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: 256 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': 256, '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})
)

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("ahmedHamdi/narrative-similarity-fr-en-gemma-masked-NE")
# Run inference
sentences = [
    "A young man announces to his closest friends that he is about to marry a young woman from a wealthy family who has rejected him without a second thought. To mark the end of his bachelorhood, they decide to throw him a party he won't soon forget...",
    "ORG animal PERSON (PERSON), who makes his living as a Catholic-school bus driver, decides to settle down and marry his girlfriend PERSON (PERSON). After learning the news of the engagement, PERSON's shocked friends, led by PERSON (PERSON), decide to throw him an epic bachelor party. The bride's wealthy, conservative parents are unhappy with her decision, and her father enlists the help of ORG's ex-boyfriend PERSON (PERSON) to sabotage her relationship with PERSON and win her back. While ORG worries and goes off to a bridal shower thrown by her friends, PERSON heads to ORG, which takes place in a lavish, spacious hotel suite, and promises to remain faithful. Both parties start off on the wrong foot because of Cole's meddling. As the bachelor party starts to heat up, ORG and the girls decide to get even with PERSON and his friends by having a party of their own. Both parties eventually collide, leading to ORG accusing PERSON of infidelity. The bachelor party becomes a wild, drunken orgy and the hotel room is trashed, which infuriates the hotel's frustrated manager (PERSON). Adding to the confusion is PERSON's friend PERSON, who has become despondent over the breakup of his marriage and botches several suicide attempts. PERSON convinces PERSON of his love and faithfulness just as the party is raided by the police. In the ensuing melee, PERSON and ORG become separated and Cole kidnaps ORG, so PERSON and his friends chase after them. The chase culminates in a showdown between PERSON and Cole in a 36-screen movie theater, with a fist fight taking place in synchronization with a similar fight being shown in a 3D film projected behind them; the audience believes that the real fight is an extraordinary 3D effect. PERSON wins the fight and is reunited with ORG. After the wedding, PERSON and PERSON are driven to the airport for their honeymoon in PERSON's school bus, which is driven by a laughing PERSON.",
    'Film historian PERSON provides a synopsis of it, a film no longer known to exist in any archive:',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000,  0.6971,  0.0240],
#         [ 0.6971,  1.0000, -0.2039],
#         [ 0.0240, -0.2039,  1.0000]])

Training Details

Training Dataset

Unnamed Dataset

Size: 17,628 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1
type string string
details
min: 7 tokens
mean: 122.58 tokens
max: 256 tokens
min: 3 tokens
mean: 226.23 tokens
max: 256 tokens

Samples:

sentence_0	sentence_1
`In GPE, a lonely boy befriends his high school security guard who decides to help him get into hip-hop.`	In GPE, a single mother, PERSON, has cooked dinner and sends her daughter, PERSON, out to find her brother, August. PERSON and August are the victims of a shooting and PERSON dies. 18 months later August suffers from PTSD and anxiety attacks due to the incident and spends most of his time in his room. PERSON goes to work every day and is overprotective of August. August pines for the girl of his dreams PERSON and watches from his window as his best friend Laz sell drugs. School principal PERSON, tells her staff that she is going to cut personnel if attendance does not improve. She hires her soon to be ex-husband, PERSON, as the school security guard and tasks him with trying to get August, who has not been to school for months, to return to class. PERSON is unaware of August's issues and scares him when he enters his bedroom after hearing August working on a piece of music. PERSON secretly befriends August and sees the boy as his ticket back to the hip-hop music scene and a way to keep...
`In the early 1960s, British businessman PERSON (PERSON), due to his frequent travels to LOC, is recruited by PERSON, in conjunction with the ORG. He is tasked with going to the GPE to make contact with PERSON (PERSON), a colonel in ORG military intelligence service (GRU). Penkovsky fears that PERSON impulsiveness might lead him to trigger a nuclear conflict with LOC.`	In 1960, PERSON, a high ranking Soviet official and ORG intelligence officer with access to top secret nuclear information, is disillusioned with PERSON's leadership in light of the growing threat of a nuclear war with GPE. He reaches out to the ORG and offers to provide information that could help de-escalate the situation. The ORG and PERSON decide that it would be better not to use an officer and instead have an ordinary businessman act as an intermediary. They approach salesman PERSON to go to GPE under the pretense of exploring commercial opportunities. PERSON establishes seemingly normal business relations with PERSON and the latter makes arrangements with western intelligence agencies to feed them information. He asks that they continue to use PERSON as their regular courier, reasoning that he will be under the Soviets' radar. Initially opposed to the task, PERSON eventually agrees, partly after ORG officer PERSON emphasizes that his efforts could help prevent a nuclear war and ...
During the Warring States period, when GPE was divided into seven constantly warring regions, the great King of PERSON, through numerous battles and violent invasions, dreamed of reunifying his now-fragmented empire. PERSON, PERSON, and ORG, three of the greatest warriors of the PERSON kingdom, also known as the most active and dangerous conspirators, were said to have been slain by a man about whom little was known: Nameless. To reward him properly, PERSON wished to receive the mysterious hero. While no one was allowed to approach the King within one hundred paces, ORG, upon arriving at the palace, received gold and land and was permitted to stand twenty paces away from him, then ten, for having defeated PERSON and Snowflake. The King then asked him to tell his story. Nameless explains that PERSON and PERSON were lovers, but that ORG had had an affair with PERSON. PERSON hadn't forgiven him, so Nameless knew that to separate PERSON and PERSON, he first had to defeat PERSON. Nameless c...	In the middle of the Warring States period, ORG, a PERSON prefect, arrives at the PERSON capital city to meet the king, who has survived multiple attempts on his life by the assassins Long Sky, Flying Snow and PERSON. As a result, the king has implemented extreme security measures: no visitors are allowed to approach the king within 100 paces. Nameless claims that he has slain the three assassins, and their weapons are displayed before the king, who allows the former to approach within ten paces and tell him his story. Nameless recounts first killing PERSON at a gaming house, before traveling to meet ORG and PERSON, who have taken refuge at a calligraphy school in the PERSON state, in a city which was under siege by the PERSON army. He tells PERSON that he is there to commission a calligraphy scroll with the character for PERSON (劍), secretly seeking to learn PERSON's skill through his calligraphy. Nameless also learns that PERSON and PERSON, formerly lovers, have grown distant. Once ...

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim", 4 "gather_across_devices": false 5}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 4
per_device_eval_batch_size: 4
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 4
per_device_eval_batch_size: 4
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: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 3
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_ratio: 0.0
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: False
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: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.1135	500	0.0453
0.2269	1000	0.07
0.3404	1500	0.1489
0.4538	2000	0.1628
0.5673	2500	0.1959
0.6807	3000	0.1272
0.7942	3500	0.1977
0.9076	4000	0.173
1.0211	4500	0.1484
1.1346	5000	0.0902
1.2480	5500	0.11
1.3615	6000	0.1178
1.4749	6500	0.1181
1.5884	7000	0.0747
1.7018	7500	0.0857
1.8153	8000	0.1008
1.9287	8500	0.0765
2.0422	9000	0.0602
2.1557	9500	0.0294
2.2691	10000	0.0431
2.3826	10500	0.0323
2.4960	11000	0.0365
2.6095	11500	0.0421
2.7229	12000	0.0267
2.8364	12500	0.0214
2.9499	13000	0.0268

Framework Versions

Python: 3.9.18
Sentence Transformers: 5.1.2
Transformers: 4.57.6
PyTorch: 2.8.0+cu128
Accelerate: 1.10.1
Datasets: 4.5.0
Tokenizers: 0.22.2

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",
}

MultipleNegativesRankingLoss

bibtex

@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}
}

	sentence_0	sentence_1
type	string	string
details	min: 7 tokens mean: 122.58 tokens max: 256 tokens	min: 3 tokens mean: 226.23 tokens max: 256 tokens