Chemlactica and Chemma are causal language models for small-molecule generation, property conditioning, and molecular optimization. They use structured text tags for SMILES, molecular properties, and molecular similarity.
| Model | Base model | Model card |
|---|---|---|
| Chemlactica-125M | Galactica-125M | Hugging Face |
| Chemlactica-1.3B | Galactica-1.3B | Hugging Face |
| Chemma-2B | Gemma-2B | Hugging Face |
The models were trained for 40 billion tokens over more than 100 million PubChem molecules. The public training corpus is PubChemForLM.
A structured prefix such as
[SAS]2.25[/SAS][SIMILAR]CC(=O)OC1=CC=CC=C1C(=O)O 0.62[/SIMILAR][START_SMILES]
asks the model to generate a molecule with the requested synthetic-accessibility score and similarity. The tags used for optimization must come from the training format; arbitrary oracle names are not natural-language instructions to these models.
The pinned Conda environment includes the training and optimization dependencies:
conda env create -f environment.yml
conda activate gemma_env_new
pip install -e .Model inference requires hardware appropriate for the selected checkpoint. The optimization code also requires RDKit.
The optimization loop is in
chemlactica/mol_opt/optimization.py, and
example_run.py shows the oracle interface
and configuration flow.
PMO experiments support two explicit information regimes:
- Task-agnostic: the objective is a black-box oracle. No task-derived structured prefix is added.
- Task-informed: public structure of the objective is encoded with the
model's trained tags, such as
[QED],[FORMULA],[TPSA],[CLOGP], and[SIMILAR]. Unsupported classifier names (DRD2,GSK3B, andJNK3) are not inserted as text.
Use the canonical registry rather than maintaining per-run prompt variants:
from chemlactica.mol_opt.optimization import optimize
from chemlactica.mol_opt.pmo_prompts import get_pmo_additional_properties
task_name = "qed"
prompt_mode = "task-informed" # or "task-agnostic"
optimize(
model,
tokenizer,
oracle,
config,
additional_properties=get_pmo_additional_properties(task_name, prompt_mode),
)The Valsartan SMARTS task has no trained SMARTS tag. Its exact SMARTS literal is
therefore kept unchanged as the reference inside [SIMILAR]; it is not expanded
into a molecule. The full registry and its rationale are implemented in
pmo_prompts.py.
The current canonical reproduction values below are sums of Top-10 AUC over all 23 Practical Molecular Optimization tasks. Each entry is the mean of five seeds evaluated over a 10,000-call horizon. Some saturated QED trajectories were stopped early and flat-filled to the horizon, following the benchmark's stopping convention.
| Model | Task-agnostic | Task-informed |
|---|---|---|
| Chemlactica-125M | 16.944598 | 20.429421 |
| Chemlactica-1.3B | 17.168555 | 20.192229 |
| Chemma-2B | 17.500147 | 20.550791 |
The two columns expose different amounts of task information and should be treated as separate comparison regimes. The reproductions use the corrected Sitagliptin oracle and a fixed prompt registry; online oracle calls are counted in the PMO budget. The public benchmark implementations are maintained in ChemLacticaTestSuite, while the canonical task-informed prompt definitions are maintained in this repository.
The current manuscript is available on arXiv.
Training entry points and model-specific configurations are under
chemlactica/. These files reflect the original research
workflows and may require adapting paths, distributed settings, and data locations
for a new cluster. The Hugging Face model cards are the recommended starting point
for inference.
Run the lightweight unit suite with:
python confirm_tests.py --run unitThe task-informed prompt tests verify coverage of all 23 PMO tasks, exact trained tag syntax, an empty task-agnostic path, and the Valsartan SMARTS handling.
Read Small Molecule Optimization with Large Language Models for the method and experimental context. The software in this repository is released under the MIT License.