Papers › TOMG-Bench: Evaluating LLMs on Text-based Open Molecule Generation

TOMG-Bench: Evaluating LLMs on Text-based Open Molecule Generation

19 Dec 2024arXiv:2412.14642archive 2025-07-28

Jiatong Li, Junxian Li, Yunqing Liu, Dongzhan Zhou, Qing Li

In this paper, we propose Text-based Open Molecule Generation Benchmark (TOMG-Bench), the first benchmark to evaluate the open-domain molecule generation capability of LLMs. TOMG-Bench encompasses a dataset of three major tasks: molecule editing (MolEdit), molecule optimization (MolOpt), and customized molecule generation (MolCustom). Each major task further contains three subtasks, while each subtask comprises 5,000 test samples. Given the inherent complexity of open molecule generation evaluation, we also developed an automated evaluation system that helps measure both the quality and the accuracy of the generated molecules. Our comprehensive benchmarking of 25 LLMs reveals the current limitations as well as potential areas for improvement in text-guided molecule discovery. Furthermore, we propose OpenMolIns, a specialized instruction tuning dataset established for solving challenges raised by TOMG-Bench. Fine-tuned on OpenMolIns, Llama3.1-8B could outperform all the open-source general LLMs, even surpassing GPT-3.5-turbo by 46.5\% on TOMG-Bench. Our codes and datasets are available through https://github.com/phenixace/TOMG-Bench.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

phenixace/tomg-bench officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

BenchmarkingDescription-guided molecule generation

Datasets

Introduced by this paper, per the archive.

TOMG-Bench

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Description-guided molecule generation TOMG-Bench Claude-3.5 wAcc 35.92 #1 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Gemini-1.5-pro wAcc 34.80 #2 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench GPT-4-turbo wAcc 34.23 #3 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench GPT-4o wAcc 32.29 #4 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Claude-3 wAcc 30.47 #5 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Llama-3.1-8B (OpenMolIns-large) wAcc 27.22 #6 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Galactica-125M (OpenMolIns-xlarge) wAcc 25.73 #7 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Llama3-70B-Instruct (INT4) wAcc 23.93 #8 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Galactica-125M (OpenMolIns-large) wAcc 23.42 #9 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Galactica-125M (OpenMolIns-medium) wAcc 19.89 #10 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench GPT-3.5-turbo wAcc 18.58 #11 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Galactica-125M (OpenMolIns-small) wAcc 15.18 #12 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Llama3.1-8B-Instruct wAcc 14.09 #13 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Llama3-8B-Instruct wAcc 13.75 #14 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench chatglm-9B wAcc 13.137 #15 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Galactica-125M (OpenMolIns-light) wAcc 13.136 #16 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Llama3.2-1B (OpenMolIns-large) wAcc 8.10 #17 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench yi-1.5-9B wAcc 7.32 #18 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Mistral-7B-Instruct-v0.2 wAcc 4.81 #19 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench BioT5-base wAcc 4.21 #20 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench MolT5-large wAcc 2.89 #21 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Llama-3.1-1B-Instruct wAcc 1.99 #22 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench MolT5-base wAcc 1.300 #23 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench MolT5-small wAcc 1.299 #24 of 25 Archive leaderboard report
Description-guided molecule generation TOMG-Bench Qwen2-7B-Instruct wAcc 0.15 #25 of 25 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamAttentionAttention DropoutBPECosine AnnealingDense ConnectionsDropoutGPT-3Layer NormalizationLinear LayerLinear Warmup With Cosine AnnealingMulti-Head AttentionResidual ConnectionSoftmaxWeight Decay

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections