Instructions to use Trelis/ms-marco-MiniLM-L-6-v2-2-constant-ep-MNRLpairs-2e-5-batch32-cuda-overlap with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Trelis/ms-marco-MiniLM-L-6-v2-2-constant-ep-MNRLpairs-2e-5-batch32-cuda-overlap with sentence-transformers:

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("Trelis/ms-marco-MiniLM-L-6-v2-2-constant-ep-MNRLpairs-2e-5-batch32-cuda-overlap")

sentences = [
"What happens if a player in possession enters the defending team's seven-metre zone?",
"10. 8 if a touch is made in the in - goal area before the ball is grounded, the player in possession is to perform a rollball seven ( 7 ) metres from the team ’ s attacking try line, provided it is not the sixth touch and the player is not half. 10. 9 if a player in possession is touched while on or behind their defending try line, the touch counts and once the referee sets the mark seven ( 7 ) metres directly forward of the contact point from the defending team ’ s try line, a rollball is performed. 10. 10 if a player in possession intentionally makes a touch on an offside defender who is making every effort to retire and remain out of play, the touch counts. fit playing rules - 5th edition copyright © touch football australia 2020 9 10. 11 if a touch is made on a player in possession while the player is juggling the ball in an attempt to maintain control of it, the touch counts if the attacking player following the touch retains possession.",
"9. 2 on the change of possession due to an intercept, the first touch will be zero ( 0 ) touch. 9. 3 following the sixth touch or a loss of possession due to any other means, the ball must be returned to the mark without delay. ruling = a deliberate delay in the changeover procedure will result in a penalty awarded to the non - offending team ten ( 10 ) metres forward of the mark for the change of possession. 9. 4 if the ball is dropped or passed and goes to ground during play, a change of possession results. ruling = the mark for the change of possession is where the ball makes initial contact with the ground. 9. 5 if the ball, while still under the control of the half, contacts the ground in the in - goal area, possession is lost. ruling = play will restart with a rollball at the nearest point on the seven ( 7 ) metre line. fit playing rules - 5th edition 8 copyright © touch football australia 2020 9. 6 if a player mishandles the ball and even if in an effort to gain control, the ball is accidentally knocked forward into any other player, a change of possession results.",
"fit playing rules - 5th edition copyright © touch football australia 2020 9 10. 11 if a touch is made on a player in possession while the player is juggling the ball in an attempt to maintain control of it, the touch counts if the attacking player following the touch retains possession. 10. 12 if a player in possession is touched and subsequently makes contact with either the sideline, a field marker or the ground outside the field of play, the touch counts and play continues with a rollball at the mark where the touch occurred. 10. 13 when a player from the defending team enters its defensive seven metre zone, the defending team must move forward at a reasonable pace until a touch is imminent or made. ruling = a penalty to the attacking team at the point of the infringement. 10. 14 when a player in possession enters the defending teams ’ seven metre zone the defending team is not obliged to move forward but cannot retire back towards their try line until a touch is imminent or made. ruling = a penalty to the attacking team at the seven ( 7 ) metre line in line with the point of the infringement."
]
embeddings = model.encode(sentences)

similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [4, 4]

Notebooks
Google Colab
Kaggle

SentenceTransformer based on cross-encoder/ms-marco-MiniLM-L-6-v2

This is a sentence-transformers model finetuned from cross-encoder/ms-marco-MiniLM-L-6-v2. 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: cross-encoder/ms-marco-MiniLM-L-6-v2
Maximum Sequence Length: 512 tokens
Output Dimensionality: 384 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': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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("Trelis/ms-marco-MiniLM-L-6-v2-2-constant-ep-MNRLpairs-2e-5-batch32-cuda-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: 2
lr_scheduler_type: constant
warmup_ratio: 0.3
bf16: True

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: 2
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: True
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	3.7086	3.1734
0.5714	4	3.4714	3.0742
0.8571	6	3.41	3.0204
1.1429	8	3.042	2.9657
1.4286	10	3.3335	2.9125
1.7143	12	3.2224	2.8573
2.0	14	2.9969	2.8300

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}
}