SentenceTransformer based on Sami92/multilingual-e5-large-instruct-eu-parl-de-v2
This is a sentence-transformers model finetuned from Sami92/multilingual-e5-large-instruct-eu-parl-de-v2. 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.
Model Details
Model Description
Model Sources
Full Model Architecture
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})
)
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
model = SentenceTransformer("Sami92/claim-matching-multiling-e5-large-instruct-eu-parl-de-v2")
sentences = [
'Häuser und Wohnungen sind deutlich günstiger geworden. Doch einen Kollaps der Immobilienpreise erwartet der Chef der Förderbank KfW nicht. Ihm macht etwas anderes Sorgen.',
'OMG, Häuser und Wohnungen sind soooo viel günstiger jetzt! 😱 Aber der Chef der KfW glaubt nicht, dass die Preise total abstürzen werden. #WorriedAboutSomethingElse',
'OMG, habt ihr schon das neue Video von Lisa gesehen? 😂🤣 Es ist einfach zu gut! #MustWatch #EpicFail Ich kann nicht glauben, wie sie es geschafft hat, das zu filmen! 🤳🎥 Die Kommentare sind auch der Hammer, Leute! 💬💯 Schaut es euch unbedingt an und lasst ein Like da! 👍❤️',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
Evaluation
Metrics
Binary Classification
Metric |
Value |
cosine_accuracy |
0.948 |
cosine_accuracy_threshold |
0.8376 |
cosine_f1 |
0.9491 |
cosine_f1_threshold |
0.8327 |
cosine_precision |
0.9291 |
cosine_recall |
0.9699 |
cosine_ap |
0.9841 |
dot_accuracy |
0.9321 |
dot_accuracy_threshold |
429.1976 |
dot_f1 |
0.9351 |
dot_f1_threshold |
427.4893 |
dot_precision |
0.8945 |
dot_recall |
0.9795 |
dot_ap |
0.9622 |
manhattan_accuracy |
0.9486 |
manhattan_accuracy_threshold |
337.3625 |
manhattan_f1 |
0.9495 |
manhattan_f1_threshold |
342.0415 |
manhattan_precision |
0.9303 |
manhattan_recall |
0.9696 |
manhattan_ap |
0.984 |
euclidean_accuracy |
0.9487 |
euclidean_accuracy_threshold |
13.4905 |
euclidean_f1 |
0.9498 |
euclidean_f1_threshold |
13.5163 |
euclidean_precision |
0.9297 |
euclidean_recall |
0.9708 |
euclidean_ap |
0.9842 |
max_accuracy |
0.9487 |
max_accuracy_threshold |
429.1976 |
max_f1 |
0.9498 |
max_f1_threshold |
427.4893 |
max_precision |
0.9303 |
max_recall |
0.9795 |
max_ap |
0.9842 |
Triplet
Metric |
Value |
cosine_accuracy |
0.9969 |
dot_accuracy |
0.0049 |
manhattan_accuracy |
0.9965 |
euclidean_accuracy |
0.9968 |
max_accuracy |
0.9969 |
Binary Classification
Metric |
Value |
cosine_accuracy |
0.7816 |
cosine_accuracy_threshold |
0.8564 |
cosine_f1 |
0.5625 |
cosine_f1_threshold |
0.8259 |
cosine_precision |
0.675 |
cosine_recall |
0.4821 |
cosine_ap |
0.6249 |
dot_accuracy |
0.7471 |
dot_accuracy_threshold |
496.7446 |
dot_f1 |
0.5333 |
dot_f1_threshold |
456.797 |
dot_precision |
0.5714 |
dot_recall |
0.5 |
dot_ap |
0.5831 |
manhattan_accuracy |
0.7759 |
manhattan_accuracy_threshold |
321.3442 |
manhattan_f1 |
0.5556 |
manhattan_f1_threshold |
343.157 |
manhattan_precision |
0.7353 |
manhattan_recall |
0.4464 |
manhattan_ap |
0.6236 |
euclidean_accuracy |
0.7759 |
euclidean_accuracy_threshold |
13.2242 |
euclidean_f1 |
0.5556 |
euclidean_f1_threshold |
13.4257 |
euclidean_precision |
0.7353 |
euclidean_recall |
0.4464 |
euclidean_ap |
0.6209 |
max_accuracy |
0.7816 |
max_accuracy_threshold |
496.7446 |
max_f1 |
0.5625 |
max_f1_threshold |
456.797 |
max_precision |
0.7353 |
max_recall |
0.5 |
max_ap |
0.6249 |
Information Retrieval
Metric |
Value |
cosine_accuracy@1 |
1.0 |
cosine_accuracy@3 |
1.0 |
cosine_accuracy@5 |
1.0 |
cosine_accuracy@10 |
1.0 |
cosine_precision@1 |
1.0 |
cosine_precision@3 |
0.63 |
cosine_precision@5 |
0.402 |
cosine_precision@10 |
0.21 |
cosine_recall@1 |
0.4825 |
cosine_recall@3 |
0.8992 |
cosine_recall@5 |
0.9492 |
cosine_recall@10 |
0.99 |
cosine_ndcg@10 |
0.9518 |
cosine_mrr@10 |
1.0 |
cosine_map@100 |
0.9077 |
dot_accuracy@1 |
1.0 |
dot_accuracy@3 |
1.0 |
dot_accuracy@5 |
1.0 |
dot_accuracy@10 |
1.0 |
dot_precision@1 |
1.0 |
dot_precision@3 |
0.6067 |
dot_precision@5 |
0.394 |
dot_precision@10 |
0.207 |
dot_recall@1 |
0.4825 |
dot_recall@3 |
0.8667 |
dot_recall@5 |
0.9333 |
dot_recall@10 |
0.9775 |
dot_ndcg@10 |
0.9396 |
dot_mrr@10 |
1.0 |
dot_map@100 |
0.8919 |
Training Details
Training Dataset
Unnamed Dataset
Evaluation Dataset
Unnamed Dataset
Training Hyperparameters
Non-Default Hyperparameters
eval_strategy
: steps
per_device_train_batch_size
: 128
per_device_eval_batch_size
: 128
gradient_accumulation_steps
: 8
learning_rate
: 0.0001
num_train_epochs
: 1
fp16
: True
load_best_model_at_end
: True
push_to_hub
: True
hub_model_id
: Sami92/claim-matching-multiling-e5-large-instruct-eu-parl-de-v2
gradient_checkpointing
: True
push_to_hub_model_id
: claim-matching-multiling-e5-large-instruct-eu-parl-de-v2
All Hyperparameters
Click to expand
overwrite_output_dir
: False
do_predict
: False
eval_strategy
: steps
prediction_loss_only
: True
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
: 8
eval_accumulation_steps
: None
learning_rate
: 0.0001
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
: 0.0
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
: 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
: 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
: Sami92/claim-matching-multiling-e5-large-instruct-eu-parl-de-v2
hub_strategy
: every_save
hub_private_repo
: False
hub_always_push
: False
gradient_checkpointing
: True
gradient_checkpointing_kwargs
: None
include_inputs_for_metrics
: False
eval_do_concat_batches
: True
fp16_backend
: auto
push_to_hub_model_id
: claim-matching-multiling-e5-large-instruct-eu-parl-de-v2
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 |
loss |
claim-matching-synthetic-binary_max_ap |
claim-matching-synthetic-triplet_max_accuracy |
fc-de-binary_max_ap |
fc-de-retrieval_dot_map@100 |
0 |
0 |
- |
0.9801 |
0.9962 |
0.6312 |
0.8854 |
0.1047 |
5 |
0.1165 |
0.9762 |
0.9940 |
0.6142 |
0.8898 |
0.2094 |
10 |
0.1113 |
0.9828 |
0.9966 |
0.6302 |
0.8912 |
0.3141 |
15 |
0.1112 |
0.9828 |
0.9967 |
0.6437 |
0.8923 |
0.4188 |
20 |
0.1015 |
0.9842 |
0.9969 |
0.6626 |
0.8872 |
0.5236 |
25 |
0.1043 |
0.9847 |
0.9970 |
0.6662 |
0.8968 |
0.6283 |
30 |
0.1001 |
0.9847 |
0.9970 |
0.6547 |
0.8970 |
0.7330 |
35 |
0.0949 |
0.9841 |
0.9969 |
0.6282 |
0.8858 |
0.8377 |
40 |
0.0965 |
0.9838 |
0.9967 |
0.6238 |
0.8894 |
0.9424 |
45 |
0.0988 |
0.9842 |
0.9969 |
0.6249 |
0.8919 |
Framework Versions
- Python: 3.10.12
- Sentence Transformers: 3.0.1
- Transformers: 4.41.2
- PyTorch: 2.3.1+cu121
- Accelerate: 0.32.0
- Datasets: 2.20.0
- 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",
}
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}
}