ejun26/minilm-mrpc-clean-retrieval

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2 on the nyu-mll/glue dataset. 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: sentence-transformers/all-MiniLM-L6-v2
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 384 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • nyu-mll/glue
• Language: en

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("ejun26/minilm-mrpc-clean-retrieval")
# Run inference
sentences = [
    'In a statement later , he said it appeared his side may have fallen a bit short .',
    'Zilkha conceded in a statement issued today that his group may have fallen " a bit short . "',
    "U.S. law enforcement officials are sneering at Dar Heatherington 's version of of the events -- including a police conspiracy to discredit her -- which thrust her into the public spotlight .",
]
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]

Evaluation

Metrics

Clean Information Retrieval

• Dataset: mrpc-validation-clean-v2
• Evaluated with models.evaluator.CleanInformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.0
cosine_precision@3	0.3357
cosine_precision@5	0.2014
cosine_precision@10	0.1007
cosine_recall@1	0.0
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.6309
cosine_mrr@10	0.4994
cosine_map@100	0.5
dot_accuracy@1	0.0
dot_accuracy@3	1.0
dot_accuracy@5	1.0
dot_accuracy@10	1.0
dot_precision@1	0.0
dot_precision@3	0.3357
dot_precision@5	0.2014
dot_precision@10	0.1007
dot_recall@1	0.0
dot_recall@3	1.0
dot_recall@5	1.0
dot_recall@10	1.0
dot_ndcg@10	0.6309
dot_mrr@10	0.4994
dot_map@100	0.5

Clean Information Retrieval

• Dataset: mrpc-test-clean-v2
• Evaluated with models.evaluator.CleanInformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0
cosine_accuracy@3	0.9799
cosine_accuracy@5	0.9895
cosine_accuracy@10	0.9965
cosine_precision@1	0.0
cosine_precision@3	0.3275
cosine_precision@5	0.1984
cosine_precision@10	0.0999
cosine_recall@1	0.0
cosine_recall@3	0.9799
cosine_recall@5	0.9891
cosine_recall@10	0.9961
cosine_ndcg@10	0.623
cosine_mrr@10	0.4912
cosine_map@100	0.4915
dot_accuracy@1	0.0
dot_accuracy@3	0.9799
dot_accuracy@5	0.9895
dot_accuracy@10	0.9965
dot_precision@1	0.0
dot_precision@3	0.3275
dot_precision@5	0.1984
dot_precision@10	0.0999
dot_recall@1	0.0
dot_recall@3	0.9799
dot_recall@5	0.9891
dot_recall@10	0.9961
dot_ndcg@10	0.623
dot_mrr@10	0.4912
dot_map@100	0.4915

Training Details

Training Dataset

nyu-mll/glue

• Dataset: nyu-mll/glue at bcdcba7
• Size: 3,668 training samples
• Columns: text1, text2, and label
• Approximate statistics based on the first 1000 samples:

  	text1	text2	label
  type	string	string	int
  details	min: 9 tokens mean: 27.16 tokens max: 47 tokens	min: 11 tokens mean: 26.88 tokens max: 49 tokens	0: ~33.70% 1: ~66.30%

• Samples:

  text1	text2	label
  Amrozi accused his brother , whom he called " the witness " , of deliberately distorting his evidence .	Referring to him as only " the witness " , Amrozi accused his brother of deliberately distorting his evidence .	1
  Yucaipa owned Dominick 's before selling the chain to Safeway in 1998 for $ 2.5 billion .	Yucaipa bought Dominick 's in 1995 for $ 693 million and sold it to Safeway for $ 1.8 billion in 1998 .	0
  They had published an advertisement on the Internet on June 10 , offering the cargo for sale , he added .	On June 10 , the ship 's owners had published an advertisement on the Internet , offering the explosives for sale .	1

• Loss: OnlineContrastiveLoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 32
• learning_rate: 2e-05
• num_train_epochs: 5
• warmup_ratio: 0.1
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 32
• 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
• 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: 5
• 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
• 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: 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: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: False
• 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: proportional

Training Logs

Epoch	Step	Training Loss	mrpc-test-clean-v2_cosine_map@100	mrpc-validation-clean-v2_cosine_map@100
0	0	-	-	0.4961
0.0870	10	1.6166	-	-
0.1739	20	1.668	-	-
0.2609	30	1.5081	-	-
0.3478	40	1.3996	-	-
0.4348	50	1.2969	-	0.4985
0.5217	60	1.1771	-	-
0.6087	70	0.9977	-	-
0.6957	80	1.1213	-	-
0.7826	90	1.139	-	-
0.8696	100	1.0821	-	0.5
0.9565	110	1.1488	-	-
1.0435	120	0.932	-	-
1.1304	130	0.794	-	-
1.2174	140	0.9996	-	-
1.3043	150	0.9328	-	0.5
1.3913	160	1.1032	-	-
1.4783	170	0.9692	-	-
1.5652	180	0.9501	-	-
1.6522	190	0.7863	-	-
1.7391	200	0.8454	-	0.5
1.8261	210	0.9311	-	-
1.9130	220	0.8134	-	-
2.0	230	1.0013	-	-
2.0870	240	0.7564	-	-
2.1739	250	0.9165	-	0.5
2.2609	260	0.7668	-	-
2.3478	270	0.6587	-	-
2.4348	280	0.5904	-	-
2.5217	290	0.7431	-	-
2.6087	300	0.6133	-	0.5
2.6957	310	0.5994	-	-
2.7826	320	0.6256	-	-
2.8696	330	0.7294	-	-
2.9565	340	0.7527	-	-
3.0435	350	0.6908	-	0.5
3.1304	360	0.6455	-	-
3.2174	370	0.3765	-	-
3.3043	380	0.5955	-	-
3.3913	390	0.6239	-	-
3.4783	400	0.6666	-	0.5
3.5652	410	0.6498	-	-
3.6522	420	0.6363	-	-
3.7391	430	0.7046	-	-
3.8261	440	0.4384	-	-
3.9130	450	0.6721	-	0.5
4.0	460	0.5341	-	-
4.0870	470	0.4459	-	-
4.1739	480	0.4153	-	-
4.2609	490	0.5116	-	-
4.3478	500	0.4221	-	0.5
4.4348	510	0.4696	-	-
4.5217	520	0.4552	-	-
4.6087	530	0.5403	-	-
4.6957	540	0.367	-	-
4.7826	550	0.3275	-	0.5
4.8696	560	0.4016	-	-
4.9565	570	0.4889	-	-
5.0	575	-	0.4915	-

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 3.0.1
• Transformers: 4.41.0
• PyTorch: 2.2.2+cu121
• Accelerate: 1.12.0
• Datasets: 3.3.2
• Tokenizers: 0.19.1

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",
}