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, text classification, clustering, and more.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: google/embeddinggemma-300m
Maximum Sequence Length: 2048 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 2048, 'do_lower_case': False, 'architecture': 'Gemma3TextModel'})
  (1): Pooling({'word_embedding_dimension': 768, '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): Dense({'in_features': 768, 'out_features': 3072, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (3): Dense({'in_features': 3072, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (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("ndsanjana/embedgemma_ns")
# Run inference
queries = [
    "Theme: Ethics of de\u2011extinction and scientific responsibility, Human ambition versus natural limits, Emergence of higher intelligence in extinct species, Corporate militarization of biological research, Coexistence and harmony between ancient and modern life forms",
]
documents = [
    'Theme: The ethical limits of scientific ambition, The moral implications of resurrecting extinct species, The clash between corporate exploitation and scientific integrity, The unexpected cognitive complexity of prehistoric life, The possibility of coexistence between past and present ecosystems',
    'Actions: Dr. Sarah Chen extracts viable DNA from a Triceratops fossil. -> She creates the first living dinosaur in 65 million years, nicknamed Trinity. -> The creature is publicly revealed, sparking global debate on de‑extinction ethics. -> Trinity exhibits unexpected higher intelligence. -> Biotech magnate Marcus Voss attempts to weaponize the research for military use. -> A confrontation occurs at the research facility. -> Trinity escapes into the nearby wilderness and encounters modern wildlife. -> Dr. Chen decides to destroy her research data to prevent further exploitation. -> Trinity disappears into a remote forest preserve. -> Final scene shows Trinity peacefully coexisting with a herd of elk.',
    '85_theme_vs_action',
]
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([[ 0.7758,  0.1831, -0.0576]])

Training Details

Training Dataset

Unnamed Dataset

Size: 1,000 training samples
Columns: anchor, positive, negative, triplet_id, and source

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative	triplet_id	source
type	string	string	string	string	string
details	min: 18 tokens mean: 80.7 tokens max: 204 tokens	min: 19 tokens mean: 81.97 tokens max: 201 tokens	min: 17 tokens mean: 83.77 tokens max: 230 tokens	min: 7 tokens mean: 9.25 tokens max: 11 tokens	min: 3 tokens mean: 3.0 tokens max: 3 tokens

Samples:

anchor	positive	negative	triplet_id	source
`Theme: inheritance, haunted house, supernatural, grief and loss, revenge, family dynamics, possession, exorcism, unresolved trauma, moral choice`	`Theme: Inheritance of legacy and the weight of family history, Supernatural haunting as a manifestation of unresolved trauma, The conflict between self-preservation and compassion, The cyclical nature of guilt and the desire for redemption, The tension between rational action and inexplicable forces`	`Theme: grief and avoidance, emotional healing, isolation and its psychological effects, responsibility toward family, the interplay between scientific curiosity and personal emotion, the reflective power of nature, guilt and unresolved conflict`	`0_theme_cross`	`unknown`
Actions: Family moves into inherited Victorian mansion -> Strange occurrences begin immediately -> Teenage daughter becomes primary target of supernatural activity -> Family researches property’s past and learns about reclusive widow and lost daughter -> Paranormal events intensify, threatening family safety -> Father attempts exorcism using items from hidden basement -> Exorcism angers the entity further -> Mother faces a critical choice: flee or help the spirit find peace by reuniting her with her daughter's remains -> Mother chooses to help the spirit	Actions: A newlywed couple inherits a sprawling ranch house in the desert from an estranged uncle. -> From the first night, bizarre phenomena (whispers, self-opening doors, sudden cold rooms) plague the household. -> The wife becomes the focal point of the disturbances, experiencing terrifying visions and speaking in unfamiliar voices. -> The couple investigates the property's history and learns that the former owner, an elderly hermit, died under suspicious circumstances after his young son accidentally died on the grounds. -> They discover that the hermit's ghost is desperately seeking someone to take his boy's place. -> The husband attempts to banish the spirit using ritual objects found in a concealed cellar. -> The ritual backfires, provoking the entity to greater violence and intensifying the supernatural assault. -> During the final confrontation, the wife faces an impossible decision: escape with her husband to safety or help the anguished ghost locate his son's hidden grave to...	Actions: Marine biologist accepts a research position at an isolated underwater station studying deep‑sea thermal vents. -> She leaves behind her estranged teenage son, who blames her for his father's recent death. -> During her six‑month assignment she discovers unusual bioluminescent organisms that respond to human emotions and memories. -> She spends more time observing the creatures, which triggers vivid recollections of her late husband and the unresolved guilt surrounding their final argument before his fatal accident. -> The organisms feed on her emotional energy, growing brighter and more active as her psychological state deteriorates. -> Her research partner becomes concerned about her erratic behavior and threatens to abort the mission. -> She realizes that her obsession with the creatures is a way of avoiding her grief and responsibility to her son. -> In the final act, she chooses to surface early and return home, accepting that healing requires facing her loss rather than ...	`0_action_cross`	`unknown`
`Outcomes: The mother’s decision to reunite the widow’s daughter’s remains brings peace to the spirit, ending the haunting. The family remains safe and can continue living in the house.`	`Outcomes: The story concludes with the wife's decision, leaving the haunting either unresolved if they escape or potentially resolved if they help the ghost find the grave. The final state is ambiguous, reflecting the unresolved tension between survival and compassion.`	`Outcomes: She returns home, confronts her grief and responsibility toward her son, and begins the process of healing.`	`0_outcome_cross`	`unknown`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false
}

Training Hyperparameters

Non-Default Hyperparameters

learning_rate: 2e-05
num_train_epochs: 10
warmup_ratio: 0.1
fp16: True
prompts: task: sentence similarity | query:

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
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: 2e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 10
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
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: None
jit_mode_eval: False
bf16: False
fp16: True
fp16_opt_level: O1
half_precision_backend: auto
bf16_full_eval: False
fp16_full_eval: False
tf32: None
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: False
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}
parallelism_config: None
deepspeed: None
label_smoothing_factor: 0.0
optim: adamw_torch_fused
optim_args: None
adafactor: False
group_by_length: False
length_column_name: length
project: huggingface
trackio_space_id: trackio
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
dataloader_pin_memory: True
dataloader_persistent_workers: False
skip_memory_metrics: True
use_legacy_prediction_loop: False
push_to_hub: False
resume_from_checkpoint: None
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: no
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: True
prompts: task: sentence similarity | query:
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.8	100	0.0664
1.6	200	0.017
2.4	300	0.018
3.2	400	0.005
4.0	500	0.026
4.8	600	0.0119
5.6	700	0.0083
6.4	800	0.0198
7.2	900	0.0217
8.0	1000	0.0123
8.8	1100	0.0174
9.6	1200	0.0112

Framework Versions

Python: 3.11.14
Sentence Transformers: 5.1.2
Transformers: 4.57.1
PyTorch: 2.9.1+cu128
Accelerate: 1.12.0
Datasets: 4.4.1
Tokenizers: 0.22.1

Citation

BibTeX

Sentence Transformers

@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",
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}