This is a sentence-transformers model finetuned from nomic-ai/modernbert-embed-base on the touch-rugby-modernbert-pairs dataset. 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: nomic-ai/modernbert-embed-base
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- touch-rugby-modernbert-pairs

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': 8192, 'do_lower_case': False}) with Transformer model: ModernBertModel 
  (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("Trelis/modernbert-embed-base-touch-rugby-ft-v2")
# Run inference
sentences = [
    'Who indicates to commence play at the start of a Touch Rugby match?',
    '6.2\tThe Team coach(s) and Team officials may move from one position to the other \nbut shall do so without delay.While in a position at the end of the Field of Play, \nthe Team coach(s) or Team official must remain no closer than five (5) metres \nfrom the Dead Ball Line and must not coach or communicate (verbal or non-\nverbal) with either Team or the Referees.7\u2002 Commencement and Recommencement of Play  \n7.1\tTeam captains are to toss a coin in the presence of the Referee(s) with the \nwinning captain’s Team having the choice of the direction the Team wishes \nto run in the first half; the choice of Interchange Areas for the duration of the \nmatch, including any extra time; and the choice of which team will commence \nthe match in Possession.7.2\tA player of the Attacking Team is to commence the match with a Tap at the \ncentre of the Halfway Line following the indication to commence play from the \nReferee.',
    'See Appendix 1.Forced Interchange\nWhen a player is required to undertake a compulsory Interchange for \nan Infringement ruled more serious than a Penalty but less serious \nthan a Permanent Interchange, Sin Bin or Dismissal.Forward\nA position or direction towards the Dead Ball Line beyond the Team’s \nAttacking Try Line.Full Time\nThe expiration of the second period of time allowed for play.Half\nThe player who takes Possession following a Rollball.Half Time\nThe break in play between the two halves of a match.Imminent\nAbout to occur, it is almost certain to occur.Infringement\nThe action of a player contrary to the Rules of the game.In-Goal Area\nThe area in the Field of Play bounded by the Sidelines, the Try Lines \nand the Dead Ball Lines.There are two (2), one (1) at each end of the \nField of Play.See Appendix 1.Interchange\nThe act of an on-field player leaving the Field of Play to be replaced \nby an off-field player entering the Field of Play.',
]
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

touch-rugby-modernbert-pairs

Dataset: touch-rugby-modernbert-pairs at 7cb0ae2
Size: 305 training samples
Columns: question and related_chunk
Approximate statistics based on the first 305 samples:
question related_chunk
type string string
details
min: 10 tokens
mean: 18.68 tokens
max: 36 tokens
min: 147 tokens
mean: 231.42 tokens
max: 319 tokens

Samples:

question	related_chunk
`When may Onside players of the Defending Team move forward if the Half is not within one metre of the Rollball?`	13.10 A player ceases to be the Half once the ball is passed to another player.13.11 Defending players are not to interfere with the performance of the Rollball or the [object Object]Half.Ruling = A Penalty to the Attacking Team at a point ten (10) metres directly Forward of the [object Object]Infringement.13.12 Players of the Defending Team must not move Forward of the Onside position [object Object]until the Half has made contact with the ball, unless directed to do so by the [object Object]Referee or in accordance with 13.12.1.13.12.1 When the Half is not within one (1) metre of the Rollball, Onside players [object Object]of the Defending Team may move Forward as soon as the player [object Object]performing the Rollball releases the ball.If the Half is not in position and [object Object]a defending player moves Forward and makes contact with the ball, a [object Object]Change of Possession results.
`Besides awarding tries, what other scoring-related task does the Referee perform?`	An approach may only be made during a break in play or at [object Object]the discretion of the Referee.FIT Playing Rules - 5th Edition[object Object]18[object Object]COPYRIGHT © Touch Football Australia 2020[object Object]HALFWAY LINE[object Object]SIN BIN AREAS[object Object]IN-GOAL AREA[object Object]TRY LINE[object Object]7 M ZONE[object Object]DEAD BALL LINE[object Object]PERIMETER[object Object]INTERCHANGE[object Object]AREA[object Object]20M[object Object]10M[object Object]10M[object Object]1M[object Object]5M[object Object]7 M[object Object]7 M[object Object]7 M[object Object]7 M[object Object]50M[object Object]3M[object Object]70M[object Object]INTERCHANGE[object Object]AREA[object Object] Appendix 1 – Field of Play[object Object]FIT Playing Rules - 5th Edition[object Object]COPYRIGHT © Touch Football Australia 2020[object Object]19[object Object]FEDERATION OF INTERNATIONAL TOUCH
`What happens if a team has fewer than four players on the field during a match?`	FIT Playing Rules - 5th Edition[object Object]COPYRIGHT © Touch Football Australia 2020[object Object]7[object Object]7.6 A Tap may not be taken until at least four (4) defending players are in an Onside [object Object]position or unless directed to so by the Referee.Where the number of players [object Object]on the field from the Defending Team falls below four (4), all players must be in [object Object]an Onside position for a Tap to be taken unless directed to do so by the Referee.Ruling = The Player will be directed to return to the Mark and to take the Tap again.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 [object Object] [object Object]8.1 A match is 40 minutes in duration, consisting of two (2) x 20 minute halves with [object Object]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.

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

Evaluation Dataset

touch-rugby-modernbert-pairs

Dataset: touch-rugby-modernbert-pairs at 7cb0ae2
Size: 305 evaluation samples
Columns: question and related_chunk
Approximate statistics based on the first 305 samples:
question related_chunk
type string string
details
min: 11 tokens
mean: 18.06 tokens
max: 32 tokens
min: 173 tokens
mean: 228.39 tokens
max: 260 tokens

Samples:

question	related_chunk
`What is the definition of the 'Defending Team' in Touch Rugby Rules 5th Edition?`	Except as permitted under the [object Object]Copyright Act, these Rules must not be reproduced by any process, electronic or otherwise, without the written [object Object]permission of Touch Football Australia.Attacking Try Line[object Object]The line on or over which a player has to place the ball to [object Object]score a Try.Attacking Team[object Object]The Team which has or is gaining Possession.Behind[object Object]A position or direction towards a Team’s Defending Try Line.Change of Possession[object Object]The act of moving control of the ball from one Team to the other.Dead/Dead Ball[object Object]When the ball is out of play including the period following a Try and [object Object]until the match is recommenced and when the ball goes to ground [object Object]and/or outside the boundaries of the Field of Play prior to the [object Object]subsequent Rollball.Dead Ball Line[object Object]The end boundaries of the Field of Play.There is one at each end of [object Object]the Field of Play.See Appendix 1.Defending Try Line[object Object]The line which a Team has to defend to prevent a Try.Defending Team[object Object]The Team without or which is losing Possession.
`What is the minimum number of players required on the field for a touch rugby match to begin or continue?`	FIT Playing Rules - 5th Edition[object Object]COPYRIGHT © Touch Football Australia 2020[object Object]7[object Object]7.6 A Tap may not be taken until at least four (4) defending players are in an Onside [object Object]position or unless directed to so by the Referee.Where the number of players [object Object]on the field from the Defending Team falls below four (4), all players must be in [object Object]an Onside position for a Tap to be taken unless directed to do so by the Referee.Ruling = The Player will be directed to return to the Mark and to take the Tap again.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 [object Object] [object Object]8.1 A match is 40 minutes in duration, consisting of two (2) x 20 minute halves with [object Object]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.
`What are the possible outcomes of a Referee's Ruling?`	See Appendix 1.Forced Interchange[object Object]When a player is required to undertake a compulsory Interchange for [object Object]an Infringement ruled more serious than a Penalty but less serious [object Object]than a Permanent Interchange, Sin Bin or Dismissal.Forward[object Object]A position or direction towards the Dead Ball Line beyond the Team’s [object Object]Attacking Try Line.Full Time[object Object]The expiration of the second period of time allowed for play.Half[object Object]The player who takes Possession following a Rollball.Half Time[object Object]The break in play between the two halves of a match.Imminent[object Object]About to occur, it is almost certain to occur.Infringement[object Object]The action of a player contrary to the Rules of the game.In-Goal Area[object Object]The area in the Field of Play bounded by the Sidelines, the Try Lines [object Object]and the Dead Ball Lines.There are two (2), one (1) at each end of the [object Object]Field of Play.See Appendix 1.Interchange[object Object]The act of an on-field player leaving the Field of Play to be replaced [object Object]by an off-field player entering the Field of Play.

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 32
per_device_eval_batch_size: 32
learning_rate: 5e-06
num_train_epochs: 1
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
torch_empty_cache_steps: None
learning_rate: 5e-06
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 1
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: 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: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss
0.2222	2	2.8177	2.5945
0.4444	4	2.9155	2.5693
0.6667	6	2.9114	2.5402
0.8889	8	2.7999	2.5098

Framework Versions

Python: 3.12.4
Sentence Transformers: 3.3.1
Transformers: 4.48.0
PyTorch: 2.5.1
Accelerate: 1.3.0
Datasets: 2.17.1
Tokenizers: 0.21.0

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}
}

	question	related_chunk
type	string	string
details	min: 10 tokens mean: 18.68 tokens max: 36 tokens	min: 147 tokens mean: 231.42 tokens max: 319 tokens