Model Card for ChemAligner-T5

How to Get Started with the Model

Below is an example of how to load and generate outputs with this model:

import torch
import transformers
from huggingface_hub import login
from transformers import AutoTokenizer
from transformers.models.t5 import T5ForConditionalGeneration
import torch

login('<your_hf_token>')
device = "cuda" if torch.cuda.is_available() else "cpu"

tokenizer = AutoTokenizer.from_pretrained("Neeze/ChemAligner-T5")
model = T5ForConditionalGeneration.from_pretrained("Neeze/ChemAligner-T5").to(device)

sample_caption = (
    "The molecule is a energy storage and a fat storage, which impacts cardiovascular "
    "disease, cancer, and metabolic syndrome, and is characterized as thyroxine treatment. "
    "The molecule is a membrane stabilizer and inflammatory, and it impacts pancreatitis. "
    "The molecule is a energy source and a nutrient, impacting both obesity and atherosclerosis."
)

task_definition = (
    "Definition: You are given a molecule description in English. "
    "Your job is to generate the corresponding molecule in SELFIES representation.\n\n"
)

task_input = (
    f"{task_definition}"
    f"Now complete the following example -\n"
    f"Input: {sample_caption}\nOutput: "
)

inputs = tokenizer(
    task_input,
    return_tensors="pt",
    truncation=True,
    max_length=512,
).to(device)

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_length=512,
        num_beams=1,
        do_sample=False,
        temperature=1.0,
        decoder_start_token_id=0,
        eos_token_id=1,
        pad_token_id=0
    )

outputs = [
    s.replace("<unk>", "").replace("<pad>", "").replace("</s>", "").strip()
    for s in tokenizer.batch_decode(outputs)
]

print(*outputs)

# [C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][=Branch1][C][=O][O][C][C@@H1][Branch2][Ring1][=Branch2][C][O][C][=Branch1][C][=O][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][O][C][=Branch1][C][=O][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][C][Branch1][C][C][C][C]

@inproceedings{Phan2026ChemAlignerT5,
  title     = {ChemAligner-T5: A Unified Text-to-Molecule Model via Representation Alignment},
  author    = {Nam, Van Hai Phan and
               Khoa, Minh Nguyen and
               Phu, Nguyen Ngoc Thien and
               Nguyen, Doan Hieu Nguyen and
               Tri, Minh Pham and
               Duc, Dang Ngoc Minh},
  booktitle = {Proceedings of the 2nd International Conference on Computational Intelligence in Engineering Science},
  year      = {2026},
  month     = apr,
  address   = {Nha Trang, Khanh Hoa, Vietnam}
}