SentenceTransformer based on sentence-transformers/multi-qa-MiniLM-L6-dot-v1

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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("Trelis/multi-qa-MiniLM-L6-dot-v1-ft-pairs-4-cst-epoch-s1")
# Run inference
sentences = [
    'What is the consequence if a defending team is penalized three times in their seven-meter zone during a single possession?',
    '18. 5 the mark must be indicated by the referee before a penalty tap is taken. 18. 6 the penalty tap must be performed without delay after the referee indicates the mark. ruling = a penalty to the non - offending team at the point of infringement. 18. 7 a player may perform a rollball instead of a penalty tap and the player who receives the ball does not become the half. 18. 8 if the defending team is penalised three ( 3 ) times upon entering their seven metre zone during a single possession, the last offending player will be given an exclusion until the end of that possession. 18. 9 a penalty try is awarded if any action by a player, team official or spectator, deemed by the referee to be contrary to the rules or spirit of the game clearly prevents the attacking team from scoring a try. fit playing rules - 5th edition copyright © touch football australia 2020 15 19 advantage 19. 1 where a defending team player is offside at a tap or rollball and attempts to interfere with play, the referee will allow advantage or award a penalty, whichever is of greater advantage to the attacking team.',
    '5th edition rules touch football tion rules touch football touch football australia ( tfa ) undertook an extensive internal review of their domestic playing rules throughout 2018 and 2019. the review was led by an vastly experienced group of current and past players, coaches, referees and administrators of the sport from community competitions to the elite international game. this group consulted broadly within the australian community to develop a set of playing rules that could be applied across all levels of the sport. the result was the tfa 8th edition playing rules. at the federation of international touch paris convention held in october 2019 touch football australia presented the tfa 8th edition playing rules and subsequently offered fit and all national touch associations ( ntas ) royalty free rights to use the newly developed rules. consequently, the fit board resolved to adopt the tfa 8th edition playing rules as the 5th edition fit playing rules to be used across all levels of the game internationally. fit and its members acknowledge and thank touch football australia for the rights to use these rules. whilst consistency in the application of the rules of the game is important, fit encourages its members to offer features in local competition rules to ensure that all participants enjoy a high quality experience.',
]
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
0.3333	2	1.7279	-
0.5	3	-	1.3621
0.6667	4	1.4819	-
1.0	6	1.5272	1.2755
1.3333	8	1.2528	-
1.5	9	-	1.2600
1.6667	10	1.421	-
2.0	12	1.1836	1.2422
2.3333	14	1.2527	-
2.5	15	-	1.2317
2.6667	16	1.485	-
3.0	18	0.8239	1.1883
3.3333	20	1.1028	-
3.5	21	-	1.1533
3.6667	22	0.9746	-
4.0	24	0.816	1.1237

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

@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}
}

Trelis
/

multi-qa-MiniLM-L6-dot-v1-ft-pairs-4-cst-epoch-s1