Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
Paper
•
1908.10084
•
Published
•
9
This is a sentence-transformers model finetuned from intfloat/multilingual-e5-large-instruct. It maps sentences & paragraphs to a 1024-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.
SentenceTransformer(
(0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: XLMRobertaModel
(1): Pooling({'word_embedding_dimension': 1024, '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()
)
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("lg-software/multilingual-e5-large-instruct-ft")
# Run inference
sentences = [
'What must you do first when checking the connector?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]
# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]
InformationRetrievalEvaluator| Metric | Value |
|---|---|
| cosine_accuracy@1 | 0.1491 |
| cosine_accuracy@3 | 0.2481 |
| cosine_accuracy@5 | 0.2986 |
| cosine_accuracy@10 | 0.3725 |
| cosine_precision@1 | 0.1491 |
| cosine_precision@3 | 0.0827 |
| cosine_precision@5 | 0.0597 |
| cosine_precision@10 | 0.0373 |
| cosine_recall@1 | 0.1491 |
| cosine_recall@3 | 0.2481 |
| cosine_recall@5 | 0.2986 |
| cosine_recall@10 | 0.3725 |
| cosine_ndcg@10 | 0.2511 |
| cosine_mrr@10 | 0.2134 |
| cosine_map@100 | 0.2241 |
sentence_0 and sentence_1| sentence_0 | sentence_1 | |
|---|---|---|
| type | string | string |
| details |
|
|
| sentence_0 | sentence_1 |
|---|---|
What does ESM stand for? |
C11 2008年01月 (平成20年01月) ESM --- --- ティーダ / ティーダ ラティオ (C11) - 2008年01月 |
How many steps are in the connector inspection process? |
C11 2008年01月 (平成20年01月) ESM --- --- ティーダ / ティーダ ラティオ (C11) - 2008年01月 |
What does the first step involve? |
C11 2008年01月 (平成20年01月) ESM --- --- ティーダ / ティーダ ラティオ (C11) - 2008年01月 |
MultipleNegativesRankingLoss with these parameters:{
"scale": 20.0,
"similarity_fct": "cos_sim"
}
eval_strategy: stepsper_device_train_batch_size: 10per_device_eval_batch_size: 10num_train_epochs: 2multi_dataset_batch_sampler: round_robinoverwrite_output_dir: Falsedo_predict: Falseeval_strategy: stepsprediction_loss_only: Trueper_device_train_batch_size: 10per_device_eval_batch_size: 10per_gpu_train_batch_size: Noneper_gpu_eval_batch_size: Nonegradient_accumulation_steps: 1eval_accumulation_steps: Nonetorch_empty_cache_steps: Nonelearning_rate: 5e-05weight_decay: 0.0adam_beta1: 0.9adam_beta2: 0.999adam_epsilon: 1e-08max_grad_norm: 1num_train_epochs: 2max_steps: -1lr_scheduler_type: linearlr_scheduler_kwargs: {}warmup_ratio: 0.0warmup_steps: 0log_level: passivelog_level_replica: warninglog_on_each_node: Truelogging_nan_inf_filter: Truesave_safetensors: Truesave_on_each_node: Falsesave_only_model: Falserestore_callback_states_from_checkpoint: Falseno_cuda: Falseuse_cpu: Falseuse_mps_device: Falseseed: 42data_seed: Nonejit_mode_eval: Falseuse_ipex: Falsebf16: Falsefp16: Falsefp16_opt_level: O1half_precision_backend: autobf16_full_eval: Falsefp16_full_eval: Falsetf32: Nonelocal_rank: 0ddp_backend: Nonetpu_num_cores: Nonetpu_metrics_debug: Falsedebug: []dataloader_drop_last: Falsedataloader_num_workers: 0dataloader_prefetch_factor: Nonepast_index: -1disable_tqdm: Falseremove_unused_columns: Truelabel_names: Noneload_best_model_at_end: Falseignore_data_skip: Falsefsdp: []fsdp_min_num_params: 0fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}fsdp_transformer_layer_cls_to_wrap: Noneaccelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}deepspeed: Nonelabel_smoothing_factor: 0.0optim: adamw_torchoptim_args: Noneadafactor: Falsegroup_by_length: Falselength_column_name: lengthddp_find_unused_parameters: Noneddp_bucket_cap_mb: Noneddp_broadcast_buffers: Falsedataloader_pin_memory: Truedataloader_persistent_workers: Falseskip_memory_metrics: Trueuse_legacy_prediction_loop: Falsepush_to_hub: Falseresume_from_checkpoint: Nonehub_model_id: Nonehub_strategy: every_savehub_private_repo: Nonehub_always_push: Falsegradient_checkpointing: Falsegradient_checkpointing_kwargs: Noneinclude_inputs_for_metrics: Falseinclude_for_metrics: []eval_do_concat_batches: Truefp16_backend: autopush_to_hub_model_id: Nonepush_to_hub_organization: Nonemp_parameters: auto_find_batch_size: Falsefull_determinism: Falsetorchdynamo: Noneray_scope: lastddp_timeout: 1800torch_compile: Falsetorch_compile_backend: Nonetorch_compile_mode: Nonedispatch_batches: Nonesplit_batches: Noneinclude_tokens_per_second: Falseinclude_num_input_tokens_seen: Falseneftune_noise_alpha: Noneoptim_target_modules: Nonebatch_eval_metrics: Falseeval_on_start: Falseuse_liger_kernel: Falseeval_use_gather_object: Falseaverage_tokens_across_devices: Falseprompts: Nonebatch_sampler: batch_samplermulti_dataset_batch_sampler: round_robin| Epoch | Step | Training Loss | cosine_ndcg@10 |
|---|---|---|---|
| 0.0628 | 500 | 0.9035 | 0.1946 |
| 0.1256 | 1000 | 0.4735 | 0.1945 |
| 0.1885 | 1500 | 0.4639 | 0.1870 |
| 0.2513 | 2000 | 0.4581 | 0.1974 |
| 0.3141 | 2500 | 0.4027 | 0.1904 |
| 0.3769 | 3000 | 0.3761 | 0.1848 |
| 0.4398 | 3500 | 0.3527 | 0.2005 |
| 0.5026 | 4000 | 0.3654 | 0.1995 |
| 0.5654 | 4500 | 0.3344 | 0.2045 |
| 0.6282 | 5000 | 0.3373 | 0.2114 |
| 0.6910 | 5500 | 0.3258 | 0.2033 |
| 0.7539 | 6000 | 0.3086 | 0.2145 |
| 0.8167 | 6500 | 0.3046 | 0.2246 |
| 0.8795 | 7000 | 0.2963 | 0.2194 |
| 0.9423 | 7500 | 0.2846 | 0.2225 |
| 1.0 | 7959 | - | 0.2187 |
| 1.0052 | 8000 | 0.2965 | 0.2263 |
| 1.0680 | 8500 | 0.2417 | 0.2321 |
| 1.1308 | 9000 | 0.228 | 0.2247 |
| 1.1936 | 9500 | 0.2298 | 0.2268 |
| 1.2564 | 10000 | 0.2494 | 0.2303 |
| 1.3193 | 10500 | 0.2105 | 0.2333 |
| 1.3821 | 11000 | 0.2139 | 0.2278 |
| 1.4449 | 11500 | 0.2064 | 0.2412 |
| 1.5077 | 12000 | 0.2052 | 0.2413 |
| 1.5705 | 12500 | 0.2002 | 0.2445 |
| 1.6334 | 13000 | 0.216 | 0.2451 |
| 1.6962 | 13500 | 0.2066 | 0.2495 |
| 1.7590 | 14000 | 0.2095 | 0.2478 |
| 1.8218 | 14500 | 0.2062 | 0.2492 |
| 1.8847 | 15000 | 0.203 | 0.2507 |
| 1.9475 | 15500 | 0.1793 | 0.2504 |
| 2.0 | 15918 | - | 0.2511 |
@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",
}
@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}
}
Base model
intfloat/multilingual-e5-large-instruct