dolly-V2-12B', 'Release Date': ''},\n",
|
1294 |
+
" {'key': 'llama-13b', 'Model': 'LLaMA-13B', 'Release Date': ''}]"
|
1295 |
+
]
|
1296 |
+
},
|
1297 |
+
"execution_count": 119,
|
1298 |
+
"metadata": {},
|
1299 |
+
"output_type": "execute_result"
|
1300 |
+
}
|
1301 |
+
],
|
1302 |
+
"source": [
|
1303 |
+
"t.to_dict(orient=\"records\")"
|
1304 |
+
]
|
1305 |
+
},
|
1306 |
+
{
|
1307 |
+
"cell_type": "markdown",
|
1308 |
+
"metadata": {},
|
1309 |
+
"source": [
|
1310 |
+
"## Build plot"
|
1311 |
+
]
|
1312 |
+
},
|
1313 |
+
{
|
1314 |
+
"cell_type": "markdown",
|
1315 |
+
"metadata": {},
|
1316 |
+
"source": []
|
1317 |
+
}
|
1318 |
+
],
|
1319 |
+
"metadata": {
|
1320 |
+
"kernelspec": {
|
1321 |
+
"display_name": ".venv",
|
1322 |
+
"language": "python",
|
1323 |
+
"name": "python3"
|
1324 |
+
},
|
1325 |
+
"language_info": {
|
1326 |
+
"codemirror_mode": {
|
1327 |
+
"name": "ipython",
|
1328 |
+
"version": 3
|
1329 |
+
},
|
1330 |
+
"file_extension": ".py",
|
1331 |
+
"mimetype": "text/x-python",
|
1332 |
+
"name": "python",
|
1333 |
+
"nbconvert_exporter": "python",
|
1334 |
+
"pygments_lexer": "ipython3",
|
1335 |
+
"version": "3.10.11"
|
1336 |
+
}
|
1337 |
+
},
|
1338 |
+
"nbformat": 4,
|
1339 |
+
"nbformat_minor": 2
|
1340 |
+
}
|
release_date_mapping.json
ADDED
@@ -0,0 +1,457 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
[
|
2 |
+
{
|
3 |
+
"key": "gpt-4-turbo-2024-04-09",
|
4 |
+
"Model": "GPT-4-Turbo-2024-04-09",
|
5 |
+
"Release Date": "2024-04-09"
|
6 |
+
},
|
7 |
+
{
|
8 |
+
"key": "gpt-4-1106-preview",
|
9 |
+
"Model": "GPT-4-1106-preview",
|
10 |
+
"Release Date": "2024-11-06"
|
11 |
+
},
|
12 |
+
{
|
13 |
+
"key": "claude-3-opus-20240229",
|
14 |
+
"Model": "Claude 3 Opus",
|
15 |
+
"Release Date": "2024-02-29"
|
16 |
+
},
|
17 |
+
{
|
18 |
+
"key": "gemini-1.5-pro-api-0409-preview",
|
19 |
+
"Model": "Gemini 1.5 Pro API-0409-Preview",
|
20 |
+
"Release Date": "2024-04-09"
|
21 |
+
},
|
22 |
+
{
|
23 |
+
"key": "gpt-4-0125-preview",
|
24 |
+
"Model": "GPT-4-0125-preview",
|
25 |
+
"Release Date": "2024-01-25"
|
26 |
+
},
|
27 |
+
{
|
28 |
+
"key": "bard-jan-24-gemini-pro",
|
29 |
+
"Model": "Bard (Gemini Pro)",
|
30 |
+
"Release Date": "2024-02-01"
|
31 |
+
},
|
32 |
+
{
|
33 |
+
"key": "llama-3-70b-instruct",
|
34 |
+
"Model": "Llama-3-70b-Instruct",
|
35 |
+
"Release Date": "2024-04-18"
|
36 |
+
},
|
37 |
+
{
|
38 |
+
"key": "claude-3-sonnet-20240229",
|
39 |
+
"Model": "Claude 3 Sonnet",
|
40 |
+
"Release Date": "2024-02-29"
|
41 |
+
},
|
42 |
+
{
|
43 |
+
"key": "command-r-plus",
|
44 |
+
"Model": "Command R+",
|
45 |
+
"Release Date": "2024-04-04"
|
46 |
+
},
|
47 |
+
{
|
48 |
+
"key": "gpt-4-0314",
|
49 |
+
"Model": "GPT-4-0314",
|
50 |
+
"Release Date": "2024-03-14"
|
51 |
+
},
|
52 |
+
{
|
53 |
+
"key": "claude-3-haiku-20240307",
|
54 |
+
"Model": "Claude 3 Haiku",
|
55 |
+
"Release Date": "2024-03-07"
|
56 |
+
},
|
57 |
+
{
|
58 |
+
"key": "gpt-4-0613",
|
59 |
+
"Model": "GPT-4-0613",
|
60 |
+
"Release Date": "2023-06-13"
|
61 |
+
},
|
62 |
+
{
|
63 |
+
"key": "mistral-large-2402",
|
64 |
+
"Model": "Mistral-Large-2402",
|
65 |
+
"Release Date": "2024-02-24"
|
66 |
+
},
|
67 |
+
{
|
68 |
+
"key": "qwen1.5-72b-chat",
|
69 |
+
"Model": "Qwen1.5-72B-Chat",
|
70 |
+
"Release Date": "2024-02-04"
|
71 |
+
},
|
72 |
+
{
|
73 |
+
"key": "reka-flash-21b-20240226-online",
|
74 |
+
"Model": "Reka-Flash-21B-online",
|
75 |
+
"Release Date": "2024-02-26"
|
76 |
+
},
|
77 |
+
{
|
78 |
+
"key": "claude-1",
|
79 |
+
"Model": "Claude-1",
|
80 |
+
"Release Date": "2023-03-14"
|
81 |
+
},
|
82 |
+
{
|
83 |
+
"key": "reka-flash-21b-20240226",
|
84 |
+
"Model": "Reka-Flash-21B",
|
85 |
+
"Release Date": "2024-02-26"
|
86 |
+
},
|
87 |
+
{
|
88 |
+
"key": "command-r",
|
89 |
+
"Model": "Command R",
|
90 |
+
"Release Date": "2024-03-11"
|
91 |
+
},
|
92 |
+
{
|
93 |
+
"key": "mistral-medium",
|
94 |
+
"Model": "Mistral Medium",
|
95 |
+
"Release Date": "2023-12-11"
|
96 |
+
},
|
97 |
+
{
|
98 |
+
"key": "mixtral-8x22b-instruct-v0.1",
|
99 |
+
"Model": "Mixtral-8x22b-Instruct-v0.1",
|
100 |
+
"Release Date": "2024-04-17"
|
101 |
+
},
|
102 |
+
{
|
103 |
+
"key": "llama-3-8b-instruct",
|
104 |
+
"Model": "Llama-3-8b-Instruct",
|
105 |
+
"Release Date": "2024-04-18"
|
106 |
+
},
|
107 |
+
{
|
108 |
+
"key": "gemini-pro-dev-api",
|
109 |
+
"Model": "Gemini Pro (Dev API)",
|
110 |
+
"Release Date": "2023-12-13"
|
111 |
+
},
|
112 |
+
{
|
113 |
+
"key": "qwen1.5-32b-chat",
|
114 |
+
"Model": "Qwen1.5-32B-Chat",
|
115 |
+
"Release Date": "2024-02-04"
|
116 |
+
},
|
117 |
+
{
|
118 |
+
"key": "claude-2.0",
|
119 |
+
"Model": "Claude-2.0",
|
120 |
+
"Release Date": "2023-07-11"
|
121 |
+
},
|
122 |
+
{
|
123 |
+
"key": "mistral-next",
|
124 |
+
"Model": "Mistral-Next",
|
125 |
+
"Release Date": "2024-02-17"
|
126 |
+
},
|
127 |
+
{
|
128 |
+
"key": "zephyr-orpo-141b-A35b-v0.1",
|
129 |
+
"Model": "Zephyr-ORPO-141b-A35b-v0.1",
|
130 |
+
"Release Date": "2024-04-12"
|
131 |
+
},
|
132 |
+
{
|
133 |
+
"key": "gpt-3.5-turbo-0613",
|
134 |
+
"Model": "GPT-3.5-Turbo-0613",
|
135 |
+
"Release Date": "2023-06-13"
|
136 |
+
},
|
137 |
+
{
|
138 |
+
"key": "claude-2.1",
|
139 |
+
"Model": "Claude-2.1",
|
140 |
+
"Release Date": "2023-11-21"
|
141 |
+
},
|
142 |
+
{
|
143 |
+
"key": "qwen1.5-14b-chat",
|
144 |
+
"Model": "Qwen1.5-14B-Chat",
|
145 |
+
"Release Date": "2024-02-04"
|
146 |
+
},
|
147 |
+
{
|
148 |
+
"key": "starling-lm-7b-beta",
|
149 |
+
"Model": "Starling-LM-7B-beta",
|
150 |
+
"Release Date": "2024-03-20"
|
151 |
+
},
|
152 |
+
{
|
153 |
+
"key": "gemini-pro",
|
154 |
+
"Model": "Gemini Pro",
|
155 |
+
"Release Date": "2023-12-13"
|
156 |
+
},
|
157 |
+
{
|
158 |
+
"key": "mixtral-8x7b-instruct-v0.1",
|
159 |
+
"Model": "Mixtral-8x7b-Instruct-v0.1",
|
160 |
+
"Release Date": "2023-12-11"
|
161 |
+
},
|
162 |
+
{
|
163 |
+
"key": "claude-instant-1",
|
164 |
+
"Model": "Claude-Instant-1",
|
165 |
+
"Release Date": "2023-03-14"
|
166 |
+
},
|
167 |
+
{
|
168 |
+
"key": "yi-34b-chat",
|
169 |
+
"Model": "Yi-34B-Chat",
|
170 |
+
"Release Date": "2024-01-23"
|
171 |
+
},
|
172 |
+
{
|
173 |
+
"key": "gpt-3.5-turbo-0314",
|
174 |
+
"Model": "GPT-3.5-Turbo-0314",
|
175 |
+
"Release Date": "2024-03-14"
|
176 |
+
},
|
177 |
+
{
|
178 |
+
"key": "wizardlm-70b",
|
179 |
+
"Model": "WizardLM-70B-v1.0",
|
180 |
+
"Release Date": "2023-08-09"
|
181 |
+
},
|
182 |
+
{
|
183 |
+
"key": "gpt-3.5-turbo-0125",
|
184 |
+
"Model": "GPT-3.5-Turbo-0125",
|
185 |
+
"Release Date": "2024-01-25"
|
186 |
+
},
|
187 |
+
{
|
188 |
+
"key": "tulu-2-dpo-70b",
|
189 |
+
"Model": "Tulu-2-DPO-70B",
|
190 |
+
"Release Date": "2023-11-12"
|
191 |
+
},
|
192 |
+
{
|
193 |
+
"key": "dbrx-instruct-preview",
|
194 |
+
"Model": "DBRX-Instruct-Preview",
|
195 |
+
"Release Date": "2024-03-27"
|
196 |
+
},
|
197 |
+
{
|
198 |
+
"key": "openchat-3.5-0106",
|
199 |
+
"Model": "OpenChat-3.5-0106",
|
200 |
+
"Release Date": "2024-01-06"
|
201 |
+
},
|
202 |
+
{
|
203 |
+
"key": "vicuna-33b",
|
204 |
+
"Model": "Vicuna-33B",
|
205 |
+
"Release Date": "2023-06-21"
|
206 |
+
},
|
207 |
+
{
|
208 |
+
"key": "starling-lm-7b-alpha",
|
209 |
+
"Model": "Starling-LM-7B-alpha",
|
210 |
+
"Release Date": "2023-11-25"
|
211 |
+
},
|
212 |
+
{
|
213 |
+
"key": "llama-2-70b-chat",
|
214 |
+
"Model": "Llama-2-70b-chat",
|
215 |
+
"Release Date": "2023-07-18"
|
216 |
+
},
|
217 |
+
{
|
218 |
+
"key": "nous-hermes-2-mixtral-8x7b-dpo",
|
219 |
+
"Model": "Nous-Hermes-2-Mixtral-8x7B-DPO",
|
220 |
+
"Release Date": "2024-01-13"
|
221 |
+
},
|
222 |
+
{
|
223 |
+
"key": "gemma-1.1-7b-it",
|
224 |
+
"Model": "Gemma-1.1-7B-it",
|
225 |
+
"Release Date": "2024-04-09"
|
226 |
+
},
|
227 |
+
{
|
228 |
+
"key": "llama2-70b-steerlm-chat",
|
229 |
+
"Model": "NV-Llama2-70B-SteerLM-Chat",
|
230 |
+
"Release Date": "2023-11-24"
|
231 |
+
},
|
232 |
+
{
|
233 |
+
"key": "deepseek-llm-67b-chat",
|
234 |
+
"Model": "DeepSeek-LLM-67B-Chat",
|
235 |
+
"Release Date": "2023-11-29"
|
236 |
+
},
|
237 |
+
{
|
238 |
+
"key": "openhermes-2.5-mistral-7b",
|
239 |
+
"Model": "OpenHermes-2.5-Mistral-7b",
|
240 |
+
"Release Date": "2023-10-29"
|
241 |
+
},
|
242 |
+
{
|
243 |
+
"key": "openchat-3.5",
|
244 |
+
"Model": "OpenChat-3.5",
|
245 |
+
"Release Date": "2023-11-16"
|
246 |
+
},
|
247 |
+
{
|
248 |
+
"key": "pplx-70b-online",
|
249 |
+
"Model": "pplx-70b-online",
|
250 |
+
"Release Date": "2023-11-29"
|
251 |
+
},
|
252 |
+
{
|
253 |
+
"key": "mistral-7b-instruct-v0.2",
|
254 |
+
"Model": "Mistral-7B-Instruct-v0.2",
|
255 |
+
"Release Date": "2023-12-11"
|
256 |
+
},
|
257 |
+
{
|
258 |
+
"key": "qwen1.5-7b-chat",
|
259 |
+
"Model": "Qwen1.5-7B-Chat",
|
260 |
+
"Release Date": "2024-02-04"
|
261 |
+
},
|
262 |
+
{
|
263 |
+
"key": "gpt-3.5-turbo-1106",
|
264 |
+
"Model": "GPT-3.5-Turbo-1106",
|
265 |
+
"Release Date": "2023-11-06"
|
266 |
+
},
|
267 |
+
{
|
268 |
+
"key": "dolphin-2.2.1-mistral-7b",
|
269 |
+
"Model": "Dolphin-2.2.1-Mistral-7B",
|
270 |
+
"Release Date": "2023-10-30"
|
271 |
+
},
|
272 |
+
{
|
273 |
+
"key": "solar-10.7b-instruct-v1.0",
|
274 |
+
"Model": "SOLAR-10.7B-Instruct-v1.0",
|
275 |
+
"Release Date": "2023-12-13"
|
276 |
+
},
|
277 |
+
{
|
278 |
+
"key": "phi-3-mini-128k-instruct",
|
279 |
+
"Model": "Phi-3-Mini-128k-Instruct",
|
280 |
+
"Release Date": "2024-04-23"
|
281 |
+
},
|
282 |
+
{
|
283 |
+
"key": "wizardlm-13b",
|
284 |
+
"Model": "WizardLM-13b-v1.2",
|
285 |
+
"Release Date": "2023-07-25"
|
286 |
+
},
|
287 |
+
{
|
288 |
+
"key": "llama-2-13b-chat",
|
289 |
+
"Model": "Llama-2-13b-chat",
|
290 |
+
"Release Date": "2023-07-18"
|
291 |
+
},
|
292 |
+
{
|
293 |
+
"key": "zephyr-7b-beta",
|
294 |
+
"Model": "Zephyr-7b-beta",
|
295 |
+
"Release Date": "2023-10-26"
|
296 |
+
},
|
297 |
+
{
|
298 |
+
"key": "codellama-70b-instruct",
|
299 |
+
"Model": "CodeLlama-70B-instruct",
|
300 |
+
"Release Date": "2024-01-29"
|
301 |
+
},
|
302 |
+
{
|
303 |
+
"key": "mpt-30b-chat",
|
304 |
+
"Model": "MPT-30B-chat",
|
305 |
+
"Release Date": "2023-06-09"
|
306 |
+
},
|
307 |
+
{
|
308 |
+
"key": "vicuna-13b",
|
309 |
+
"Model": "Vicuna-13B",
|
310 |
+
"Release Date": "2023-07-23"
|
311 |
+
},
|
312 |
+
{
|
313 |
+
"key": "codellama-34b-instruct",
|
314 |
+
"Model": "CodeLlama-34B-instruct",
|
315 |
+
"Release Date": "2023-08-24"
|
316 |
+
},
|
317 |
+
{
|
318 |
+
"key": "gemma-7b-it",
|
319 |
+
"Model": "Gemma-7B-it",
|
320 |
+
"Release Date": "2024-02-21"
|
321 |
+
},
|
322 |
+
{
|
323 |
+
"key": "pplx-7b-online",
|
324 |
+
"Model": "pplx-7b-online",
|
325 |
+
"Release Date": "2023-11-29"
|
326 |
+
},
|
327 |
+
{
|
328 |
+
"key": "zephyr-7b-alpha",
|
329 |
+
"Model": "Zephyr-7b-alpha",
|
330 |
+
"Release Date": "2023-10-09"
|
331 |
+
},
|
332 |
+
{
|
333 |
+
"key": "llama-2-7b-chat",
|
334 |
+
"Model": "Llama-2-7b-chat",
|
335 |
+
"Release Date": "2023-07-18"
|
336 |
+
},
|
337 |
+
{
|
338 |
+
"key": "qwen-14b-chat",
|
339 |
+
"Model": "Qwen-14B-Chat",
|
340 |
+
"Release Date": "2023-09-24"
|
341 |
+
},
|
342 |
+
{
|
343 |
+
"key": "falcon-180b-chat",
|
344 |
+
"Model": "falcon-180b-chat",
|
345 |
+
"Release Date": "2023-09-05"
|
346 |
+
},
|
347 |
+
{
|
348 |
+
"key": "guanaco-33b",
|
349 |
+
"Model": "Guanaco-33B",
|
350 |
+
"Release Date": "2023-05-22"
|
351 |
+
},
|
352 |
+
{
|
353 |
+
"key": "stripedhyena-nous-7b",
|
354 |
+
"Model": "StripedHyena-Nous-7B",
|
355 |
+
"Release Date": "2023-12-07"
|
356 |
+
},
|
357 |
+
{
|
358 |
+
"key": "olmo-7b-instruct",
|
359 |
+
"Model": "OLMo-7B-instruct",
|
360 |
+
"Release Date": "2024-02-23"
|
361 |
+
},
|
362 |
+
{
|
363 |
+
"key": "gemma-1.1-2b-it",
|
364 |
+
"Model": "Gemma-1.1-2B-it",
|
365 |
+
"Release Date": "2024-04-09"
|
366 |
+
},
|
367 |
+
{
|
368 |
+
"key": "mistral-7b-instruct",
|
369 |
+
"Model": "Mistral-7B-Instruct-v0.1",
|
370 |
+
"Release Date": "2023-09-27"
|
371 |
+
},
|
372 |
+
{
|
373 |
+
"key": "palm-2",
|
374 |
+
"Model": "PaLM-Chat-Bison-001",
|
375 |
+
"Release Date": "2023-07-10"
|
376 |
+
},
|
377 |
+
{
|
378 |
+
"key": "vicuna-7b",
|
379 |
+
"Model": "Vicuna-7B",
|
380 |
+
"Release Date": "2023-07-29"
|
381 |
+
},
|
382 |
+
{
|
383 |
+
"key": "qwen1.5-4b-chat",
|
384 |
+
"Model": "Qwen1.5-4B-Chat",
|
385 |
+
"Release Date": "2024-02-04"
|
386 |
+
},
|
387 |
+
{
|
388 |
+
"key": "gemma-2b-it",
|
389 |
+
"Model": "Gemma-2B-it",
|
390 |
+
"Release Date": "2024-02-21"
|
391 |
+
},
|
392 |
+
{
|
393 |
+
"key": "koala-13b",
|
394 |
+
"Model": "Koala-13B",
|
395 |
+
"Release Date": "2023-04-03"
|
396 |
+
},
|
397 |
+
{
|
398 |
+
"key": "chatglm3-6b",
|
399 |
+
"Model": "ChatGLM3-6B",
|
400 |
+
"Release Date": "2023-10-25"
|
401 |
+
},
|
402 |
+
{
|
403 |
+
"key": "gpt4all-13b-snoozy",
|
404 |
+
"Model": "GPT4All-13B-Snoozy",
|
405 |
+
"Release Date": "2023-04-24"
|
406 |
+
},
|
407 |
+
{
|
408 |
+
"key": "chatglm2-6b",
|
409 |
+
"Model": "ChatGLM2-6B",
|
410 |
+
"Release Date": "2023-06-25"
|
411 |
+
},
|
412 |
+
{
|
413 |
+
"key": "mpt-7b-chat",
|
414 |
+
"Model": "MPT-7B-Chat",
|
415 |
+
"Release Date": "2023-05-04"
|
416 |
+
},
|
417 |
+
{
|
418 |
+
"key": "RWKV-4-Raven-14B",
|
419 |
+
"Model": "RWKV-4-Raven-14B",
|
420 |
+
"Release Date": "2023-05-22"
|
421 |
+
},
|
422 |
+
{
|
423 |
+
"key": "alpaca-13b",
|
424 |
+
"Model": "Alpaca-13B",
|
425 |
+
"Release Date": "2023-03-13"
|
426 |
+
},
|
427 |
+
{
|
428 |
+
"key": "oasst-pythia-12b",
|
429 |
+
"Model": "OpenAssistant-Pythia-12B",
|
430 |
+
"Release Date": "2023-04-03"
|
431 |
+
},
|
432 |
+
{
|
433 |
+
"key": "chatglm-6b",
|
434 |
+
"Model": "ChatGLM-6B",
|
435 |
+
"Release Date": "2023-03-13"
|
436 |
+
},
|
437 |
+
{
|
438 |
+
"key": "fastchat-t5-3b",
|
439 |
+
"Model": "FastChat-T5-3B",
|
440 |
+
"Release Date": "2023-04-27"
|
441 |
+
},
|
442 |
+
{
|
443 |
+
"key": "stablelm-tuned-alpha-7b",
|
444 |
+
"Model": "StableLM-Tuned-Alpha-7B",
|
445 |
+
"Release Date": "2023-04-19"
|
446 |
+
},
|
447 |
+
{
|
448 |
+
"key": "dolly-v2-12b",
|
449 |
+
"Model": "Dolly-V2-12B",
|
450 |
+
"Release Date": "2023-04-12"
|
451 |
+
},
|
452 |
+
{
|
453 |
+
"key": "llama-13b",
|
454 |
+
"Model": "LLaMA-13B",
|
455 |
+
"Release Date": "2023-02-27"
|
456 |
+
}
|
457 |
+
]
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
huggingface_hub
|
2 |
+
pandas
|
3 |
+
plotly
|
4 |
+
gradio
|