Shuu12121 / CodeEmbed-ModernBERT-Owl-Preview

SentenceTransformer based on Shuu12121/CodeModernBERT-Owl

This is a sentence-transformers model finetuned from Shuu12121/CodeModernBERT-Owl . 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: Shuu12121/CodeModernBERT-Owl
• 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}) with Transformer model: ModernBertModel 
  (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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("sentence_transformers_model_id")
# Run inference
sentences = [
    'Specify whether MediaConvert generates images for trick play. Keep the default value, None, to not generate any images. Choose Thumbnail to generate tiled thumbnails. Choose Thumbnail and full frame to generate tiled thumbnails and full-resolution images of single frames. MediaConvert adds an entry in the .mpd manifest for each set of images that you generate. A common application for these images is Roku trick mode. The thumbnails and full-frame images that MediaConvert creates with this feature are compatible with this Roku specification: https://developer.roku.com/docs/developer-program/media-playback/trick-mode/hls-and-dash.md',
    '    pub fn set_image_based_trick_play(mut self, input: ::std::option::Option<crate::types::DashIsoImageBasedTrickPlay>) -> Self {\n        self.image_based_trick_play = input;\n        self\n    }',
    'def reset_key\n      properties.key.zip(key) do |property, value|\n        property.set!(self, value)\n      end\n    end',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Training Details

Training Dataset

Unnamed Dataset

• Size: 773,718 training samples
• Columns: sentence_0 , sentence_1 , and label
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1	label
  type	string	string	float
  details	min: 3 tokens mean: 60.75 tokens max: 1758 tokens	min: 15 tokens mean: 168.69 tokens max: 2048 tokens	min: 1.0 mean: 1.0 max: 1.0

• Samples:

  sentence_0	sentence_1	label
  Returns the value previously set for the given variable. @param name the variable to lookup @return the value associated with the given name. Uses the parent scope if no variable with the given name exists. Returns a {@link Value} with "" as content, in case the value is completely unknown.	public Expression get(String name) { if (variables.containsKey(name)) { return variables.get(name); } if (parent == null) { return new Value(""); } return parent.get(name); }	1.0
  The recommended vertical offset in font design units for superscripts for this font.	pub fn y_superscript_y_offset(&self) -> i16 { let range = self.shape.y_superscript_y_offset_byte_range(); self.data.read_at(range.start).unwrap() }	1.0
  // Render renders the generated markup for this view	func (v *View) Render(m ...string) trees.Markup { if len(m) <= 0 { m = []string{"."} } v.Engine().All(m[0]) if v.rview == nil { return elems.Div() } dom := v.rview.Render(m...) if dom == nil { return elems.Div() } // // swap the uid for the new dom // // to ensure we keep the sync between backend and frontend in sync. v.backdoor.M = dom v.backdoor.SwapUID(v.uid) v.backdoor.M = nil if v.liveMarkup != nil { dom.Reconcile(v.liveMarkup) } dom.UseEventManager(v.events) v.events.LoadUpEvents() v.liveMarkup = dom return dom }	1.0

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy : steps
• per_device_train_batch_size : 24
• per_device_eval_batch_size : 24
• fp16 : True
• multi_dataset_batch_sampler : round_robin

All Hyperparameters

Training Logs

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.50.0
• PyTorch: 2.6.0+cu124
• Accelerate: 1.5.2
• Datasets: 3.4.1
• Tokenizers: 0.21.1

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}
}