This is a sentence-transformers model finetuned from sentence-transformers/multi-qa-MiniLM-L6-dot-v1. It maps sentences & paragraphs to a 384-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/multi-qa-MiniLM-L6-dot-v1
Maximum Sequence Length: 512 tokens
Output Dimensionality: 384 tokens
Similarity Function: Dot Product

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: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, '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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("Trelis/multi-qa-MiniLM-L6-dot-v1-ft-pairs-4-cst-epoch-s1-overlap")
# Run inference
sentences = [
    'What happens if a team is leading at the end of the two-minute period of extra time?',
    '24. 1. 2 the drop - off commences with a tap from the centre of the halfway line by the team that did not commence the match with possession. 24. 1. 3 the drop - off will commence with a two ( 2 ) minute period of extra time. 24. 1. 4 should a team be leading at the expiration of the two ( 2 ) minute period of extra time then that team will be declared the winner and match complete. 24. 1. 5 should neither team be leading at the expiration of two ( 2 ) minutes, a signal is given and the match will pause at the next touch or dead ball. each team will then remove another player from the field of play. 24. 1. 6 the match will recommence immediately after the players have left the field at the same place where it paused ( i. e. the team retains possession at the designated number of touches, or at change of possession due to some infringement or the sixth touch ) and the match will continue until a try is scored. 24. 1. 7 there is no time off during the drop - off and the clock does not stop at the two ( 2 ) minute interval.',
    '7. 7 the tap to commence or recommence play must be performed without delay. ruling = a penalty to the non - offending team at the centre of the halfway line. 8 match duration 8. 1 a match is 40 minutes in duration, consisting of two ( 2 ) x 20 minute halves with a half time break. 8. 1. 1 there is no time off for injury during a match. 8. 2 local competition and tournament conditions may vary the duration of a match. 8. 3 when time expires, play is to continue until the next touch or dead ball and end of play is signaled by the referee. 8. 3. 1 should a penalty be awarded during this period, the penalty is to be taken. 8. 4 if a match is abandoned in any circumstances other than those referred to in clause 24. 1. 6 the nta or nta competition provider in its sole discretion shall determine the result of the match. 9 possession 9. 1 the team with the ball is entitled to six ( 6 ) touches prior to a change of possession. 9. 2 on the change of possession due to an intercept, the first touch will be zero ( 0 ) touch.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

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

Training Details

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: constant
warmup_ratio: 0.3

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
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: 4
max_steps: -1
lr_scheduler_type: constant
lr_scheduler_kwargs: {}
warmup_ratio: 0.3
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: False
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: 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: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss
0.2857	2	1.7892	-
0.5714	4	1.5998	1.3002
0.8571	6	1.5637	-
1.1429	8	1.3347	1.1748
1.4286	10	1.4256	-
1.7143	12	1.2205	1.1085
2.0	14	1.1307	-
2.2857	16	1.119	1.0558
2.5714	18	1.2639	-
2.8571	20	1.2834	1.0108
3.1429	22	0.9248	-
3.4286	24	1.1527	1.0074
3.7143	26	0.8702	-
4.0	28	0.725	1.0124

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.0.1
Transformers: 4.42.3
PyTorch: 2.1.1+cu121
Accelerate: 0.31.0
Datasets: 2.17.1
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",
}

MultipleNegativesRankingLoss

bibtex

@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}
}