SentenceTransformer based on google/embeddinggemma-300m

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 semantic textual similarity, semantic search, paraphrase mining, classification, clustering, and more.

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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("sentence_transformers_model_id")
# Run inference
queries = [
    'CC(C)c1cc(c(-c2ccc(F)cc2)n1\\C=C\\[C@@H](O)C[C@@H](O)CC([O-])=O)-c1ccccn1',
]
documents = [
    'CC(C)=CCC\\C(C)=C\\CCC1=C[C@@H](OC1=O)c1ccccc1O',
    'CCCCc1cc(OC)c(OC)cc1OC',
    'CO[C@@H]1C[C@H](OC(=O)C1)\\C=C\\c1c(C)cc(C)cc1-c1ccc(F)c(C)c1',
]
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([[-0.0571, -0.0621,  0.5212]])

Training Details

Training Dataset

Unnamed Dataset

Size: 42,498 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: 11 tokens mean: 57.24 tokens max: 110 tokens	min: 11 tokens mean: 51.97 tokens max: 110 tokens	0: ~47.12% 2: ~52.88%

Samples:

premise	hypothesis	label
`CCC(C)(C)C(=O)O[C@H]1CC@(CSc2ccccc2)O[C@@H]2CCC@HC@HC12`	`Cc1nnnn1C(\C=C[C@@H](O)CC@@HCC([O-])=O)=C(c1ccc(F)cc1)c1ccc(F)cc1`	`2`
`CCn1c(CCC@@HCC@@HCC([O-])=O)c(c(C)c1C(=O)Nc1ccccc1)-c1ccc(F)cc1`	`O[C@@H]1CC@H\C=C\c1c(Cl)cc(Cl)cc1OCc1cccnc1`	`2`
`CC(C)c1c(c(c(-c2ccc(F)cc2)n1CCC@@HCC@@HCC([O-])=O)-c1ccc(F)cc1)S(=O)(=O)Nc1cccc(CC(N)=O)c1`	`COc1ccc(OC)c(c1)[C@@H]1OC(=O)C(CC\C=C(/C)CCC=C(C)C)=C1`	`2`

Loss: SoftmaxLoss with these parameters:

{
    "num_labels": 3,
    "concatenation_sent_rep": true,
    "concatenation_sent_difference": true,
    "concatenation_sent_multiplication": false
}

Evaluation Dataset

Unnamed Dataset

Size: 7,500 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: 16 tokens mean: 57.12 tokens max: 93 tokens	min: 11 tokens mean: 52.61 tokens max: 90 tokens	0: ~43.27% 2: ~56.73%

Samples:

premise	hypothesis	label
`O[C@@H]1CC(CCc2ccccc2-c2ccccc2)OC(=O)C1`	`O[C@@H]1CC@H\C=C\c1ccc(Cl)cc1Cl`	`0`
`CC(C)n1c(CCC@@HCC@@HCC([O-])=O)c(c2CCCN(c3ccccc3)C(=O)c12)-c1ccc(F)cc1`	`CC(C)c1ccc(Sc2cc3c(Sc4ccc(cc4)C(C)C)c(\C=C[C@@H]4CC@@HCC(=O)O4)cnc3c(Sc3ccc(cc3)C(C)C)c2Sc2ccc(cc2)C(C)C)cc1`	`2`
`CC(C)c1c(cc(-c2ccc(F)cc2)n1CC[C@@H]1CC@@HCC(=O)O1)-c1ccccc1`	`CC(CCc1ccc2c(c1)-c1ccccc1S2(=O)=O)CC(O)=O`	`2`

Loss: SoftmaxLoss with these parameters:

{
    "num_labels": 3,
    "concatenation_sent_rep": true,
    "concatenation_sent_difference": true,
    "concatenation_sent_multiplication": false
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 32
warmup_steps: 10
optim: adafactor
weight_decay: 0.01
bf16: True
per_device_eval_batch_size: 32
load_best_model_at_end: True
dataloader_pin_memory: False

All Hyperparameters

Click to expand

per_device_train_batch_size: 32
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: 32
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
0.0150	20	0.9755	-
0.0301	40	0.9023	-
0.0376	50	-	0.8855
0.0451	60	0.8828	-
0.0602	80	0.8538	-
0.0752	100	0.8304	0.8613
0.0903	120	0.8119	-
0.1053	140	0.7917	-
0.1129	150	-	0.7640
0.1204	160	0.7857	-
0.1354	180	0.7664	-
0.1505	200	0.7574	0.7473
0.1655	220	0.7350	-
0.1806	240	0.7296	-
0.1881	250	-	0.7173
0.1956	260	0.7332	-
0.2107	280	0.7386	-
0.2257	300	0.7079	0.7244
0.2408	320	0.6684	-
0.2558	340	0.6670	-
0.2634	350	-	0.6760
0.2709	360	0.6600	-
0.2859	380	0.6630	-
0.3010	400	0.6050	0.6227
0.3160	420	0.6080	-
0.3311	440	0.6267	-
0.3386	450	-	0.6153
0.3461	460	0.6197	-
0.3612	480	0.6111	-
0.3762	500	0.5984	0.5897
0.3913	520	0.6073	-
0.4063	540	0.5479	-
0.4138	550	-	0.5703
0.4214	560	0.5916	-
0.4364	580	0.5567	-
0.4515	600	0.6482	0.7078
0.4665	620	0.5991	-
0.4816	640	0.5820	-
0.4891	650	-	0.5583
0.4966	660	0.5783	-
0.5117	680	0.5352	-
0.5267	700	0.5344	0.5442
0.5418	720	0.5527	-
0.5568	740	0.5349	-
0.5643	750	-	0.5376
0.5719	760	0.5310	-
0.5869	780	0.5384	-
0.6020	800	0.4973	0.5414
0.6170	820	0.5212	-
0.6321	840	0.4855	-
0.6396	850	-	0.5125
0.6471	860	0.5038	-
0.6622	880	0.5001	-
0.6772	900	0.4910	0.5051
0.6922	920	0.5000	-
0.7073	940	0.4682	-
0.7148	950	-	0.4980
0.7223	960	0.4830	-
0.7374	980	0.5125	-
0.7524	1000	0.5118	0.4886
0.7675	1020	0.4507	-
0.7825	1040	0.4715	-
0.7901	1050	-	0.4867
0.7976	1060	0.5020	-
0.8126	1080	0.4806	-
0.8277	1100	0.4737	0.4858
0.8427	1120	0.4677	-
0.8578	1140	0.4579	-
0.8653	1150	-	0.4536
0.8728	1160	0.4481	-
0.8879	1180	0.4897	-
0.9029	1200	0.4500	0.4570
0.9180	1220	0.4647	-
0.9330	1240	0.4498	-
0.9406	1250	-	0.4436
0.9481	1260	0.4429	-
0.9631	1280	0.4748	-
0.9782	1300	0.4139	0.4568
0.9932	1320	0.4340	-
1.0083	1340	0.4404	-
1.0158	1350	-	0.4689
1.0233	1360	0.4709	-
1.0384	1380	0.4103	-
1.0534	1400	0.3984	0.4238
1.0685	1420	0.4082	-
1.0835	1440	0.4505	-
1.091	1450	-	0.3957
1.0986	1460	0.4392	-
1.1136	1480	0.3966	-
1.1287	1500	0.3717	0.4049
1.1437	1520	0.3994	-
1.1588	1540	0.4006	-
1.1663	1550	-	0.4244
1.1738	1560	0.4005	-
1.1889	1580	0.3951	-
1.2039	1600	0.4065	0.4066

The bold row denotes the saved checkpoint.

Training Time

Training: 28.8 minutes

Framework Versions

Python: 3.12.13
Sentence Transformers: 5.7.0
Transformers: 5.13.1
PyTorch: 2.11.0+cu128
Accelerate: 1.14.0
Datasets: 5.0.1
Tokenizers: 0.22.2

Additional Resources

Training and Finetuning Embedding Models with Sentence Transformers: the end-to-end guide for training or finetuning Sentence Transformer models.
Introduction to Matryoshka Embedding Models: variable-size embeddings that can be truncated with minimal quality loss.
Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval: post-training compression of embedding vectors.
Multimodal Embedding & Reranker Models with Sentence Transformers: use text, image, audio, and video models through the same API.
Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers: train multimodal embedding models, with a Visual Document Retrieval walkthrough.

Citation

BibTeX

Sentence Transformers and SoftmaxLoss

@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",
}

Downloads last month: -

Safetensors

Model size

0.3B params

Tensor type

F32

Model tree for cafierom/smiles_embedding_gemma_FT_full

Base model

google/embeddinggemma-300m

Finetuned

(268)

this model

Paper for cafierom/smiles_embedding_gemma_FT_full

Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

Paper • 1908.10084 • Published Aug 27, 2019 • 16