You can deploy this model and start earning money today!

ChemMRL: SMILES Matryoshka Representation Learning Embedding Transformer

Name: ChemMRL-beta
Rating: 2.5 (39 reviews)

This is a Chem-MRL (sentence-transformers) model finetuned from Derify/ChemBERTa-druglike on the pubchem_10m_genmol_similarity dataset. It maps SMILES to a 1024-dimensional dense vector space and can be used for molecular similarity, semantic search, database indexing, molecular classification, clustering, and more.

Model Details

Model Description

Model Type: ChemMRL (Sentence Transformer)
Base model: Derify/ChemBERTa-druglike
Maximum Sequence Length: 128 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Tanimoto
Training Dataset:
- pubchem_10m_genmol_similarity
License: apache-2.0

Model Sources

Repository: Chem-MRL on GitHub
Demo App Repository: Chem-MRL-demo on GitHub

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False, 'architecture': 'RobertaModel'})
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

Usage

Direct Usage (Chem-MRL)

First install the Chem-MRL library:

pip install -U chem-mrl>=0.7.3

Then you can load this model and run inference.

python

from chem_mrl import ChemMRL

# Download from the 🤗 Hub
model = ChemMRL("Derify/ChemMRL-beta")
# Run inference
sentences = [
    "Clc1nccc(C#CCCc2nc3ccccc3o2)n1",
    "O=Cc1nc2ccccc2o1",
    "O[C@H]1CN(C(Cc2ccccc2)c2ccccc2)C[C@@H]1Cc1cnc[nH]1",
]
embeddings = model.backbone.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.backbone.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.3200, 0.1209],
#         [0.3200, 1.0000, 0.0950],
#         [0.1209, 0.0950, 1.0000]])

# Load the model with half precision
model = ChemMRL("Derify/ChemMRL-beta", use_half_precision=True)
sentences = [
    "Clc1nccc(C#CCCc2nc3ccccc3o2)n1",
    "O=Cc1nc2ccccc2o1",
    "O[C@H]1CN(C(Cc2ccccc2)c2ccccc2)C[C@@H]1Cc1cnc[nH]1",
]
embeddings = model.embed(sentences)  # Use the embed method for half precision
print(embeddings.shape)
# [3, 1024]

Evaluation

Metrics

Semantic Similarity

Dataset: pubchem_10m_genmol_similarity
Evaluated with chem_mrl.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator with these parameters:
json
1{ 2 "precision": "float32" 3}

Split	Metric	Value
validation	spearman	0.993212
test	spearman	0.993243

Training Details

Training Dataset

pubchem_10m_genmol_similarity

Dataset: pubchem_10m_genmol_similarity at f68d779
Size: 19,692,766 training samples
Columns: smiles_a, smiles_b, and label
Approximate statistics based on the first 1000 samples:
smiles_a smiles_b label
type string string float
details
min: 17 tokens
mean: 39.66 tokens
max: 119 tokens
min: 11 tokens
mean: 38.29 tokens
max: 115 tokens
min: 0.02
mean: 0.57
max: 1.0
Loss: Matryoshka2dLoss with these parameters:

Click to expand

json
1{ 2 "loss": "TanimotoSentLoss", 3 "n_layers_per_step": -1, 4 "last_layer_weight": 2.0, 5 "prior_layers_weight": 1.0, 6 "kl_div_weight": 0.5, 7 "kl_temperature": 0.3, 8 "matryoshka_dims": [ 9 1024, 10 512, 11 256, 12 128, 13 64, 14 32, 15 16, 16 8 17 ], 18 "matryoshka_weights": [ 19 1, 20 1, 21 1, 22 1, 23 1, 24 1, 25 1, 26 1 27 ], 28 "n_dims_per_step": -1 29}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 128
learning_rate: 8e-06
weight_decay: 6.505130550397454e-06
warmup_ratio: 0.2
data_seed: 42
fp16: True
tf32: True
load_best_model_at_end: True
optim: adamw_apex_fused
dataloader_pin_memory: False

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 64
per_device_eval_batch_size: 128
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 1
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 8e-06
weight_decay: 6.505130550397454e-06
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 3
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.2
warmup_steps: 0
log_level: passive
log_level_replica: warning
log_on_each_node: True
logging_nan_inf_filter: True
save_safetensors: True
save_on_each_node: False
save_only_model: False
restore_callback_states_from_checkpoint: False
no_cuda: False
use_cpu: False
use_mps_device: False
seed: 42
data_seed: 42
jit_mode_eval: False
use_ipex: False
bf16: False
fp16: True
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: True
local_rank: 0
ddp_backend: None
tpu_num_cores: None
tpu_metrics_debug: False
debug: []
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_prefetch_factor: None
past_index: -1
disable_tqdm: False
remove_unused_columns: True
label_names: None
load_best_model_at_end: True
ignore_data_skip: False
fsdp: []
fsdp_min_num_params: 0
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
fsdp_transformer_layer_cls_to_wrap: None
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_apex_fused
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: False
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
hub_model_id: None
hub_strategy: every_save
hub_private_repo: None
hub_always_push: False
hub_revision: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
include_for_metrics: []
eval_do_concat_batches: True
fp16_backend: auto
push_to_hub_model_id: None
push_to_hub_organization: None
mp_parameters:
auto_find_batch_size: False
full_determinism: False
torchdynamo: None
ray_scope: last
ddp_timeout: 1800
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
include_tokens_per_second: False
include_num_input_tokens_seen: False
neftune_noise_alpha: None
optim_target_modules: None
batch_eval_metrics: False
eval_on_start: False
use_liger_kernel: False
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss	pubchem_10m_genmol_similarity_spearman
0.0796	24500	121.4633	-
0.08	24616	-	0.9739
0.1592	49000	118.6111	-
0.16	49232	-	0.9817
0.2389	73500	117.491	-
0.24	73848	-	0.9848
0.3185	98000	116.3786	-
0.32	98464	-	0.9865
0.3997	123000	115.9773	-
0.4	123080	-	0.9873
0.4794	147500	115.2441	-
0.48	147696	-	0.9885
0.5590	172000	114.8674	-
0.56	172312	-	0.9887
0.6386	196500	114.6483	-
0.64	196928	-	0.9892
0.7199	221500	114.0507	-
0.72	221544	-	0.9898
0.7995	246000	113.5606	-
0.8	246160	-	0.9902
0.8791	270500	113.2762	-
0.88	270776	-	0.9907
0.9587	295000	113.3295	-
0.96	295392	-	0.9908
1.0400	320000	112.9253	-
1.04	320008	-	0.9909
1.1196	344500	112.584	-
1.12	344624	-	0.9910
1.1992	369000	112.616	-
1.2	369240	-	0.9916
1.2788	393500	112.4692	-
1.28	393856	-	0.9914
1.3585	418000	112.2679	-
1.3600	418472	-	0.9917
1.4397	443000	112.1639	-
1.44	443088	-	0.9919
1.5193	467500	112.1139	-
1.52	467704	-	0.9921
1.5990	492000	111.8096	-
1.6	492320	-	0.9923
1.6786	516500	111.8252	-
1.6800	516936	-	0.9922
1.7598	541500	111.836	-
1.76	541552	-	0.9924
1.8395	566000	111.8471	-
1.8400	566168	-	0.9924
1.9191	590500	111.7778	-
1.92	590784	-	0.9925
1.9987	615000	111.4892	-
2.0	615400	-	0.9927
2.0799	640000	111.2659	-
2.08	640016	-	0.9928
2.1596	664500	111.3635	-
2.16	664632	-	0.9927
2.2392	689000	111.0114	-
2.24	689248	-	0.9928
2.3188	713500	111.0559	-
2.32	713864	-	0.9929
2.3984	738000	110.5276	-
2.4	738480	-	0.9929
2.4797	763000	110.9828	-
2.48	763096	-	0.9930
2.5593	787500	110.8404	-
2.56	787712	-	0.9930
2.6389	812000	111.1937	-
2.64	812328	-	0.9931
2.7186	836500	110.6662	-
2.7200	836944	-	0.9931
2.7998	861500	110.7714	-
2.8	861560	-	0.9932
2.8794	886000	110.7638	-
2.88	886176	-	0.9932
2.9591	910500	110.7021	-
2.96	910792	-	0.9932
2.9997	923000	110.6097	-

Training Hardware

On Cloud: No
GPU Model: 1 x NVIDIA GeForce RTX 3090
CPU Model: AMD Ryzen 7 3700X 8-Core Processor
RAM Size: 62.70 GB

Framework Versions

Python: 3.12.11
Sentence Transformers: 5.0.0
Transformers: 4.53.3
PyTorch: 2.7.1+cu126
Accelerate: 1.9.0
Datasets: 3.6.0
Tokenizers: 0.21.2

Citation

BibTeX

Sentence Transformers

bibtex

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

Matryoshka2dLoss

bibtex

@misc{li20242d,
    title={2D Matryoshka Sentence Embeddings},
    author={Xianming Li and Zongxi Li and Jing Li and Haoran Xie and Qing Li},
    year={2024},
    eprint={2402.14776},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

MatryoshkaLoss

bibtex

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

CoSENTLoss

bibtex

@online{kexuefm-8847,
    title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},
    author={Su Jianlin},
    year={2022},
    month={Jan},
    url={https://kexue.fm/archives/8847},
}

TanimotoSentLoss

bibtex

@online{cortes-2025-tanimotosentloss,
    title={TanimotoSentLoss: Tanimoto Loss for SMILES Embeddings},
    author={Emmanuel Cortes},
    year={2025},
    month={Jan},
    url={https://github.com/emapco/chem-mrl},
}

Model Card Authors

@eacortes

Model Card Contact

Manny Cortes (manny@derifyai.com)

	smiles_a	smiles_b	label
type	string	string	float
details	min: 17 tokens mean: 39.66 tokens max: 119 tokens	min: 11 tokens mean: 38.29 tokens max: 115 tokens	min: 0.02 mean: 0.57 max: 1.0