-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathviewer.html
199 lines (185 loc) · 47.5 KB
/
viewer.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8" />
<title>BALROG Model Viewer</title>
<meta name="description" content="BALROG: Benchmarking Agentic LLM/VLM Reasoning On Games - Model Viewer" />
<link rel="shortcut icon" href="img/logo.png" />
<link rel="icon" href="img/logo.png" />
<link rel="stylesheet" href="css/normalize.css" />
<link rel="stylesheet" href="css/fonts.css" />
<link rel="stylesheet" href="css/styles.css" />
<link rel="stylesheet" href="css/viewer.css" />
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.2.0/css/all.min.css"
integrity="..." crossorigin="anonymous" />
<script src="https://cdn.jsdelivr.net/npm/marked/marked.min.js"></script>
<!-- Include a sanitizer for security -->
<script src="https://cdn.jsdelivr.net/npm/dompurify@2.3.6/dist/purify.min.js"></script>
</head>
<body>
<div style="padding-bottom: 50px">
<!-- Top Section with BALROG Title and Image -->
<section style="background-color: var(--dark_accent_color)">
<div class="content-wrapper title-wrapper" style="flex-direction: column">
<div style="
display: flex;
flex-direction: row;
align-items: center;
padding-bottom: 15px;
">
<h1 style="font-size: 60px; padding-top: 0.4em">BALROG</h1>
<img src="img/logo.png" style="height: 120px; padding-top: 0em; padding-left: 0.5em" />
</div>
<h3>Benchmarking Agentic LLM/VLM Reasoning On Games</h3>
<!-- Model Name -->
<h2 id="modelName" style="font-size: 40px; padding-top: 0.4em"></h2>
<p id="modelDate">Date: </p>
<!-- Navigation Buttons -->
<div class="content-wrapper" style="margin-top: 2em">
<a href="index.html">
<button class="outline"
style="flex-direction: row; display: flex; justify-content: center; align-items: center;">
<img src="img/logo.png"
style="height: 1.8em; margin-left: -0.3em; margin-right: 0.3em; margin-bottom: 0em;" />
Home
</button>
</a>
<a href="https://arxiv.org/abs/2411.13543">
<button class="outline">
<i class="fa fa-paperclip"></i> Paper
</button>
</a>
<a href="https://github.com/balrog-ai/BALROG">
<button class="outline">
<i class="fab fa-github"></i> Code
</button>
</a>
<a href="submit.html">
<button class="outline">
<i class="fa fa-upload"></i> Submit
</button>
</a>
</div>
</div>
</section>
<!-- About Section -->
<section class="main-container">
<!-- Display README.md content -->
<div class="content-wrapper">
<div class="content-box" id="readme">
Loading README.md...
</div>
</div>
<!-- Trajectories Section (Always displayed) -->
<div class="content-wrapper" id="trajectoriesLinkContainer">
<div class="content-box">
<h2 class="text-title">Trajectories</h2>
<p class="text-content" id="trajectoriesText">
<!-- Content will be updated via JavaScript -->
</p>
</div>
</div>
<!-- Display Performance Data -->
<div class="content-wrapper">
<h3>Performance Data</h3>
<table class="table-results" id="performanceTable">
<thead>
<tr>
<th>Environment</th>
<th>Progress (%)</th>
<th>Standard Error</th>
<th>Number of Episodes</th>
</tr>
</thead>
<tbody>
<!-- Data will be inserted here -->
</tbody>
</table>
</div>
</section>
</div>
<!-- Embed the data variable -->
<script>
var data = {"leaderboards": [{"name": "LLM", "results": [{"average": [16.18, 1.64], "babaisai": [19.327731092436977, 3.6197542978761432, 119], "babyai": [34.0, 6.69925368977769, 50], "crafter": [27.272727272727277, 3.5790944881871867, 10], "date": "2024-11-25", "folder": "submissions/LLM/20241115_naive-Qwen2.5-72B-Instruct", "minihack": [5.0, 3.44, 40], "name": "Qwen2.5-72B-it", "nle": [0.30784538150722346, 0.2753452798438033, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [11.176470588235295, 3.835827381829256, 30], "trajs": false, "verified": true}, {"average": [32.64471266283788, 1.9342003853175014], "babaisai": [37.5, 4.419417382415922, 120], "babyai": [68.0, 6.596969000988254, 50], "crafter": [32.72727272727273, 3.2012394293925466, 10], "date": "2024-11-11", "folder": "submissions/LLM/20241103_Claude-3.5-Sonnet", "minihack": [15.0, 5.645794895318107, 40], "name": "Claude-3.5-Sonnet-2024-10-22", "nle": [0.5821797203427749, 0.5207173719233047, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [42.05882352941177, 5.406976071199611, 60], "trajs": true, "verified": true}, {"average": [14.633333333333333, 1.372992473880805], "babaisai": [12.8, 2.33, 200], "babyai": [50.0, 7.07, 50], "crafter": [20.0, 0.74, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240924_naive-Gemini-1.5-Flash", "minihack": [5.0, 3.45, 40], "name": "Gemini-1.5-Flash-002", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [0.0, 0.0, 60], "trajs": false, "verified": true}, {"average": [15.144385026737966, 1.5514263407931903], "babaisai": [18.333333333333332, 3.532258746447074, 120], "babyai": [36.0, 6.788225099390856, 50], "crafter": [25.454545454545453, 3.226952608963427, 10], "date": "2024-11-11", "folder": "submissions/LLM/20241101_naive-Llama-3.1-8B-Instruct", "minihack": [5.0, 3.446012188022554, 40], "name": "Llama-3.1-8B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [6.07843137254902, 2.409186144801279, 30], "trajs": false, "verified": true}, {"average": [7.797534165181225, 1.0978942439497628], "babaisai": [12.5, 3.0190368221228, 120], "babyai": [14.000000000000002, 4.907137658554117, 50], "crafter": [16.363636363636363, 3.0287874998104876, 10], "date": "2024-11-25", "folder": "submissions/LLM/20241115_naive-Qwen2.5-7B-it", "minihack": [0.0, 0.0, 40], "name": "Qwen-2.5-7B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [3.9215686274509802, 1.0125446656751418, 30], "trajs": true, "verified": true}, {"average": [11.642008318478906, 1.4408567366363194], "babaisai": [16.666666666666664, 3.4020690871988615, 120], "babyai": [32.0, 6.596969000988255, 50], "crafter": [13.636363636363635, 2.6504326794751365, 10], "date": "2025-01-13", "folder": "submissions/LLM/20250113_robust_naive_microsoft_phi-4", "minihack": [5.0, 3.4460121880225536, 40], "name": "Microsoft-Phi-4", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Robust Naive Baseline\nWe use naive zero-shot agents as the baseline for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate this robust_naive baselines, use `agent.type=robust_naive`:\n\n```\npython eval.py \\\n agent.type=robust_naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=claude \\\n client.model_id=claude-3-5-haiku-20241022\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/docs/evaluation.md)\n- [Paper](https://arxiv.org/abs/2411.13543)\n- [BALROG](https://balrogai.com)", "site": null, "textworld": [2.5490196078431375, 0.8628936295197067, 30], "trajs": true, "verified": true}, {"average": [3.665881504116798, 0.7993593693789691], "babaisai": [7.563025210084033, 2.4238020876994986, 119], "babyai": [4.0, 2.771281292110205, 50], "crafter": [6.363636363636363, 1.7248787237282064, 10], "date": "2024-11-25", "folder": "submissions/LLM/20241115_naive-Qwen2-VL-7B-instruct", "minihack": [2.5, 2.4685522072664368, 40], "name": "Qwen2-VL-7B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)", "site": null, "textworld": [1.5686274509803921, 0.6159068752622091, 30], "trajs": false, "verified": true}, {"average": [10.133689839572192, 1.280367136777229], "babaisai": [17.5, 3.4686092313779033, 120], "babyai": [20.0, 5.656854249492381, 50], "crafter": [17.27272727272727, 2.7872199485010714, 10], "date": "2024-11-11", "folder": "submissions/LLM/20241030_naive-Llama-3.2-3B-Instruct", "minihack": [2.5, 2.4685522072664368, 40], "name": "Llama-3.2-3B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [3.5294117647058822, 1.0595494969495132, 30], "trajs": true, "verified": true}, {"average": [23.03203437290035, 1.6588823778056299], "babaisai": [29.166666666666668, 4.149269235080457, 120], "babyai": [66.0, 6.699253689777692, 50], "crafter": [28.636363636363633, 4.0681183336844455, 10], "date": "2024-12-09", "folder": "submissions/LLM/20241209_naive_Llama-3.3-70B-Instruct", "minihack": [5.0, 3.446012188022554, 40], "name": "Llama-3.3-70B-it", "nle": [0.36956809123452017, 0.33055174972609247, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nWe use naive zero-shot agents as the baseline for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with no sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`:\n\n```\nvllm serve meta-llama/Llama-3.3-70B-Instruct --port 8080\n\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n agent.max_history=16 \\\n eval.num_workers=32 \\\n client.client_name=vllm \\\n client.model_id=meta-llama/Llama-3.3-70B-Instruct \\\n client.base_url=http://0.0.0.0:8080/v1\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper](https://arxiv.org/abs/2411.13543)\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [9.019607843137255, 2.9047857650598834, 30], "trajs": true, "verified": true}, {"average": [32.34, 1.49], "babaisai": [33.66, 3.3, 200], "babyai": [77.6, 3.73, 125], "crafter": [33.1, 2.32, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240924_naive-GPT4o", "minihack": [10.0, 4.74, 40], "name": "GPT-4o-2024-05-13", "nle": [0.37, 0.37, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [39.31, 5.24, 60], "trajs": false, "verified": true}, {"average": [21.0, 1.18], "babaisai": [32.02, 3.26, 200], "babyai": [58.4, 4.41, 125], "crafter": [30.21, 2.86, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240924_naive-Gemini-1.5-Pro", "minihack": [5.0, 3.45, 40], "name": "Gemini-1.5-Pro-002", "nle": [0.37, 0.37, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [0.0, 0.0, 60], "trajs": false, "verified": true}, {"average": [12.75, 1.56], "babaisai": [10.83, 2.84, 120], "babyai": [24.0, 6.04, 50], "crafter": [22.72, 2.73, 10], "date": "2024-11-25", "folder": "submissions/LLM/20241115_naive-Qwen2-VL-72B-Instruct", "minihack": [2.5, 2.47, 40], "name": "Qwen2-VL-72B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [16.47, 5.39, 30], "trajs": false, "verified": true}, {"average": [17.646375893946985, 1.4794126953475895], "babaisai": [20.833333333333336, 3.7073188375108734, 120], "babyai": [50.0, 7.0710678118654755, 50], "crafter": [27.727272727272727, 2.685282815953078, 10], "date": "2024-12-09", "folder": "submissions/LLM/20241209_naive-mistral-nemo-instruct", "minihack": [2.5, 2.4685522072664368, 40], "name": "Mistral-Nemo-it-2407", "nle": [0.30784538150722346, 0.2753452798438033, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nWe use naive zero-shot agents as the baseline for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with no sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`:\n\n```\nvllm serve mistralai/Mistral-Nemo-Instruct-2407 --port 8080\n\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n agent.max_history=16 \\\n eval.num_workers=32 \\\n client.client_name=vllm \\\n client.model_id=mistralai/Mistral-Nemo-Instruct-2407 \\\n client.base_url=http://0.0.0.0:8080/v1\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper](https://arxiv.org/abs/2411.13543)\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [4.509803921568627, 1.2912425975989794, 30], "trajs": true, "verified": true}, {"average": [27.876666666666665, 1.4263453688048735], "babaisai": [40.0, 3.42, 200], "babyai": [73.2, 3.96, 125], "crafter": [31.21, 2.68, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240930_naive-Llama-3.1-70B-Instruct", "minihack": [7.5, 4.16, 40], "name": "Llama-3.1-70B-it", "nle": [0.35, 0.35, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [15.0, 4.61, 60], "trajs": false, "verified": true}, {"average": [17.36, 1.35], "babaisai": [15.6, 2.53, 200], "babyai": [50.4, 4.47, 125], "crafter": [15.9, 2.05, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240924_naive-GPT4o-mini", "minihack": [10.0, 4.74, 40], "name": "GPT-4o-mini-2024-07-18", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [12.25, 3.55, 60], "trajs": false, "verified": true}, {"average": [27.293333333333337, 1.4426730591355603], "babaisai": [43.9, 3.47, 200], "babyai": [72.0, 6.35, 50], "crafter": [31.68, 1.36, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240930_naive-Llama-3.2-90B-Instruct", "minihack": [5.0, 3.44, 40], "name": "Llama-3.2-90B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [11.18, 2.98, 60], "trajs": false, "verified": true}, {"average": [19.31525904701802, 1.8294268418571227], "babaisai": [8.333333333333332, 2.5230419617479125, 120], "babyai": [52.0, 7.065408693062279, 50], "crafter": [26.36363636363636, 2.7872199485010705, 10], "date": "2024-12-11", "folder": "submissions/LLM/20241209_naive-claude-3-5-haiku", "minihack": [10.0, 4.743416490252569, 40], "name": "Claude-3.5-Haiku-2024-10-22", "nle": [1.1553688988639115, 0.423598115933611, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nWe use naive zero-shot agents as the baseline for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate this naive baselines, use `agent.type=naive`:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=claude \\\n client.model_id=claude-3-5-haiku-20241022\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper](https://arxiv.org/abs/2411.13543)\n- [BALROG](https://balrogai.com)", "site": null, "textworld": [18.03921568627451, 5.809571530119767, 30], "trajs": true, "verified": true}, {"average": [16.825, 1.470721667149234], "babaisai": [15.6, 2.5, 200], "babyai": [50.0, 7.07, 50], "crafter": [26.19, 3.29, 10], "date": "2024-11-11", "folder": "submissions/LLM/20240930_naive-Llama-3.2-11B-Instruct", "minihack": [2.5, 2.47, 40], "name": "Llama-3.2-11B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [6.66, 2.17, 60], "trajs": false, "verified": true}, {"average": [6.648989898989899, 1.0422013437544229], "babaisai": [10.833333333333334, 2.837211398277984, 120], "babyai": [8.0, 3.8366652186501793, 50], "crafter": [12.727272727272727, 1.9069251784911847, 10], "date": "2024-11-11", "folder": "submissions/LLM/20241030_naive-Llama-3.2-1B-Instruct", "minihack": [5.0, 3.446012188022554, 40], "name": "Llama-3.2-1B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "textworld": [3.3333333333333335, 0.9063547420103955, 30], "trajs": true, "verified": true}]}, {"name": "VLM", "results": [{"average": [35.47637553683924, 2.0209436790267405], "babaisai": [34.45378151260504, 4.3563101057350435, 119], "babyai": [82.0, 5.433231082882449, 50], "crafter": [37.272727272727266, 3.1360342383018796, 10], "date": "2024-11-11", "folder": "submissions/VLM/20241103_Claude-3.5-Sonnet", "minihack": [22.5, 6.602556323122129, 40], "name": "Claude-3.5-Sonnet-2024-10-22", "nle": [1.1553688988639113, 0.423598115933611, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": true, "verified": true}, {"average": [14.940000000000001, 1.3992655216219687], "babaisai": [8.3, 1.92, 200], "babyai": [43.2, 4.43, 125], "crafter": [20.7, 4.42, 10], "date": "2024-11-11", "folder": "submissions/VLM/20240924_naive-Gemini-1.5-Flash", "minihack": [2.5, 2.47, 40], "name": "Gemini-1.5-Flash-002", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}, {"average": [4.3799560913219935, 0.8492538039049989], "babaisai": [11.864406779661017, 2.976854968958091, 118], "babyai": [2.0, 1.9798989873223312, 50], "crafter": [5.454545454545454, 0.8624393618641033, 10], "date": "2024-11-25", "folder": "submissions/VLM/20241115_naive-Qwen2-VL-7B-instruct", "minihack": [5.0, 3.446012188022554, 40], "name": "Qwen2-VL-7B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)", "site": null, "textworld": [1.9607843137254901, 0.7509232217696767, 30], "trajs": false, "verified": true}, {"average": [22.560000000000002, 1.444882002102594], "babaisai": [18.62, 2.72, 200], "babyai": [62.0, 4.34, 125], "crafter": [26.81, 3.74, 10], "date": "2024-11-11", "folder": "submissions/VLM/20240930_naive-gpt-4o", "minihack": [5.0, 3.44, 40], "name": "GPT-4o-2024-05-13", "nle": [0.37, 0.37, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}, {"average": [25.7568, 1.3608593755417937], "babaisai": [31.4, 3.24, 200], "babyai": [58.4, 4.41, 125], "crafter": [33.5, 2.07, 10], "date": "2024-11-11", "folder": "submissions/VLM/20240924_naive-Gemini-1.5-Pro", "minihack": [5.0, 3.44, 40], "name": "Gemini-1.5-Pro-002", "nle": [0.484, 0.484, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}, {"average": [12.212, 1.5919912060058623], "babaisai": [5.93, 2.18, 120], "babyai": [34.0, 6.7, 50], "crafter": [18.63, 2.76, 10], "date": "2024-11-25", "folder": "submissions/VLM/20241115_naive-Qwen2-VL-72B-Instruct", "minihack": [2.5, 2.47, 40], "name": "Qwen2-VL-72B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}, {"average": [20.988, 1.582188357939724], "babaisai": [21.9, 2.89, 200], "babyai": [66.0, 6.7, 50], "crafter": [14.54, 1.8, 10], "date": "2024-11-11", "folder": "submissions/VLM/20240930_naive-Llama-3.2-90B-Instruct", "minihack": [2.5, 2.47, 40], "name": "Llama-3.2-90B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}, {"average": [8.434, 1.2576008905849263], "babaisai": [5.76, 1.63, 200], "babyai": [18.0, 5.43, 50], "crafter": [15.91, 1.16, 10], "date": "2024-11-11", "folder": "submissions/VLM/20240930_naive-Llama-3.2-11B-Instruct", "minihack": [2.5, 2.46, 40], "name": "Llama-3.2-11B-it", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}, {"average": [15.3632, 1.2875014563098561], "babaisai": [16.41, 2.59, 200], "babyai": [38.0, 4.34, 125], "crafter": [19.906, 3.13, 10], "date": "2024-11-11", "folder": "submissions/VLM/20240930_naive-gpt-4o-mini", "minihack": [2.5, 2.47, 40], "name": "GPT-4o-mini-2024-07-18", "nle": [0.0, 0.0, 5], "org_logo": "https://github.com/balrog-ai.png?size=200", "oss": true, "readme_content": "# BALROG Naive Baseline\nAs part of the original BALROG paper, we use naive zero-shot agents as the baseline approach for all the environments in BALROG.\n\nSpecifically:\n\n1. The agent is first given the rules of the games and available actions\n2. The agent is then shown the history of previous 16 observations-actions in a chat format, \n3. Finally, the agent is asked to only output a single action, with not sophisticated reasoning mechanism. \n\nTo replicate the naive baselines, use `agent.type=naive`, an example with GPT4o-mini is:\n\n```\npython eval.py \\\n agent.type=naive \\\n agent.max_image_history=0 \\\n eval.num_workers=32 \\\n client.client_name=openai \\\n client.model_id=gpt-4o-mini-2024-07-18\n```\n\nFor more information, visit:\n\n- [Evaluation tutorial](https://github.com/balrog-ai/BALROG/blob/main/assets/evaluation.md)\n- [Paper]()\n- [BALROG](https://balrogai.com)\n\n", "site": null, "trajs": false, "verified": true}]}]};
</script>
<script>
// Function to read query parameters
function getQueryParams() {
const params = new URLSearchParams(window.location.search);
return {
model: params.get('model'),
leaderboard: params.get('leaderboard')
};
}
// Function to display the model data
function displayModelData() {
const { model, leaderboard } = getQueryParams();
if (!model || !leaderboard) {
document.getElementById('modelName').textContent = 'Model not specified';
return;
}
const leaderboards = data.leaderboards;
let modelData = null;
// Find the model data
for (const lb of leaderboards) {
if (lb.name === leaderboard) {
modelData = lb.results.find(item => item.name === model);
break;
}
}
if (!modelData) {
document.getElementById('modelName').textContent = 'Model not found';
return;
}
// Update page content
document.getElementById('modelName').textContent = modelData.name;
document.getElementById('modelDate').textContent = 'Date: ' + modelData.date;
// Display performance data
const environments = ['babyai', 'crafter', 'textworld', 'babaisai', 'minihack', 'nle'];
const tbody = document.querySelector('#performanceTable tbody');
tbody.innerHTML = '';
for (const env of environments) {
if (env in modelData) {
const [progress, stdError, episodes] = modelData[env];
const row = document.createElement('tr');
row.innerHTML = `
<td>${env.charAt(0).toUpperCase() + env.slice(1)}</td>
<td>${progress.toFixed(2)}</td>
<td>${stdError.toFixed(2)}</td>
<td>${episodes}</td>
`;
tbody.appendChild(row);
}
}
// Add average
if ('average' in modelData) {
const [avgProgress, avgStdError] = modelData.average;
const avgRow = document.createElement('tr');
avgRow.innerHTML = `
<td><strong>Average</strong></td>
<td>${avgProgress.toFixed(2)}</td>
<td>${avgStdError.toFixed(2)}</td>
<td>-</td>
`;
tbody.appendChild(avgRow);
}
// Display README.md content
const readmeContent = modelData.readme_content || 'README.md not provided.';
const sanitizedContent = DOMPurify.sanitize(readmeContent);
document.getElementById('readme').innerHTML = marked.parse(sanitizedContent);
// Display Trajectories Information
const trajectoriesLinkContainer = document.getElementById('trajectoriesLinkContainer');
const trajectoriesText = document.getElementById('trajectoriesText');
if (modelData.trajs === true) {
// Construct the GitHub URL
const githubUrl = 'https://github.com/balrog-ai/experiments/tree/main/' + modelData.folder;
trajectoriesText.innerHTML = `Trajectories are available for this model. You can view them on GitHub: <a href="${githubUrl}" target="_blank">View Trajectories</a>`;
} else {
trajectoriesText.textContent = 'Trajectories are not available for this model.';
}
}
// Call the function on page load
displayModelData();
</script>
</body>
</html>