This is a sentence-transformers model finetuned from sentence-transformers/all-mpnet-base-v2. It maps sentences & paragraphs to a 768-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-mpnet-base-v2
Maximum Sequence Length: 384 tokens
Output Dimensionality: 768 tokens
Similarity Function: Cosine Similarity

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': 384, 'do_lower_case': False}) with Transformer model: MPNetModel 
  (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})
  (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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("s2593817/sft-sql-embedding")
# Run inference
sentences = [
    'SELECT alias1.col1 FROM table1 AS alias1 JOIN table2 AS alias2 ON alias1.col2 = alias2.col2 JOIN table3 AS alias3 ON alias2.col3 = alias3.col3 WHERE alias3.col4 = str INTERSECT SELECT alias1.col1 FROM table1 AS alias1 JOIN table2 AS alias2 ON alias1.col2 = alias2.col2 JOIN table3 AS alias3 ON alias2.col3 = alias3.col3 WHERE alias3.col4 = str',
    'SELECT count(col1) FROM table1 WHERE col2 = num',
    'SELECT count(DISTINCT col1) FROM table1',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Training Details

Training Dataset

Unnamed Dataset

Size: 300,000 training samples
Columns: sentence1, sentence2, and score
Approximate statistics based on the first 1000 samples:
sentence1 sentence2 score
type string string float
details
min: 8 tokens
mean: 38.49 tokens
max: 189 tokens
min: 7 tokens
mean: 37.44 tokens
max: 153 tokens
min: 0.04
mean: 0.36
max: 1.0

Samples:

sentence1	sentence2	score
`SELECT DISTINCT count(DISTINCT alias4.col1) , alias3.col2 FROM table1 AS alias1 JOIN table2 AS alias2 ON alias1.col3 = alias2.col3 JOIN table3 AS alias3 ON alias3.col4 = alias1.col4 JOIN table4 AS alias4 ON alias3.col4 = alias4.col5 WHERE alias2.col6 = str GROUP BY alias3.col2 ORDER BY count(DISTINCT alias4.col1) DESC`	`SELECT count(*) FROM table1 WHERE col1 = str`	`0.14221014492753623`
`SELECT DISTINCT count(alias2.col1) FROM table1 AS alias1 JOIN table2 AS alias2 ON alias1.col2 = alias2.col2 WHERE alias1.col3 = str`	`SELECT alias3.col1 FROM table1 AS alias1 JOIN table2 AS alias2 ON alias1.col2 = alias2.col2 JOIN table3 AS alias3 ON alias2.col3 = alias3.col3 WHERE alias1.col4 = str AND alias1.col5 = str`	`0.5468686868686868`
`SELECT count(*) FROM table1`	`SELECT count(*) FROM table1 WHERE col1 LIKE str`	`0.6269230769230769`

Loss: CoSENTLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "pairwise_cos_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 160
learning_rate: 2e-05
num_train_epochs: 8
warmup_ratio: 0.2
fp16: True
dataloader_num_workers: 16
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 160
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: 8
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.2
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: 16
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: 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
eval_on_start: False
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss
0.0533	100	12.0379
0.1067	200	9.2042
0.16	300	8.6521
0.2133	400	8.5353
0.2667	500	8.4472
0.32	600	8.4105
0.3733	700	8.3927
0.4267	800	8.3553
0.48	900	8.3326
0.5333	1000	8.3168
0.5867	1100	8.2941
0.64	1200	6.0021
0.6933	1300	5.3802
0.7467	1400	5.3282
0.8	1500	5.2365
0.8533	1600	5.0198
0.9067	1700	4.899
0.96	1800	4.8887
1.0133	1900	4.7603
1.0667	2000	4.6292
1.12	2100	4.4811
1.1733	2200	4.2841
1.2267	2300	4.2251
1.28	2400	4.0261
1.3333	2500	3.8628
1.3867	2600	3.8404
1.44	2700	3.6471
1.4933	2800	3.6673
1.5467	2900	3.5626
1.6	3000	3.5391
1.6533	3100	3.5629
1.7067	3200	3.4787
1.76	3300	3.4401
1.8133	3400	3.491
1.8667	3500	3.3358
1.92	3600	3.3555
1.9733	3700	3.161
2.0267	3800	3.1708
2.08	3900	3.1678
2.1333	4000	3.1348
2.1867	4100	2.9159
2.24	4200	2.8359
2.2933	4300	2.8359
2.3467	4400	2.796
2.4	4500	2.8483
2.4533	4600	2.7774
2.5067	4700	2.7766
2.56	4800	2.7185
2.6133	4900	2.778
2.6667	5000	2.7114
2.72	5100	2.6623
2.7733	5200	2.5093
2.8267	5300	2.4835
2.88	5400	2.2851
2.9333	5500	2.1488
2.9867	5600	2.2175
3.04	5700	2.0813
3.0933	5800	2.1489
3.1467	5900	2.1337
3.2	6000	2.2258
3.2533	6100	2.1601
3.3067	6200	1.9266
3.36	6300	1.8427
3.4133	6400	1.8434
3.4667	6500	1.917
3.52	6600	1.8204
3.5733	6700	2.0209
3.6267	6800	1.7852
3.68	6900	1.9566
3.7333	7000	1.852
3.7867	7100	1.8562
3.84	7200	1.7595
3.8933	7300	1.4295
3.9467	7400	1.2669
4.0	7500	1.2029
4.0533	7600	1.3074
4.1067	7700	1.435
4.16	7800	1.5712
4.2133	7900	1.2366
4.2667	8000	1.526
4.32	8100	1.2565
4.3733	8200	1.4546
4.4267	8300	1.374
4.48	8400	1.3387
4.5333	8500	1.3776
4.5867	8600	1.3984
4.64	8700	1.3577
4.6933	8800	1.2393
4.7467	8900	1.4125
4.8	9000	1.6127
4.8533	9100	1.6897
4.9067	9200	1.1217
4.96	9300	1.406
5.0133	9400	1.4641
5.0667	9500	1.48
5.12	9600	1.3367
5.1733	9700	1.4681
5.2267	9800	1.4628
5.28	9900	1.32
5.3333	10000	1.448
5.3867	10100	1.2516
5.44	10200	1.4421
5.4933	10300	1.2542
5.5467	10400	1.4545
5.6	10500	1.1441
5.6533	10600	1.251
5.7067	10700	1.3396
5.76	10800	1.0305
5.8133	10900	1.0155
5.8667	11000	0.9871
5.92	11100	1.074
5.9733	11200	0.4534
6.0267	11300	0.1965
6.08	11400	0.1822
6.1333	11500	0.2101
6.1867	11600	0.2326
6.24	11700	0.4126
6.2933	11800	0.4871
6.3467	11900	0.2012
6.4	12000	0.2113
6.4533	12100	0.1788
6.5067	12200	0.2271
6.56	12300	0.1685
6.6133	12400	0.3347
6.6667	12500	0.123
6.72	12600	0.155
6.7733	12700	0.2476
6.8267	12800	0.1926
6.88	12900	0.1394
6.9333	13000	0.1683
6.9867	13100	0.2484
7.04	13200	0.1338
7.0933	13300	0.1568
7.1467	13400	0.1206
7.2	13500	0.1683
7.2533	13600	0.1831
7.3067	13700	0.3077
7.36	13800	0.3533
7.4133	13900	0.1165
7.4667	14000	0.2128
7.52	14100	0.236
7.5733	14200	0.3616
7.6267	14300	0.2989
7.68	14400	0.2416
7.7333	14500	0.2105
7.7867	14600	0.1575
7.84	14700	0.224
7.8933	14800	0.1593
7.9467	14900	0.1293
8.0	15000	0.0985

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.0.1
Transformers: 4.42.4
PyTorch: 2.3.1+cu121
Accelerate: 0.33.0
Datasets: 2.20.0
Tokenizers: 0.19.1

Citation

BibTeX

Sentence Transformers

bibtex

@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",
}

CoSENTLoss

bibtex

@online{kexuefm-8847,
    title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},
    author={Su Jianlin},
    year={2022},
    month={Jan},
    url={https://kexue.fm/archives/8847},
}

	sentence1	sentence2	score
type	string	string	float
details	min: 8 tokens mean: 38.49 tokens max: 189 tokens	min: 7 tokens mean: 37.44 tokens max: 153 tokens	min: 0.04 mean: 0.36 max: 1.0