This is a sentence-transformers model finetuned from google/embeddinggemma-300m. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: google/embeddinggemma-300m
Maximum Sequence Length: 128 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Supported Modality: Text

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'Gemma3TextModel'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'mean', 'include_prompt': True})
  (2): Dense({'in_features': 768, 'out_features': 3072, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity', 'module_input_name': 'sentence_embedding', 'module_output_name': 'sentence_embedding'})
  (3): Dense({'in_features': 3072, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity', 'module_input_name': 'sentence_embedding', 'module_output_name': 'sentence_embedding'})
  (4): Normalize({})
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("cafierom/smiles_embedding_gemma_FT")
# Run inference
queries = [
    'CCC(CC)(c1ccccc1)c1ccc(\\C=C\\[C@@H](O)C[C@@H](O)CC([O-])=O)c(c1)-c1ccccc1F',
]
documents = [
    'Cc1ccc(C(=C(\\C=C\\[C@@H](O)C[C@@H](O)CC([O-])=O)c2nnnn2C)c2ccc(C)cc2C)c(C)c1',
    'O[C@H](CCn1c(nc(c1-c1ccc(F)cc1)-c1ccc(F)cc1)C(F)(F)F)C[C@@H](O)CC([O-])=O',
    'CC(C)c1c(c(c(-c2ccc(F)cc2)n1CC[C@@H](O)C[C@@H](O)CC([O-])=O)-c1ccccc1)S(=O)(=O)N(C)C',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.9356, 0.9019, 0.5159]])

Training Details

Training Dataset

Unnamed Dataset

Size: 3,396 training samples
Columns: premise, hypothesis, and label
Approximate statistics based on the first 100 samples:
premise hypothesis label
type string string int
modality text text
details
min: 18 tokens
mean: 58.11 tokens
max: 117 tokens
min: 11 tokens
mean: 54.91 tokens
max: 119 tokens
0: ~55.77%
2: ~44.23%

Samples:

premise	hypothesis	label
`CC[object Object]C(=O)O[C@H]1C[object Object]\C=C\Cc1ccccc1`	`Cc1cc(-c2ccc(Cl)cc2)c(\C=C[C@@H]2C[object Object]CC(=O)O2)c(C)n1`	`2`
`CC(C)n1c(CC[object Object]C[object Object]CC([O-])=O)c(c(c1C(=O)Nc1ccccc1)-c1ccccc1)-c1ccc(F)cc1`	`CC(C)n1c(CC[object Object]C[object Object]CC([O-])=O)c(c(c1C(=O)NCc1ccccn1)-c1ccccc1)-c1ccc(F)cc1`	`0`
`CC(C)n1c(CC[object Object]C[object Object]CC([O-])=O)c(c(c1C(=O)N(C)C)-c1ccccc1)-c1ccc(F)cc1`	`COc1ccc2ccccc2c1-c1ccc(C(C)C)n1CC[C@@H]1C[object Object]CC(=O)O1`	`2`

Loss: SoftmaxLoss with these parameters:
json
1{ 2 "num_labels": 3, 3 "concatenation_sent_rep": true, 4 "concatenation_sent_difference": true, 5 "concatenation_sent_multiplication": false 6}

Evaluation Dataset

Unnamed Dataset

Size: 600 evaluation samples
Columns: premise, hypothesis, and label
Approximate statistics based on the first 100 samples:
premise hypothesis label
type string string int
modality text text
details
min: 11 tokens
mean: 56.06 tokens
max: 101 tokens
min: 12 tokens
mean: 55.85 tokens
max: 119 tokens
0: ~41.35%
2: ~58.65%

Samples:

premise	hypothesis	label
`CNc1cccc(c1)-c1nc(C(C)C)n(\C=C[C@@H](O)C[object Object]CC([O-])=O)c1-c1ccc(F)cc1`	`CC(C)c1c(CC[object Object]C[object Object]CC([O-])=O)c(cn1-c1ccccc1)-c1ccc(F)cc1`	`0`
`CC(C)c1nc(nc(-c2ccc(F)cc2)c1\C=C[C@@H]1C[object Object]CC(OCC(Cl)(Cl)Cl)O1)N(C)S(C)(=O)=O`	`CC(C)c1nn(-c2nc3ccccc3s2)c(c1\C=C[C@@H](O)C[object Object]CC([O-])=O)-c1ccc(F)cc1`	`2`
`Cc1c(\C=C[C@H](O)C[object Object]CC([O-])=O)c(cn1-c1ccccc1)-c1ccc(F)cc1`	`C[object Object][C@H]1CC(=O)[C@@]2(C)C3=C(C(=O)C[C@]12C)[C@@]1(C)CCC(=O)[object Object](COC(=O)C[object Object](O)CC(O)=O)[C@@H]1CC3=O`	`2`

Loss: SoftmaxLoss with these parameters:
json
1{ 2 "num_labels": 3, 3 "concatenation_sent_rep": true, 4 "concatenation_sent_difference": true, 5 "concatenation_sent_multiplication": false 6}

Training Hyperparameters

Non-Default Hyperparameters

warmup_steps: 10
optim: adafactor
weight_decay: 0.01
bf16: True
load_best_model_at_end: True
dataloader_pin_memory: False

All Hyperparameters

Click to expand

per_device_train_batch_size: 8
num_train_epochs: 3
max_steps: -1
learning_rate: 5e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 10
optim: adafactor
optim_args: None
weight_decay: 0.01
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 1.0
label_smoothing_factor: 0.0
bf16: True
fp16: False
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: huggingface
trackio_space_id: None
trackio_bucket_id: None
trackio_static_space_id: None
per_device_eval_batch_size: 8
prediction_loss_only: True
eval_on_start: False
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: True
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: False
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
parallelism_config: None
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: False
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: True
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_static_graph: None
ddp_backend: None
ddp_timeout: 1800
fsdp: None
fsdp_config: None
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: None
local_rank: -1
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss
0.0471	20	0.9762	-
0.0941	40	0.8931	-
0.1176	50	-	0.8680
0.1412	60	0.8805	-
0.1882	80	0.8463	-
0.2353	100	0.8186	0.8070
0.2824	120	0.8125	-
0.3294	140	0.7831	-
0.3529	150	-	0.7607
0.3765	160	0.7789	-
0.4235	180	0.7543	-
0.4706	200	0.7119	0.7341
0.5176	220	0.7230	-
0.5647	240	0.7006	-
0.5882	250	-	0.6976
0.6118	260	0.6999	-
0.6588	280	0.6649	-
0.7059	300	0.6693	0.6542
0.7529	320	0.6768	-
0.8	340	0.6203	-
0.8235	350	-	0.6334
0.8471	360	0.6820	-
0.8941	380	0.6402	-
0.9412	400	0.6327	0.6568
0.9882	420	0.5882	-
1.0353	440	0.6258	-
1.0588	450	-	0.6009
1.0824	460	0.5778	-
1.1294	480	0.6016	-
1.1765	500	0.5251	0.5823
1.2235	520	0.5810	-
1.2706	540	0.5700	-
1.2941	550	-	0.5618
1.3176	560	0.5384	-
1.3647	580	0.6171	-
1.4118	600	0.5861	0.5493
1.4588	620	0.5898	-
1.5059	640	0.5139	-
1.5294	650	-	0.5250
1.5529	660	0.5235	-
1.6	680	0.4972	-
1.6471	700	0.5056	0.5538
1.6941	720	0.5550	-
1.7412	740	0.5275	-
1.7647	750	-	0.5268
1.7882	760	0.5224	-
1.8353	780	0.4965	-
1.8824	800	0.4768	0.5072
1.9294	820	0.5545	-
1.9765	840	0.5012	-
2.0	850	-	0.5041
2.0235	860	0.5343	-
2.0706	880	0.4404	-
2.1176	900	0.4679	0.4866
2.1647	920	0.5153	-
2.2118	940	0.4592	-
2.2353	950	-	0.4959
2.2588	960	0.4473	-
2.3059	980	0.5223	-
2.3529	1000	0.4917	0.4882
2.4	1020	0.4113	-
2.4471	1040	0.5266	-
2.4706	1050	-	0.4844
2.4941	1060	0.4752	-
2.5412	1080	0.4672	-
2.5882	1100	0.5019	0.4730
2.6353	1120	0.4282	-
2.6824	1140	0.3954	-
2.7059	1150	-	0.4660
2.7294	1160	0.4608	-
2.7765	1180	0.4770	-
2.8235	1200	0.4747	0.4630
2.8706	1220	0.5013	-
2.9176	1240	0.3948	-
2.9412	1250	-	0.4606
2.9647	1260	0.4259	-
3.0	1275	-	0.4595

The bold row denotes the saved checkpoint.

Training Time

Training: 31.9 minutes
Evaluation: 19.4 minutes
Total: 51.3 minutes

Framework Versions

Python: 3.12.13
Sentence Transformers: 5.6.1
Transformers: 5.14.1
PyTorch: 2.13.0
Accelerate: 1.14.0
Datasets: 5.0.1
Tokenizers: 0.22.2

Citation

BibTeX

Sentence Transformers and SoftmaxLoss

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",
}

	premise	hypothesis	label
type	string	string	int
modality	text	text
details	min: 18 tokens mean: 58.11 tokens max: 117 tokens	min: 11 tokens mean: 54.91 tokens max: 119 tokens	0: ~55.77% 2: ~44.23%