ModernBERT Embed base fitness health Matryoshka

This is a sentence-transformers model finetuned from kokojake/modernbert-embed-base-fitness-health-matryoshka on the json 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: kokojake/modernbert-embed-base-fitness-health-matryoshka
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json
Language: en
License: apache-2.0

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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("kokojake/modernbert-embed-base-fitness-health-matryoshka-8-epochs")
# Run inference
sentences = [
    'With this policy paper we offer our teams and partners the basis for understanding what we \nbelieve physical and functional rehabilitation covers today. While we can look back and see how far we have come, we are also aware of the changes we have yet to go through. Hence it \nhighlights the directions we will be taking in the coming years, for example the user-centred approach, the quality and sustainability of services in developing countries, improving the \nprofessional training process, and connecting with user groups.\nWe have a small, highly-motivated and ambitious team that, with limited resources (always too limited!), manages daily tours-de-force to improve our practices, capitalise, train, \ninnovate, structure and improve the position of physical and functional rehabilitation in the world. This document follows naturally from that, highlighting how the medical field plays an \nessential role in enhancing the social participation of people with disabilities. The beneficiary becomes a new participant in his own health, allowing him to be a stakeholder in building the \nsocieties of today and tomorrow.\nThanks to all who contributed, and happy reading.',
    'role of medical field in social participation for disabilities',
    'selection and adjustment of assistive products for patient independence',
]
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]

Evaluation

Metrics

Information Retrieval

Dataset: dim_768
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 768
}
```

Metric	Value
cosine_accuracy@1	0.5211
cosine_accuracy@3	0.5211
cosine_accuracy@5	0.5211
cosine_accuracy@10	0.5578
cosine_precision@1	0.5211
cosine_precision@3	0.5211
cosine_precision@5	0.5211
cosine_precision@10	0.4754
cosine_recall@1	0.0649
cosine_recall@3	0.1947
cosine_recall@5	0.3245
cosine_recall@10	0.5487
cosine_ndcg@10	0.5375
cosine_mrr@10	0.5272
cosine_map@100	0.606

Information Retrieval

Dataset: dim_512
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 512
}
```

Metric	Value
cosine_accuracy@1	0.5062
cosine_accuracy@3	0.5062
cosine_accuracy@5	0.5062
cosine_accuracy@10	0.5461
cosine_precision@1	0.5062
cosine_precision@3	0.5057
cosine_precision@5	0.5056
cosine_precision@10	0.4611
cosine_recall@1	0.0636
cosine_recall@3	0.1904
cosine_recall@5	0.3173
cosine_recall@10	0.5348
cosine_ndcg@10	0.5233
cosine_mrr@10	0.5129
cosine_map@100	0.595

Information Retrieval

Dataset: dim_256
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 256
}
```

Metric	Value
cosine_accuracy@1	0.4859
cosine_accuracy@3	0.4859
cosine_accuracy@5	0.4859
cosine_accuracy@10	0.5227
cosine_precision@1	0.4859
cosine_precision@3	0.4859
cosine_precision@5	0.4859
cosine_precision@10	0.4407
cosine_recall@1	0.0612
cosine_recall@3	0.1837
cosine_recall@5	0.3062
cosine_recall@10	0.5121
cosine_ndcg@10	0.5017
cosine_mrr@10	0.4921
cosine_map@100	0.5785

Information Retrieval

Dataset: dim_128
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 128
}
```

Metric	Value
cosine_accuracy@1	0.4688
cosine_accuracy@3	0.4688
cosine_accuracy@5	0.4688
cosine_accuracy@10	0.507
cosine_precision@1	0.4688
cosine_precision@3	0.4688
cosine_precision@5	0.4686
cosine_precision@10	0.43
cosine_recall@1	0.0584
cosine_recall@3	0.1753
cosine_recall@5	0.292
cosine_recall@10	0.4963
cosine_ndcg@10	0.4854
cosine_mrr@10	0.4751
cosine_map@100	0.5618

Information Retrieval

Dataset: dim_64
Evaluated with InformationRetrievalEvaluator with these parameters:
```
{
    "truncate_dim": 64
}
```

Metric	Value
cosine_accuracy@1	0.4141
cosine_accuracy@3	0.4141
cosine_accuracy@5	0.4141
cosine_accuracy@10	0.4586
cosine_precision@1	0.4141
cosine_precision@3	0.4141
cosine_precision@5	0.4141
cosine_precision@10	0.3875
cosine_recall@1	0.0512
cosine_recall@3	0.1535
cosine_recall@5	0.2558
cosine_recall@10	0.4471
cosine_ndcg@10	0.4339
cosine_mrr@10	0.4215
cosine_map@100	0.5068

Training Details

Training Dataset

json

Dataset: json
Size: 11,518 training samples
Columns: positive and anchor
Approximate statistics based on the first 1000 samples:
positive anchor
type string string
details
min: 10 tokens
mean: 244.14 tokens
max: 412 tokens

min: 4 tokens
mean: 10.92 tokens
max: 39 tokens

	positive	anchor
type	string	string
details	min: 10 tokens mean: 244.14 tokens max: 412 tokens	min: 4 tokens mean: 10.92 tokens max: 39 tokens

Samples:

positive	anchor
certainty as rated in the Cochrane review (89) due to indirectness], 35 trials); and › improve function (trivial effect, low certainty evidence [downgraded # Review findings: values and preferences relevant to older people GRADE- CERQual assessment of confidence Explanation of confidence assessment 12 Older people emphasized the importance of continuity of physical exercises to maintain mobility and reduce pain. A lack of continuity of physical exercise and instruction could have adverse effects. LOW No/very minor concerns regarding methodological limitations, moderate concerns regarding coherence, minor concerns regarding adequacy, and minor concerns regarding relevance. 13 Older people also valued educational materials to accompany exercise programmes, such as drawings and descriptions of the exercises. LOW Minor concerns regarding methodological limitations, no/very minor concerns regarding coherence, serious concerns regarding adequacy, and serious con...	`importance of physical exercise continuity for older adults mobility and pain reduction`
Phosphodiesterase-5 inhibitors Prescription and/or administration (if injection) of the medicine and providing education and advice on the safe intake or administration (if self-directed) and potential adverse effects of the medicine. Physical exercise training A variety of physical exercises (e.g. aerobic or strengthening exercises, balance or coordination exercises, mind–body exercises), with or without weight-bearing, are suitable to improve exercise capacity, muscle strength, joint mobility, voluntary movement, balance, gait and walking, as well as helping to reduce pain and fatigue. Regular physical exercise training (including education and advice on exercises) is planned according to an individual’s needs, guided or assisted and, if feasible, performed self-directed following education and advice on the appropriate exercises.	`Phosphodiesterase-5 inhibitors prescription administration and adverse effects`
`and occasional guidelines (Bach-Faig 2011): Every day • Three main meals should contain three basic elements: o Cereals: 1-2 servings per meal (preferably whole grain), such as bread, pasta, rice, and couscous`	`three main meals basic elements cereals whole grain`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        512,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 32
per_device_eval_batch_size: 16
gradient_accumulation_steps: 16
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: True
tf32: True
load_best_model_at_end: True
optim: adamw_torch_fused
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
prediction_loss_only: True
per_device_train_batch_size: 32
per_device_eval_batch_size: 16
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 16
eval_accumulation_steps: None
torch_empty_cache_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: cosine
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
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: True
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: True
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}
tp_size: 0
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_fused
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
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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.4444	10	16.2146	-	-	-	-	-
0.8889	20	15.6805	-	-	-	-	-
1.0	23	-	0.5287	0.5234	0.5034	0.4812	0.4255
1.3111	30	13.2495	-	-	-	-	-
1.7556	40	13.4064	-	-	-	-	-
2.0	46	-	0.5320	0.5199	0.5019	0.4776	0.4306
2.1778	50	11.3483	-	-	-	-	-
2.6222	60	11.7323	-	-	-	-	-
3.0	69	-	0.5375	0.5216	0.5014	0.4816	0.4288
3.0444	70	11.3371	-	-	-	-	-
3.4889	80	10.2106	-	-	-	-	-
3.8444	88	-	0.5375	0.5233	0.5017	0.4854	0.4339

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.11
Sentence Transformers: 4.0.2
Transformers: 4.51.1
PyTorch: 2.6.0+cu124
Accelerate: 1.5.2
Datasets: 3.5.0
Tokenizers: 0.21.1

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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

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}
}

kokojake
/

modernbert-embed-base-fitness-health-matryoshka-8-epochs

ModernBERT Embed base fitness health Matryoshka

Model Details

Model Description

Model Sources

Full Model Architecture

Usage

Direct Usage (Sentence Transformers)

Evaluation

Metrics

Information Retrieval

Information Retrieval

Information Retrieval

Information Retrieval

Information Retrieval

Training Details

Training Dataset

json

Training Hyperparameters

Non-Default Hyperparameters

All Hyperparameters

Training Logs

Framework Versions

Citation

BibTeX

Sentence Transformers

MatryoshkaLoss

MultipleNegativesRankingLoss

Model tree for kokojake/modernbert-embed-base-fitness-health-matryoshka-8-epochs

Evaluation results