tomaarsen/distilroberta-base-nli-matryoshka-reduced

SentenceTransformer based on distilbert/distilroberta-base

This is a sentence-transformers model finetuned from distilbert/distilroberta-base on the sentence-transformers/all-nli dataset. It maps sentences & paragraphs to a 256-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: distilbert/distilroberta-base
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 256 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • sentence-transformers/all-nli
• 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': 512, 'do_lower_case': False}) with Transformer model: RobertaModel 
  (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})
  (reduced_dim): Dense({'in_features': 768, 'out_features': 256, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
)

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/distilroberta-base-nli-matryoshka-reduced")
# Run inference
sentences = [
    'A boy is vacuuming.',
    'A little boy is vacuuming the floor.',
    'A woman is applying eye shadow.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 256]

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

Evaluation

Metrics

Semantic Similarity

• Dataset: sts-dev-256
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.833
spearman_cosine	0.845
pearson_manhattan	0.8284
spearman_manhattan	0.8314
pearson_euclidean	0.8291
spearman_euclidean	0.8319
pearson_dot	0.7274
spearman_dot	0.7358
pearson_max	0.833
spearman_max	0.845

Semantic Similarity

• Dataset: sts-dev-128
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8266
spearman_cosine	0.8416
pearson_manhattan	0.825
spearman_manhattan	0.8277
pearson_euclidean	0.8256
spearman_euclidean	0.8285
pearson_dot	0.712
spearman_dot	0.7163
pearson_max	0.8266
spearman_max	0.8416

Semantic Similarity

• Dataset: sts-dev-64
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8171
spearman_cosine	0.8356
pearson_manhattan	0.8176
spearman_manhattan	0.8213
pearson_euclidean	0.8175
spearman_euclidean	0.8216
pearson_dot	0.6852
spearman_dot	0.6861
pearson_max	0.8176
spearman_max	0.8356

Semantic Similarity

• Dataset: sts-dev-32
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.7964
spearman_cosine	0.8244
pearson_manhattan	0.7983
spearman_manhattan	0.8049
pearson_euclidean	0.8003
spearman_euclidean	0.807
pearson_dot	0.6312
spearman_dot	0.6277
pearson_max	0.8003
spearman_max	0.8244

Semantic Similarity

• Dataset: sts-dev-16
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.7401
spearman_cosine	0.7872
pearson_manhattan	0.761
spearman_manhattan	0.7761
pearson_euclidean	0.7645
spearman_euclidean	0.7794
pearson_dot	0.5202
spearman_dot	0.5115
pearson_max	0.7645
spearman_max	0.7872

Semantic Similarity

• Dataset: sts-test-256
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8124
spearman_cosine	0.8211
pearson_manhattan	0.7835
spearman_manhattan	0.7822
pearson_euclidean	0.7852
spearman_euclidean	0.784
pearson_dot	0.5917
spearman_dot	0.5785
pearson_max	0.8124
spearman_max	0.8211

Semantic Similarity

• Dataset: sts-test-128
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8079
spearman_cosine	0.819
pearson_manhattan	0.7795
spearman_manhattan	0.7786
pearson_euclidean	0.7813
spearman_euclidean	0.7813
pearson_dot	0.5714
spearman_dot	0.5602
pearson_max	0.8079
spearman_max	0.819

Semantic Similarity

• Dataset: sts-test-64
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.7988
spearman_cosine	0.8129
pearson_manhattan	0.7728
spearman_manhattan	0.7728
pearson_euclidean	0.7735
spearman_euclidean	0.7751
pearson_dot	0.5397
spearman_dot	0.5279
pearson_max	0.7988
spearman_max	0.8129

Semantic Similarity

• Dataset: sts-test-32
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.772
spearman_cosine	0.7936
pearson_manhattan	0.7561
spearman_manhattan	0.7597
pearson_euclidean	0.7581
spearman_euclidean	0.7628
pearson_dot	0.489
spearman_dot	0.4779
pearson_max	0.772
spearman_max	0.7936

Semantic Similarity

• Dataset: sts-test-16
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.7138
spearman_cosine	0.7486
pearson_manhattan	0.7254
spearman_manhattan	0.7339
pearson_euclidean	0.7274
spearman_euclidean	0.7382
pearson_dot	0.3856
spearman_dot	0.3749
pearson_max	0.7274
spearman_max	0.7486

Training Details

Training Dataset

sentence-transformers/all-nli

• Dataset: sentence-transformers/all-nli at 65dd388
• Size: 557,850 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 7 tokens mean: 10.38 tokens max: 45 tokens	min: 6 tokens mean: 12.8 tokens max: 39 tokens	min: 6 tokens mean: 13.4 tokens max: 50 tokens

• Samples:

  anchor	positive	negative
  A person on a horse jumps over a broken down airplane.	A person is outdoors, on a horse.	A person is at a diner, ordering an omelette.
  Children smiling and waving at camera	There are children present	The kids are frowning
  A boy is jumping on skateboard in the middle of a red bridge.	The boy does a skateboarding trick.	The boy skates down the sidewalk.

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          256,
          128,
          64,
          32,
          16
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Evaluation Dataset

sentence-transformers/stsb

• Dataset: sentence-transformers/stsb at ab7a5ac
• Size: 1,500 evaluation samples
• Columns: sentence1, sentence2, and score
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	score
  type	string	string	float
  details	min: 5 tokens mean: 15.0 tokens max: 44 tokens	min: 6 tokens mean: 14.99 tokens max: 61 tokens	min: 0.0 mean: 0.47 max: 1.0

• Samples:

  sentence1	sentence2	score
  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: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          256,
          128,
          64,
          32,
          16
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• num_train_epochs: 1
• warmup_ratio: 0.1
• fp16: True
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: False
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_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: 1.0
• 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
• 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: True
• 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: None
• 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_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	sts-dev-128_spearman_cosine	sts-dev-16_spearman_cosine	sts-dev-256_spearman_cosine	sts-dev-32_spearman_cosine	sts-dev-64_spearman_cosine	sts-test-128_spearman_cosine	sts-test-16_spearman_cosine	sts-test-256_spearman_cosine	sts-test-32_spearman_cosine	sts-test-64_spearman_cosine
0.0229	100	21.0363	14.2448	0.7856	0.7417	0.7873	0.7751	0.7846	-	-	-	-	-
0.0459	200	11.1093	13.4736	0.7877	0.7298	0.7861	0.7687	0.7798	-	-	-	-	-
0.0688	300	10.1847	13.7191	0.7877	0.7284	0.7898	0.7617	0.7755	-	-	-	-	-
0.0918	400	9.356	13.2955	0.7906	0.7385	0.7914	0.7715	0.7799	-	-	-	-	-
0.1147	500	8.9318	12.8099	0.7889	0.7346	0.7910	0.7690	0.7801	-	-	-	-	-
0.1376	600	8.5293	13.7384	0.7814	0.7362	0.7866	0.7656	0.7736	-	-	-	-	-
0.1606	700	8.7589	13.4466	0.7899	0.7467	0.7945	0.7770	0.7847	-	-	-	-	-
0.1835	800	7.7941	13.6734	0.7960	0.7526	0.7986	0.7800	0.7894	-	-	-	-	-
0.2065	900	7.9183	12.9082	0.7885	0.7470	0.7966	0.7705	0.7803	-	-	-	-	-
0.2294	1000	7.3669	13.2827	0.7751	0.7181	0.7822	0.7557	0.7675	-	-	-	-	-
0.2524	1100	7.6205	13.0227	0.7875	0.7373	0.7914	0.7730	0.7828	-	-	-	-	-
0.2753	1200	7.4308	13.4980	0.7844	0.7373	0.7890	0.7709	0.7755	-	-	-	-	-
0.2982	1300	7.3625	12.8380	0.7984	0.7520	0.8032	0.7824	0.7915	-	-	-	-	-
0.3212	1400	6.9421	12.7016	0.7912	0.7358	0.7960	0.7749	0.7850	-	-	-	-	-
0.3441	1500	7.0635	13.2198	0.8018	0.7578	0.8070	0.7861	0.7961	-	-	-	-	-
0.3671	1600	6.6682	13.3225	0.7906	0.7522	0.7944	0.7763	0.7849	-	-	-	-	-
0.3900	1700	6.42	12.7381	0.7984	0.7449	0.8021	0.7806	0.7911	-	-	-	-	-
0.4129	1800	6.659	13.0247	0.7947	0.7461	0.8002	0.7808	0.7876	-	-	-	-	-
0.4359	1900	6.1664	12.6814	0.7893	0.7312	0.7959	0.7700	0.7807	-	-	-	-	-
0.4588	2000	6.392	13.0238	0.7935	0.7354	0.7987	0.7758	0.7860	-	-	-	-	-
0.4818	2100	6.177	12.8833	0.7891	0.7428	0.7924	0.7723	0.7801	-	-	-	-	-
0.5047	2200	6.0411	12.5269	0.7836	0.7400	0.7875	0.7664	0.7765	-	-	-	-	-
0.5276	2300	6.1506	13.4349	0.7741	0.7350	0.7803	0.7556	0.7634	-	-	-	-	-
0.5506	2400	6.109	12.6996	0.7808	0.7326	0.7860	0.7663	0.7735	-	-	-	-	-
0.5735	2500	6.2849	13.2831	0.7874	0.7365	0.7932	0.7727	0.7794	-	-	-	-	-
0.5965	2600	6.0658	12.9425	0.7988	0.7481	0.8042	0.7818	0.7889	-	-	-	-	-
0.6194	2700	6.0646	13.0144	0.7965	0.7509	0.8010	0.7800	0.7875	-	-	-	-	-
0.6423	2800	6.0795	12.7602	0.7912	0.7472	0.7937	0.7778	0.7818	-	-	-	-	-
0.6653	2900	6.2407	13.2381	0.7829	0.7381	0.7873	0.7664	0.7765	-	-	-	-	-
0.6882	3000	6.1872	12.9064	0.7942	0.7516	0.7965	0.7793	0.7857	-	-	-	-	-
0.7112	3100	5.8987	12.9323	0.8065	0.7585	0.8087	0.7909	0.7989	-	-	-	-	-
0.7341	3200	5.996	13.1017	0.7971	0.7566	0.8005	0.7811	0.7889	-	-	-	-	-
0.7571	3300	5.3748	12.7601	0.8398	0.7881	0.8441	0.8232	0.8337	-	-	-	-	-
0.7800	3400	4.0798	12.7221	0.8400	0.7908	0.8440	0.8255	0.8342	-	-	-	-	-
0.8029	3500	3.6024	12.5445	0.8408	0.7892	0.8447	0.8247	0.8347	-	-	-	-	-
0.8259	3600	3.4619	12.6025	0.8405	0.7883	0.8442	0.8255	0.8347	-	-	-	-	-
0.8488	3700	3.2288	12.6636	0.8388	0.7872	0.8433	0.8226	0.8330	-	-	-	-	-
0.8718	3800	3.0543	12.6475	0.8386	0.7834	0.8427	0.8229	0.8330	-	-	-	-	-
0.8947	3900	3.0368	12.5390	0.8407	0.7845	0.8444	0.8227	0.8346	-	-	-	-	-
0.9176	4000	2.9591	12.5709	0.8419	0.7864	0.8456	0.8245	0.8359	-	-	-	-	-
0.9406	4100	2.944	12.6029	0.8415	0.7868	0.8452	0.8245	0.8359	-	-	-	-	-
0.9635	4200	2.9032	12.5514	0.8423	0.7888	0.8455	0.8254	0.8363	-	-	-	-	-
0.9865	4300	2.838	12.6054	0.8416	0.7872	0.8450	0.8244	0.8356	-	-	-	-	-
1.0	4359	-	-	-	-	-	-	-	0.8190	0.7486	0.8211	0.7936	0.8129

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 0.244 kWh
• Carbon Emitted: 0.095 kg of CO2
• Hours Used: 0.923 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: 3.0.0.dev0
• Transformers: 4.41.0.dev0
• PyTorch: 2.3.0+cu121
• Accelerate: 0.26.1
• Datasets: 2.18.0
• 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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning}, 
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

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}
}