This is a sentence-transformers model finetuned from sentence-transformers/multi-qa-mpnet-base-cos-v1 on the mediclaim 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: sentence-transformers/multi-qa-mpnet-base-cos-v1
Maximum Sequence Length: 512 tokens
Output Dimensionality: 768 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- mediclaim

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: MPNetModel 
  (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("surajvbangera/mediclaim_embedding")
# Run inference
sentences = [
    'what kind of coverage is provided by insurance for medical expenses that go beyond the normal amount?',
    'health insurance cover and provides wider health protection for you and your family. In case of higher expenses \ndue to illness or accidents, Extra Care Plus policy takes care of the additional expenses. It is important to consider',
    'Age/\ndeduc-\ntible\n200000 200000 300000 200000 300000 500000 300000 500000 300000 500000 1000000 300000 500000 1000000 300000 500000 1000000\n21-25 6,544 7,011 4,345 10,389 7,490 5,127 9,839 7,283 11,767 9,087 6,289 13,419 10,054 7,343 19,518 16,543 13,717',
]
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

Datasets: dim_768, dim_512, dim_256, dim_128 and dim_64
Evaluated with InformationRetrievalEvaluator

Metric	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.3021	0.2812	0.3021	0.2708	0.25
cosine_accuracy@3	0.8021	0.7812	0.7917	0.7812	0.7292
cosine_accuracy@5	0.875	0.875	0.8854	0.8438	0.8333
cosine_accuracy@10	0.9583	0.9479	0.9375	0.9479	0.9167
cosine_precision@1	0.3021	0.2812	0.3021	0.2708	0.25
cosine_precision@3	0.2674	0.2604	0.2639	0.2604	0.2431
cosine_precision@5	0.175	0.175	0.1771	0.1687	0.1667
cosine_precision@10	0.0958	0.0948	0.0938	0.0948	0.0917
cosine_recall@1	0.3021	0.2812	0.3021	0.2708	0.25
cosine_recall@3	0.8021	0.7812	0.7917	0.7812	0.7292
cosine_recall@5	0.875	0.875	0.8854	0.8438	0.8333
cosine_recall@10	0.9583	0.9479	0.9375	0.9479	0.9167
cosine_ndcg@10	0.6498	0.6294	0.6397	0.6229	0.5922
cosine_mrr@10	0.5484	0.5251	0.541	0.5167	0.4863
cosine_map@100	0.5513	0.5287	0.5446	0.5187	0.4908

Training Details

Training Dataset

mediclaim

Dataset: mediclaim at 943cab1
Size: 956 training samples
Columns: anchor and positive
Approximate statistics based on the first 956 samples:
anchor positive
type string string
details
min: 10 tokens
mean: 23.14 tokens
max: 85 tokens
min: 6 tokens
mean: 57.2 tokens
max: 135 tokens

Samples:

anchor	positive
`Can I get a preventive health check-up covered under my insurance, and if yes, is there a limit to it?`	`by the Medical Practitioner.[object Object] vii. The Deductible shall not be applicable on this bene�t.[object Object] Stay Fit Health Check Up[object Object] The Insured may avail a health check-up, only for Preventive [object Object]Test, up to a limit speci�ed in the Policy Schedule, provided`
`Which claims are excluded if they don't follow the Transplantation of Human Organs Amendment Bill 2011?`	`4 CIN: U66010PN2000PLC015329, UIN: BAJHLIP23069V032223[object Object] Specific exclusions:[object Object] 1. Claims which have NOT been admitted under Medical expenses section[object Object] 2. Claims not in compliance with THE TRANSPLANTATION OF HUMAN ORGANS (AMENDMENT) BILL, 2011`
`Will the insurance pay for lawful abortion and related hospital stays?`	`ii. We will also cover expenses towards lawful medical termination of pregnancy during the Policy period.[object Object] iii. In patient Hospitalization Expenses of pre-natal and post-natal hospitalization`

Loss: MatryoshkaLoss with these parameters:
json
1{ 2 "loss": "MultipleNegativesRankingLoss", 3 "matryoshka_dims": [ 4 768, 5 512, 6 256, 7 128, 8 64 9 ], 10 "matryoshka_weights": [ 11 1, 12 1, 13 1, 14 1, 15 1 16 ], 17 "n_dims_per_step": -1 18}

Evaluation Dataset

mediclaim

Dataset: mediclaim at 943cab1
Size: 956 evaluation samples
Columns: anchor and positive
Approximate statistics based on the first 956 samples:
anchor positive
type string string
details
min: 10 tokens
mean: 22.4 tokens
max: 62 tokens
min: 6 tokens
mean: 56.76 tokens
max: 133 tokens

Samples:

anchor	positive
`Is there any refund for medical exams if I get a policy and it's accepted?`	`• If pre-policy checkup is conducted, 50% of the medical tests charges would be reimbursed, subject to acceptance [object Object]of proposal and policy issuance.[object Object]Age of the person [object Object]to be insured[object Object]Sum Insured Medical Examination`
`Are there any exclusions for coverage of substance abuse treatment or its consequences?`	`are payable but not the complete claim. [object Object]12. T reatment for Alcoholism, drug or substance abuse or any addictive condition and consequences thereof. [object Object](Excl12)`
`Can you tell me about the medical bills I might have within 90 days after being discharged?`	`CIN: U66010PN2000PLC015329, UIN:BAJHLIP23069V032223 3[object Object] c. Post-hospitalisation expenses[object Object] The medical expenses incurred in the 90 days immediately after you were discharged, provided that:`

Loss: MatryoshkaLoss with these parameters:
json
1{ 2 "loss": "MultipleNegativesRankingLoss", 3 "matryoshka_dims": [ 4 768, 5 512, 6 256, 7 128, 8 64 9 ], 10 "matryoshka_weights": [ 11 1, 12 1, 13 1, 14 1, 15 1 16 ], 17 "n_dims_per_step": -1 18}

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: 40
lr_scheduler_type: cosine
warmup_ratio: 0.1
fp16: 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: 40
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: False
fp16: True
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: 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}
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
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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation 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
-1	-1	-	-	0.4723	0.4748	0.5015	0.4589	0.3867
1.0	2	-	1.5925	0.4821	0.4846	0.5122	0.4604	0.3971
2.0	4	-	1.5925	0.4821	0.4846	0.5122	0.4604	0.3971
3.0	6	-	1.0402	0.5431	0.5468	0.5530	0.5009	0.4435
4.0	8	-	0.7900	0.5876	0.5926	0.6075	0.5484	0.4726
5.0	10	33.0646	0.6077	0.5890	0.6039	0.6270	0.5779	0.5072
6.0	12	-	0.5213	0.6357	0.6379	0.6522	0.5966	0.5417
7.0	14	-	0.4735	0.6425	0.6395	0.6286	0.5995	0.5795
8.0	16	-	0.4416	0.6253	0.6387	0.6227	0.5903	0.5738
9.0	18	-	0.4236	0.6303	0.6489	0.6387	0.6179	0.5670
10.0	20	8.8456	0.4115	0.6465	0.6519	0.6369	0.6112	0.572
11.0	22	-	0.4059	0.6447	0.6270	0.6318	0.6169	0.5950
12.0	24	-	0.4036	0.6382	0.6318	0.6346	0.6063	0.6026
13.0	26	-	0.4022	0.6485	0.6410	0.6441	0.6163	0.5900
14.0	28	-	0.4022	0.6520	0.6426	0.6597	0.6225	0.6001
15.0	30	4.4602	0.4033	0.6507	0.6363	0.6576	0.6217	0.6134
16.0	32	-	0.4047	0.6530	0.6389	0.6609	0.6350	0.6068
17.0	34	-	0.4058	0.6501	0.6344	0.6501	0.6281	0.5997
18.0	36	-	0.4067	0.6509	0.6333	0.6553	0.6360	0.6050
19.0	38	-	0.4070	0.6561	0.6331	0.6602	0.6397	0.6051
20.0	40	3.9605	0.4071	0.6498	0.6294	0.6397	0.6229	0.5922

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.4.1
Transformers: 4.48.3
PyTorch: 2.5.1+cu124
Accelerate: 1.3.0
Datasets: 3.3.2
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",
}

MatryoshkaLoss

bibtex

@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

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}
}

	anchor	positive
type	string	string
details	min: 10 tokens mean: 23.14 tokens max: 85 tokens	min: 6 tokens mean: 57.2 tokens max: 135 tokens