tomaarsen/mpnet-base-allnli

SentenceTransformer based on microsoft/mpnet-base

This is a sentence-transformers model finetuned from microsoft/mpnet-base on the multi_nli, snli and stsb datasets. 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: microsoft/mpnet-base
• Maximum Sequence Length: 384 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Dot Product
• Training Datasets:
  • multi_nli
  • snli
  • stsb
• Language: en
• License: apache-2.0

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': 384, 'do_lower_case': False}) with Transformer model: MPNetModel 
  (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): 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("tomaarsen/mpnet-base-allnli")
# Run inference
sentences = [
    "Rouen is the ancient center of Normandy's thriving textile industry, and the place of Joan of Arc's martyrdom ' a national symbol of resistance to tyranny.",
    'Joan of Arc sacrificed her life at Rouen, which became an enduring symbol of opposition to tyranny.',
    'The islands are part of France now instead of just colonies.',
]
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]

Evaluation

Metrics

Semantic Similarity

• Dataset: sts-dev
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8344
spearman_cosine	0.8295
pearson_manhattan	0.8317
spearman_manhattan	0.8332
pearson_euclidean	0.8273
spearman_euclidean	0.8295
pearson_dot	0.8344
spearman_dot	0.8295
pearson_max	0.8344
spearman_max	0.8332

Semantic Similarity

• Dataset: sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.7776
spearman_cosine	0.7643
pearson_manhattan	0.7788
spearman_manhattan	0.7659
pearson_euclidean	0.7763
spearman_euclidean	0.7643
pearson_dot	0.7776
spearman_dot	0.7643
pearson_max	0.7788
spearman_max	0.7659

Training Details

Training Datasets

multi_nli

• Dataset: multi_nli at da70db2
• Size: 392,702 training samples
• Columns: premise, hypothesis, and label
• Approximate statistics based on the first 1000 samples:

  	premise	hypothesis	label
  type	string	string	int
  details	min: 4 tokens mean: 26.95 tokens max: 189 tokens	min: 5 tokens mean: 14.11 tokens max: 49 tokens	0: ~34.30% 1: ~28.20% 2: ~37.50%

• Samples:

  premise	hypothesis	label
  Conceptually cream skimming has two basic dimensions - product and geography.	Product and geography are what make cream skimming work.	1
  you know during the season and i guess at at your level uh you lose them to the next level if if they decide to recall the the parent team the Braves decide to call to recall a guy from triple A then a double A guy goes up to replace him and a single A guy goes up to replace him	You lose the things to the following level if the people recall.	0
  One of our number will carry out your instructions minutely.	A member of my team will execute your orders with immense precision.	0

• Loss: SoftmaxLoss

snli

• Dataset: snli at cdb5c3d
• Size: 549,367 training samples
• Columns: snli_premise, hypothesis, and label
• Approximate statistics based on the first 1000 samples:

  	snli_premise	hypothesis	label
  type	string	string	int
  details	min: 6 tokens mean: 17.38 tokens max: 52 tokens	min: 4 tokens mean: 10.7 tokens max: 31 tokens	0: ~33.40% 1: ~33.30% 2: ~33.30%

• Samples:

  snli_premise	hypothesis	label
  A person on a horse jumps over a broken down airplane.	A person is training his horse for a competition.	1
  A person on a horse jumps over a broken down airplane.	A person is at a diner, ordering an omelette.	2
  A person on a horse jumps over a broken down airplane.	A person is outdoors, on a horse.	0

• Loss: SoftmaxLoss

stsb

• Dataset: stsb at 8913289
• Size: 5,749 training samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	float
  details	min: 6 tokens mean: 10.0 tokens max: 28 tokens	min: 5 tokens mean: 9.95 tokens max: 25 tokens	min: 0.0 mean: 0.54 max: 1.0

• Samples:

  sentence1	sentence2	label
  A plane is taking off.	An air plane is taking off.	1.0
  A man is playing a large flute.	A man is playing a flute.	0.76
  A man is spreading shreded cheese on a pizza.	A man is spreading shredded cheese on an uncooked pizza.	0.76

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Evaluation Datasets

multi_nli

• Dataset: multi_nli at da70db2
• Size: 100 evaluation samples
• Columns: premise, hypothesis, and label
• Approximate statistics based on the first 1000 samples:

  	premise	hypothesis	label
  type	string	string	int
  details	min: 5 tokens mean: 27.67 tokens max: 138 tokens	min: 6 tokens mean: 13.48 tokens max: 27 tokens	0: ~35.00% 1: ~31.00% 2: ~34.00%

• Samples:

  premise	hypothesis	label
  The new rights are nice enough	Everyone really likes the newest benefits	1
  This site includes a list of all award winners and a searchable database of Government Executive articles.	The Government Executive articles housed on the website are not able to be searched.	2
  uh i don't know i i have mixed emotions about him uh sometimes i like him but at the same times i love to see somebody beat him	I like him for the most part, but would still enjoy seeing someone beat him.	0

• Loss: SoftmaxLoss

snli

• Dataset: snli at cdb5c3d
• Size: 9,842 evaluation samples
• Columns: snli_premise, hypothesis, and label
• Approximate statistics based on the first 1000 samples:

  	snli_premise	hypothesis	label
  type	string	string	int
  details	min: 6 tokens mean: 18.44 tokens max: 57 tokens	min: 5 tokens mean: 10.57 tokens max: 25 tokens	0: ~33.10% 1: ~33.30% 2: ~33.60%

• Samples:

  snli_premise	hypothesis	label
  Two women are embracing while holding to go packages.	The sisters are hugging goodbye while holding to go packages after just eating lunch.	1
  Two women are embracing while holding to go packages.	Two woman are holding packages.	0
  Two women are embracing while holding to go packages.	The men are fighting outside a deli.	2

• Loss: SoftmaxLoss

stsb

• Dataset: stsb at 8913289
• Size: 1,500 evaluation samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	float
  details	min: 5 tokens mean: 15.1 tokens max: 45 tokens	min: 6 tokens mean: 15.11 tokens max: 53 tokens	min: 0.0 mean: 0.47 max: 1.0

• Samples:

  sentence1	sentence2	label
  A man with a hard hat is dancing.	A man wearing a hard hat is dancing.	1.0
  A young child is riding a horse.	A child is riding a horse.	0.95
  A man is feeding a mouse to a snake.	The man is feeding a mouse to the snake.	1.0

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• learning_rate: 2e-05
• num_train_epochs: 1
• warmup_ratio: 0.1
• seed: 33
• bf16: True
• load_best_model_at_end: True
• push_to_hub: True
• hub_model_id: tomaarsen/mpnet-base-allnli
• hub_private_repo: True
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_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: 1
• 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: 33
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: True
• 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: True
• 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: True
• resume_from_checkpoint: None
• hub_model_id: tomaarsen/mpnet-base-allnli
• hub_strategy: every_save
• hub_private_repo: True
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• dispatch_batches: None
• split_batches: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss	multi nli loss	snli loss	stsb loss	sts-dev_spearman_dot	sts-test_spearman_cosine
0.0370	10	0.8347	-	-	-	-	-
0.0741	20	0.8269	-	-	-	-	-
0.1111	30	0.7036	1.0978	1.0984	0.0830	0.6636	-
0.1481	40	0.7889	-	-	-	-	-
0.1852	50	0.7948	-	-	-	-	-
0.2222	60	0.688	1.0976	1.0961	0.0679	0.7124	-
0.2593	70	0.7911	-	-	-	-	-
0.2963	80	0.7847	-	-	-	-	-
0.3333	90	0.6801	1.0950	1.0942	0.0522	0.7810	-
0.3704	100	0.7837	-	-	-	-	-
0.4074	110	0.7803	-	-	-	-	-
0.4444	120	0.6756	1.0978	1.0929	0.0441	0.8157	-
0.4815	130	0.7829	-	-	-	-	-
0.5185	140	0.7789	-	-	-	-	-
0.5556	150	0.6756	1.0954	1.0911	0.0433	0.8215	-
0.5926	160	0.7802	-	-	-	-	-
0.6296	170	0.7751	-	-	-	-	-
0.6667	180	0.6679	1.0934	1.0885	0.0401	0.8235	-
0.7037	190	0.7755	-	-	-	-	-
0.7407	200	0.775	-	-	-	-	-
0.7778	210	0.6694	1.0919	1.0859	0.0377	0.8295	-
0.8148	220	0.7733	-	-	-	-	-
0.8519	230	0.772	-	-	-	-	-
0.8889	240	0.6656	1.0891	1.0838	0.0365	0.8292	-
0.9259	250	0.7726	-	-	-	-	-
0.9630	260	0.7731	-	-	-	-	-
1.0	270	0.6674	1.0888	1.0833	0.0372	0.8295	0.7643

• The bold row denotes the saved checkpoint.

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 0.018 kWh
• Carbon Emitted: 0.007 kg of CO2
• Hours Used: 0.068 hours

Training Hardware

• On Cloud: No
• GPU Model: 1 x NVIDIA GeForce RTX 3090
• CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
• RAM Size: 31.78 GB

Framework Versions

• Python: 3.11.6
• Sentence Transformers: 3.1.0.dev0
• Transformers: 4.41.2
• PyTorch: 2.3.0+cu121
• Accelerate: 0.30.1
• Datasets: 2.19.1
• Tokenizers: 0.19.1

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",
}