zacbrld/MNLP_M3_document_encoder_V1

SentenceTransformer based on zacbrld/MNLP_M2_document_encoder

This is a sentence-transformers model finetuned from zacbrld/MNLP_M2_document_encoder. It maps sentences & paragraphs to a 384-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: zacbrld/MNLP_M2_document_encoder
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 384 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}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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): 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("zacbrld/MNLP_M3_document_encoder_V1")
# Run inference
sentences = [
    'synthesis method',
    'STEMNET used to receive funding from the Department for Education and Skills Since June 2007, it receives funding from the Department for Children, Schools and Families and Department for Innovation, Universities and Skills, since STEMNET sits on the chronological dividing point (age 16) of both of the new departments',
    'The Arab States of the Persian Gulf plan to start their own joint civilian nuclear program An agreement in the final days of the Bush administration provided for cooperation between the United Arab Emirates and the United States of America in which the United States would sell the UAE nuclear reactors and nuclear fuel The UAE would, in return, renounce their right to enrich uranium for their civilian nuclear program At the time of signing, this agreement was touted as a way to reduce risks of nuclear proliferation in the Persian Gulf However, Mustafa Alani of the Dubai-based Gulf Research Center stated that, should the Nuclear Non-Proliferation Treaty collapse, nuclear reactors such as those slated to be sold to the UAE under this agreement could provide the UAE with a path toward a nuclear weapon, raising the specter of further nuclear proliferation In March 2007, foreign ministers of the six-member Gulf Cooperation Council met in Saudi Arabia to discuss progress in plans agreed in December 2006, for a joint civilian nuclear program',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Training Details

Training Dataset

Unnamed Dataset

• Size: 34,235 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: 3 tokens mean: 21.24 tokens max: 256 tokens	min: 34 tokens mean: 133.62 tokens max: 256 tokens

• Samples:

  sentence_0	sentence_1
  chemistry experiment	Since 1982, research has been conducted to develop technologies, commonly referred to as electronic noses, that could detect and recognize odors and flavors Application areas include food, medicine and the environment
  quantum physics	Hydro electric - Hydro-electric turbomachinery uses potential energy stored in water to flow over an open impeller to turn a generator which creates electricity Steam turbines - Steam turbines used in power generation come in many different variations The overall principle is high pressure steam is forced over blades attached to a shaft, which turns a generator As the steam travels through the turbine, it passes through smaller blades causing the shaft to spin faster, creating more electricity Gas turbines - Gas turbines work much like steam turbines Air is forced in through a series of blades that turn a shaft Then fuel is mixed with the air and causes a combustion reaction, increasing the power This then causes the shaft to spin faster, creating more electricity Windmills - Also known as a wind turbine, windmills are increasing in popularity for their ability to efficiently use the wind to generate electricity Although they come in many shapes and sizes, the most common one is the la...
  physics law	Backlash in gear couplings allows for slight angular misalignment There can be significant backlash in unsynchronized transmissions because of the intentional gap between the dogs in dog clutches The gap is necessary to engage dogs when input shaft (engine) speed and output shaft (driveshaft) speed are imperfectly synchronized If there was a smaller clearance, it would be nearly impossible to engage the gears because the dogs would interfere with each other in most configurations In synchronized transmissions, synchromesh solves this problem However, backlash is undesirable in precision positioning applications such as machine tool tables It can be minimized by choosing ball screws or leadscrews with preloaded nuts, and mounting them in preloaded bearings A preloaded bearing uses a spring and/or a second bearing to provide a compressive axial force that maintains bearing surfaces in contact despite reversal of the load direction

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• num_train_epochs: 2
• 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: 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: 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: 2
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• 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
• use_ipex: 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}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• 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
• 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: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss
0.1168	500	1.465
0.2336	1000	1.189
0.3505	1500	1.1209
0.4673	2000	1.0333
0.5841	2500	0.993
0.7009	3000	0.9573
0.8178	3500	0.9275
0.9346	4000	0.9177
1.0514	4500	0.8241
1.1682	5000	0.7726
1.2850	5500	0.7685
1.4019	6000	0.7623
1.5187	6500	0.7668
1.6355	7000	0.7556
1.7523	7500	0.7002
1.8692	8000	0.7363
1.9860	8500	0.7396

Framework Versions

• Python: 3.12.8
• Sentence Transformers: 3.4.1
• Transformers: 4.52.2
• PyTorch: 2.7.0+cu126
• Accelerate: 1.3.0
• Datasets: 3.2.0
• Tokenizers: 0.21.0

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}
}