You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

SentenceTransformer based on BAAI/bge-m3

This is a sentence-transformers model finetuned from BAAI/bge-m3. 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 Type: Sentence Transformer
Base model: BAAI/bge-m3
Maximum Sequence Length: 64 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
HF中国镜像站: Sentence Transformers on HF中国镜像站

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 64, '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

# Download from the 🤗 Hub
model = SentenceTransformer("disi-unibo-nlp/foodex-baseterm-retriever")
# Run inference
sentences = [
    'sauce cream salad dressing based facets desc food production commercial brandname productname known',
    'The group includes any type of Salad dressing. The part consumed/analysed is by default the whole marketed unit or a homogeneous representative portion.',
    'The group includes any type of Seasonings and extracts. The part consumed/analysed is by default the whole marketed unit or a representative portion of it.',
]
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]

Evaluation

Metrics

Device Aware Information Retrieval

Evaluated with src.utils.eval_functions.DeviceAwareInformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.9617
cosine_accuracy@3	0.9974
cosine_accuracy@5	0.9993
cosine_accuracy@10	1.0
cosine_precision@1	0.9617
cosine_precision@3	0.3347
cosine_precision@5	0.2012
cosine_precision@10	0.1008
cosine_recall@1	0.9586
cosine_recall@3	0.997
cosine_recall@5	0.9988
cosine_recall@10	1.0
cosine_ndcg@10	0.9846
cosine_mrr@10	0.9793
cosine_map@100	0.9791

Training Details

Training Dataset

Unnamed Dataset

Size: 553,000 training samples
Columns: sentence_0, sentence_1, and sentence_2

Approximate statistics based on the first 1000 samples:

	sentence_0	sentence_1	sentence_2
type	string	string	string
details	min: 3 tokens mean: 16.88 tokens max: 57 tokens	min: 11 tokens mean: 55.65 tokens max: 64 tokens	min: 12 tokens mean: 53.82 tokens max: 64 tokens

Samples:

sentence_0	sentence_1	sentence_2
`porridge made with ssmilk water`	`The group includes any type of Oat porridge in dry form to be diluted with milk or water. The term refers to e.g. ground oatmeal or rolled oat intended to be used for making oat porridge. The part consumed/analysed is by default the whole or a portion of it representing the observed heterogeneity.`	`The group includes any type of Oat flour (finely milled grains with particles not easy to distinguish). Different grades of refinement and types are all included in this group. The part consumed/analysed is by default the whole or a portion of it representing the observed heterogeneity.`
`pepper sweet red raw unprocessed fresh no treatment brand product name with skin peel crust bought chilled fresh department 2 8 c`	Fruiting vegetables from the plant classified under the species Capsicum annuum var. grossum (L.) Sendtner or Capsicum annuum var. longum Bailey, commonly known as Sweet peppers or Bell peppers or Paprika or PeppersLong or Pimento or Pimiento. The part consumed/analysed is not specified. When relevant, information on the part consumed/analysed has to be reported with additional facet descriptors. In case of data collections related to legislations, the default part consumed/analysed is the one defined in the applicable legislation.	`Spices from the fruits of the plant classified under the species Piper guineense Thonn., commonly known as West African pepper fruit. The part consumed/analysed is not specified. When relevant, information on the part consumed/analysed has to be reported with additional facet descriptors. In case of data collections related to legislations, the default part consumed/analysed is the one defined in the applicable legislation.`
`yeasted wheat bread with sourmilk sliced`	`The group includes any type of bread and rolls made with wheat flour containing high proportion of bran or wholemeal (brown or wholemeal wheat flour). The part consumed/analysed is by default the whole or a portion of it representing the observed heterogeneity.`	`The group includes any type of bread and rolls made with wheat flour containing moderate amounts of bran (semi-brown wheat flour). The part consumed/analysed is by default the whole or a portion of it representing the observed heterogeneity.`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
num_train_epochs: 5
fp16: 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
torch_empty_cache_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: 5
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: 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: 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: None
hub_always_push: False
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
include_inputs_for_metrics: False
include_for_metrics: []
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
eval_on_start: False
use_liger_kernel: False
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin

Training Logs

Epoch	Step	Training Loss	cosine_ndcg@10
0	0	-	0.2918
0.0579	500	1.2336	0.7926
0.1157	1000	0.6594	0.8605
0.1736	1500	0.5586	0.8827
0.2315	2000	0.5041	0.8998
0.2893	2500	0.4777	0.9075
0.3472	3000	0.4616	0.9172
0.4050	3500	0.4416	0.9309
0.4629	4000	0.4292	0.9279
0.5208	4500	0.4105	0.9375
0.5786	5000	0.4011	0.9384
0.6365	5500	0.4064	0.9491
0.6944	6000	0.3864	0.9504
0.7522	6500	0.3864	0.9502
0.8101	7000	0.3803	0.9546
0.8680	7500	0.3837	0.9571
0.9258	8000	0.3775	0.9606
0.9837	8500	0.3728	0.9629
1.0	8641	-	0.9621
1.0415	9000	0.3612	0.9612
1.0994	9500	0.3657	0.9650
1.1573	10000	0.3613	0.9659
1.2151	10500	0.3556	0.9630
1.2730	11000	0.3657	0.9655
1.3309	11500	0.3645	0.9697
1.3887	12000	0.351	0.9702
1.4466	12500	0.3533	0.9702
1.5045	13000	0.3505	0.9723
1.5623	13500	0.3444	0.9713
1.6202	14000	0.3517	0.9725
1.6780	14500	0.3535	0.9735
1.7359	15000	0.353	0.9726
1.7938	15500	0.3444	0.9740
1.8516	16000	0.3455	0.9785
1.9095	16500	0.3459	0.9763
1.9674	17000	0.3494	0.9787
2.0	17282	-	0.9790
2.0252	17500	0.3487	0.9794
2.0831	18000	0.3371	0.9761
2.1410	18500	0.3315	0.9788
2.1988	19000	0.3352	0.9785
2.2567	19500	0.3396	0.9763
2.3145	20000	0.3356	0.9776
2.3724	20500	0.3382	0.9811
2.4303	21000	0.34	0.9805
2.4881	21500	0.3309	0.9802
2.5460	22000	0.3353	0.9797
2.6039	22500	0.3423	0.9798
2.6617	23000	0.3289	0.9809
2.7196	23500	0.3333	0.9803
2.7775	24000	0.338	0.9815
2.8353	24500	0.336	0.9816
2.8932	25000	0.3346	0.9813
2.9510	25500	0.3311	0.9807
3.0	25923	-	0.9819
3.0089	26000	0.3302	0.9824
3.0668	26500	0.3275	0.9833
3.1246	27000	0.3331	0.9840
3.1825	27500	0.3231	0.9839
3.2404	28000	0.3308	0.9839
3.2982	28500	0.3259	0.9836
3.3561	29000	0.3252	0.9838
3.4140	29500	0.315	0.9847
3.4718	30000	0.322	0.9829
3.5297	30500	0.3323	0.9837
3.5875	31000	0.3318	0.9833
3.6454	31500	0.3307	0.9842
3.7033	32000	0.331	0.9841
3.7611	32500	0.3209	0.9849
3.8190	33000	0.3267	0.9841
3.8769	33500	0.3214	0.9846
3.9347	34000	0.3232	0.9847
3.9926	34500	0.3291	0.9848
4.0	34564	-	0.9844
4.0505	35000	0.3257	0.9843
4.1083	35500	0.3237	0.9841
4.1662	36000	0.3177	0.9837
4.2240	36500	0.3314	0.9838
4.2819	37000	0.3266	0.9842
4.3398	37500	0.3184	0.9841
4.3976	38000	0.3162	0.9844
4.4555	38500	0.3164	0.9850
4.5134	39000	0.3209	0.9849
4.5712	39500	0.3292	0.9850
4.6291	40000	0.3194	0.9850
4.6870	40500	0.3212	0.9850
4.7448	41000	0.3359	0.9849
4.8027	41500	0.3199	0.9848
4.8605	42000	0.3257	0.9847
4.9184	42500	0.3172	0.9846
4.9763	43000	0.324	0.9846
5.0	43205	-	0.9846

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.4.1
Transformers: 4.49.0
PyTorch: 2.6.0+cu124
Accelerate: 1.4.0
Datasets: 3.3.1
Tokenizers: 0.21.0

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}
}