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)c1c(C)c(C)c(-c2ccc(F)cc2)n1CC[C@@H]1C[C@@H](O)CC(=O)O1',
]
documents = [
    'O[C@@H]1C[C@H](OC(=O)C1)\\C=C\\c1c(Cl)cc(Cl)cc1OC\\C=C\\c1ccccc1',
    'COc1cccc(Sc2cc3nc(C4CC4)c(\\C=C\\[C@@H]4C[C@@H](O)CC(=O)O4)c(Sc4cccc(OC)c4)c3cc2F)c1',
    'C[C@H](CC\\C=C(/C)C(O)=O)[C@H]1CC[C@@]2(C)C3=CC[C@H]4C(C)(C)[C@@H](O)CC[C@]4(C)C3=CC[C@]12C',
]
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([[0.3374, 0.3095, 0.3194]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 84,996 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: 16 tokensmean: 58.66 tokensmax: 119 tokens	min: 16 tokensmean: 55.37 tokensmax: 108 tokens	0: ~54.81%2: ~45.19%

• Samples:

  premise	hypothesis	label
  Cc1cc(nc(-c2ccc(F)cc2)c1\C=C[C@@H](O)CC@@HCC([O-])=O)-c1ccccc1	Cc1cc(C)c(CCC2CC@@HCC(=O)O2)c(c1)-c1ccc(F)c(C)c1	0
  COC(=O)CC@HCC@H\C=C\n1c(cc(c1-c1ccc(F)cc1)-c1cccc(Br)c1)C(C)C	COc1cccc(Sc2c(\C=C[C@@H]3CC@@HCC(=O)O3)c(nc3ccc(F)cc23)C2CC2)c1	2
  CC(C)c1c(Br)c(Br)c(-c2ccc(F)cc2)n1CCC@@HCC@@HCC([O-])=O	CC(c1ccc(F)cc1)c1cc(C)cc(C)c1OCC(O)CC@@HCC([O-])=O	0

• 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: 15,000 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: 14 tokensmean: 57.03 tokensmax: 82 tokens	min: 12 tokensmean: 54.39 tokensmax: 113 tokens	0: ~50.00%2: ~50.00%

• Samples:

  premise	hypothesis	label
  CC(C)c1nc(nc(-c2ccc(F)cc2)c1\C=C[C@@H]1CC@HOCc1ccccc1)N(C)S(C)(=O)=O	O[C@@H]1CC@@HOC(=O)C1	2
  CC(C)(C)c1ccc(CC2CCCc3c2nn(c3\C=C[C@@H](O)CC@@HCC([O-])=O)-c2ccc(F)cc2)cc1	CC(C)=CCC\C(C)=C\CCC1=CC@@Hc1cc(F)ccc1F	2
  CC(C)n1c(CCC@@HCC@@HCC([O-])=O)c(c(c1C(N)=O)-c1ccccn1)-c1ccc(F)cc1	CC(C)c1nc(nc(-c2ccc(F)cc2)c1\C=C[C@@H]1CC@@HCC(=O)O1)-c1ccc(F)cc1	0

• 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

• 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

• The bold row denotes the saved checkpoint.

Training Time

• Training: 18.4 minutes
• Evaluation: 41.6 minutes
• Total: 1.0 hours

Framework Versions

• Python: 3.12.6
• Sentence Transformers: 5.7.0
• Transformers: 5.14.1
• PyTorch: 2.13.0+cu130
• 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",
}