Hugging Face's logo

Datasets:
phenixace
/
S2-TOMG-Bench

Modalities:
Tabular
Text
Formats:
csv
Languages:
English
Size:
10K - 100K
ArXiv:
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

Add task categories and paper links to dataset card

#1
by nielsr HF Staff - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+42
-53
Files changed (1) hide show
  1. README.md +42 -53
README.md CHANGED
@@ -1,54 +1,65 @@
1
  ---
2
- license: apache-2.0
3
  language:
4
  - en
 
 
 
5
  configs:
6
  - config_name: MolCustom_AtomNum
7
  data_files:
8
  - split: test
9
- path: "MolCustom_AtomNum.csv"
10
  - config_name: MolCustom_BondNum
11
  data_files:
12
  - split: test
13
- path: "MolCustom_BondNum.csv"
14
  - config_name: MolCustom_FunctionalGroup
15
  data_files:
16
  - split: test
17
- path: "MolCustom_FunctionalGroup.csv"
18
  - config_name: MolEdit_AddComponent
19
  data_files:
20
  - split: test
21
- path: "MolEdit_AddComponent.csv"
22
  - config_name: MolEdit_SubComponent
23
  data_files:
24
  - split: test
25
- path: "MolEdit_SubComponent.csv"
26
  - config_name: MolEdit_DelComponent
27
  data_files:
28
  - split: test
29
- path: "MolEdit_DelComponent.csv"
30
  - config_name: MolOpt_LogP
31
  data_files:
32
  - split: test
33
- path: "MolOpt_LogP.csv"
34
  - config_name: MolOpt_MR
35
  data_files:
36
  - split: test
37
- path: "MolOpt_MR.csv"
38
  - config_name: MolOpt_QED
39
  data_files:
40
  - split: test
41
- path: "MolOpt_QED.csv"
42
  ---
43
 
44
- # S^2-Bench Dataset (TOMG) (full version, 45k entries)
45
- Official Huggingface Datasets for [S^2-Bench](https://phenixace.github.io/tomgbench/): *"Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation"*
 
 
 
46
 
47
- Please refer to our [Github Repo](https://github.com/phenixace/S2-TOMG-Bench) for more usage and useful information.
 
 
 
 
 
 
48
 
49
  ## Configurations
50
 
51
- Each configuration represents a different task:
52
 
53
  - **MolCustom_AtomNum**: Molecular customized generation by atom number
54
  - **MolCustom_BondNum**: Molecular customized generation by bond number
@@ -77,51 +88,30 @@ datasets = {config: load_dataset("phenixace/S2-TOMG-Bench", config) for config i
77
  ```
78
 
79
  ## Citation
80
- If you use our data, please cite us in the format below:
81
  ```bibtex
82
  @article{li2024speak,
83
  title={Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation},
84
  author={Li, Jiatong and Li, Junxian and Liu, Yunqing and Zheng, Changmeng and Wei, Xiaoyong and Zhou, Dongzhan and Li, Qing},
85
- journal={arXiv preprint arXiv:2412.14642v3},
86
  year={2024}
87
  }
88
  ```
89
 
90
-
91
- ## Current Leaderboard
92
-
93
- | Rank | Model | \#Parameters (B) | $\overline{S\!R}$ (\%) | $\overline{W\!S\!R} (\%)$ | |
94
- |------|-----------------------------------------------------|------------------|------------------------|---------------------------|---|
95
- | 1 | Llama3.1-8B (OpenMolIns-xlarge) | 8 | 58.79 | 39.33 | |
96
- | 2 | Claude-3.5 | - | 51.10 | 35.92 | |
97
- | 3 | Gemini-1.5-pro | - | 52.25 | 34.80 | |
98
- | 4 | GPT-4-turbo | - | 50.74 | 34.23 | |
99
- | 5 | GPT-4o | - | 49.08 | 32.29 | |
100
- | 6 | Claude-3 | - | 46.14 | 30.47 | |
101
- | 7 | Llama3.1-8B (OpenMolIns-large) | 8 | 43.1 | 27.22 | |
102
- | 8 | Galactica-125M (OpenMolIns-xlarge) | 0.125 | 44.48 | 25.73 | |
103
- | 9 | Llama3-70B-Instruct (Int4) | 70 | 38.54 | 23.93 | |
104
- | 10 | Galactica-125M (OpenMolIns-large) | 0.125 | 39.28 | 23.42 | |
105
- | 11 | Galactica-125M (OpenMolIns-medium) | 0.125 | 34.54 | 19.89 | |
106
- | 12 | GPT-3.5-turbo | - | 28.93 | 18.58 | |
107
- | 13 | Galactica-125M (OpenMolIns-small) | 0.125 | 24.17 | 15.18 | |
108
- | 14 | Gemma3-12B | 12 | 26.28 | 15.00 | |
109
- | 15 | Deepseek-R1-distill-Qwen-7B | 7 | 25.07 | 14.61 | |
110
- | 16 | Llama3.1-8B-Instruct | 8 | 26.26 | 14.09 | |
111
- | 17 | Llama3-8B-Instruct | 8 | 26.40 | 13.75 | |
112
- | 18 | chatglm-9B | 9 | 18.50 | 13.13(7) | |
113
- | 19 | Galactica-125M (OpenMolIns-light) | 0.125 | 20.95 | 13.13(6) | |
114
- | 20 | ChemDFM-v1.5-8B | 8 | 18.24 | 12.07 | |
115
- | 21 | ChemLLM-20B | 20 | 16.23 | 9.76 | |
116
- | 22 | Llama3.2-1B (OpenMolIns-large) | 1 | 14.11 | 8.10 | |
117
- | 23 | yi-1.5-9B | 9 | 14.10 | 7.32 | |
118
- | 24 | Mistral-7B-Instruct-v0.2 | 7 | 11.17 | 4.81 | |
119
- | 25 | BioT5-base | 0.25 | 24.19 | 4.21 | |
120
- | 26 | MolT5-large | 0.78 | 23.11 | 2.89 | |
121
- | 27 | Llama3.1-1B-Instruct | 1 | 3.95 | 1.99 | |
122
- | 28 | MolT5-base | 0.25 | 11.11 | 1.30(0) | |
123
- | 29 | MolT5-small | 0.08 | 11.55 | 1.29(9) | |
124
- | 30 | Qwen2-7B-Instruct | 7 | 0.18 | 0.15 | |
125
 
126
  ## OpenMolIns
127
 
@@ -131,5 +121,4 @@ The instruction tuning datasets are also available at Hugging Face Datasets:
131
  2. [OpenMolIns-small](https://huggingface.co/datasets/phenixace/OpenMolIns-small)
132
  3. [OpenMolIns-medium](https://huggingface.co/datasets/phenixace/OpenMolIns-medium)
133
  4. [OpenMolIns-large](https://huggingface.co/datasets/phenixace/OpenMolIns-large)
134
- 5. [OpenMolIns-xlarge](https://huggingface.co/datasets/phenixace/OpenMolIns-xlarge)
135
-
 
1
  ---
 
2
  language:
3
  - en
4
+ license: apache-2.0
5
+ task_categories:
6
+ - text-generation
7
  configs:
8
  - config_name: MolCustom_AtomNum
9
  data_files:
10
  - split: test
11
+ path: MolCustom_AtomNum.csv
12
  - config_name: MolCustom_BondNum
13
  data_files:
14
  - split: test
15
+ path: MolCustom_BondNum.csv
16
  - config_name: MolCustom_FunctionalGroup
17
  data_files:
18
  - split: test
19
+ path: MolCustom_FunctionalGroup.csv
20
  - config_name: MolEdit_AddComponent
21
  data_files:
22
  - split: test
23
+ path: MolEdit_AddComponent.csv
24
  - config_name: MolEdit_SubComponent
25
  data_files:
26
  - split: test
27
+ path: MolEdit_SubComponent.csv
28
  - config_name: MolEdit_DelComponent
29
  data_files:
30
  - split: test
31
+ path: MolEdit_DelComponent.csv
32
  - config_name: MolOpt_LogP
33
  data_files:
34
  - split: test
35
+ path: MolOpt_LogP.csv
36
  - config_name: MolOpt_MR
37
  data_files:
38
  - split: test
39
+ path: MolOpt_MR.csv
40
  - config_name: MolOpt_QED
41
  data_files:
42
  - split: test
43
+ path: MolOpt_QED.csv
44
  ---
45
 
46
+ # S^2-Bench Dataset (TOMG)
47
+
48
+ Official Hugging Face dataset for **S^2-Bench**: [*"Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation"*](https://huggingface.co/papers/2412.14642).
49
+
50
+ [**Project Page**](https://phenixace.github.io/tomgbench/) | [**Github Repo**](https://github.com/phenixace/S2-TOMG-Bench) | [**Paper**](https://huggingface.co/papers/2412.14642)
51
 
52
+ ## Introduction
53
+ Speak-to-Structure (S^2-Bench), also referred to as TOMG-Bench, is a benchmark designed to evaluate Large Language Models (LLMs) in open-domain natural language-driven molecule generation. Unlike traditional one-to-one mapping benchmarks, S^2-Bench focuses on one-to-many relationships, challenging LLMs to exhibit genuine molecular understanding and creative generation capabilities.
54
+
55
+ The benchmark includes three major tasks:
56
+ - **MolEdit**: Molecule editing (AddComponent, SubComponent, DelComponent)
57
+ - **MolOpt**: Molecule optimization (LogP, MR, QED)
58
+ - **MolCustom**: Customized molecule generation (AtomNum, BondNum, FunctionalGroup)
59
 
60
  ## Configurations
61
 
62
+ Each configuration represents a different subtask:
63
 
64
  - **MolCustom_AtomNum**: Molecular customized generation by atom number
65
  - **MolCustom_BondNum**: Molecular customized generation by bond number
 
88
  ```
89
 
90
  ## Citation
91
+ If you use our data, please cite us:
92
  ```bibtex
93
  @article{li2024speak,
94
  title={Speak-to-Structure: Evaluating LLMs in Open-domain Natural Language-Driven Molecule Generation},
95
  author={Li, Jiatong and Li, Junxian and Liu, Yunqing and Zheng, Changmeng and Wei, Xiaoyong and Zhou, Dongzhan and Li, Qing},
96
+ journal={arXiv preprint arXiv:2412.14642},
97
  year={2024}
98
  }
99
  ```
100
 
101
+ ## Leaderboard
102
+
103
+ | Rank | Model | \#Parameters (B) | $\overline{S\!R}$ (%) | $\overline{W\!S\!R} (\%)$ |
104
+ |------|-----------------------------------------------------|------------------|------------------------|---------------------------|
105
+ | 1 | Llama3.1-8B (OpenMolIns-xlarge) | 8 | 58.79 | 39.33 |
106
+ | 2 | Claude-3.5 | - | 51.10 | 35.92 |
107
+ | 3 | Gemini-1.5-pro | - | 52.25 | 34.80 |
108
+ | 4 | GPT-4-turbo | - | 50.74 | 34.23 |
109
+ | 5 | GPT-4o | - | 49.08 | 32.29 |
110
+ | 6 | Claude-3 | - | 46.14 | 30.47 |
111
+ | 7 | Llama3.1-8B (OpenMolIns-large) | 8 | 43.1 | 27.22 |
112
+ | 8 | Galactica-125M (OpenMolIns-xlarge) | 0.125 | 44.48 | 25.73 |
113
+ | 9 | Llama3-70B-Instruct (Int4) | 70 | 38.54 | 23.93 |
114
+ | 10 | Galactica-125M (OpenMolIns-large) | 0.125 | 39.28 | 23.42 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
115
 
116
  ## OpenMolIns
117
 
 
121
  2. [OpenMolIns-small](https://huggingface.co/datasets/phenixace/OpenMolIns-small)
122
  3. [OpenMolIns-medium](https://huggingface.co/datasets/phenixace/OpenMolIns-medium)
123
  4. [OpenMolIns-large](https://huggingface.co/datasets/phenixace/OpenMolIns-large)
124
+ 5. [OpenMolIns-xlarge](https://huggingface.co/datasets/phenixace/OpenMolIns-xlarge)