SentenceTransformer based on nomic-ai/modernbert-embed-base

This is a sentence-transformers model finetuned from nomic-ai/modernbert-embed-base on the finqalab_embedding_finetune 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:
- finqalab_embedding_finetune

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("Ch333tah/modernbert-finqalab-embeddings")
# Run inference
sentences = [
    "Question: I'm experiencing issues with logging in to the app. What should I do?  Answer: In case you are facing any issues, Try closing the app and opening it again. Try clearing the cache or updating to the latest version. If the issue still persists, contact our customer support through Whatsapp (+923003672522). Email your query at support@finqalab.com.pk",
    "My app won't let me log in!  Help!",
    "Question: The app is not loading properly on my device. What could be the problem?  Answer: If the app isn't loading properly: Please check if you have a stable internet connection. Try refreshing the screen 2-3 times. Close the app and open it again. Try clearing the cache, check for app updates or reinstall the app. If the issue still persists, contact our customer support through Whatsapp (+923003672522) or email your query at support@finqalab.com",
]
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

Triplet

Datasets: ai-job-train, ai-job-valid, ai-job-test, ai-job-train, ai-job-valid and ai-job-test
Evaluated with TripletEvaluator

Metric	ai-job-train	ai-job-valid	ai-job-test
cosine_accuracy	0.9604	0.9524	0.913

Training Details

Training Dataset

finqalab_embedding_finetune

Dataset: finqalab_embedding_finetune at 144dee2
Size: 101 training samples
Columns: Pos_Context, Query, and Neg_Context

Approximate statistics based on the first 101 samples:

	Pos_Context	Query	Neg_Context
type	string	string	string
details	min: 27 tokens mean: 74.1 tokens max: 418 tokens	min: 7 tokens mean: 15.19 tokens max: 30 tokens	min: 25 tokens mean: 72.42 tokens max: 231 tokens

Samples:

Pos_Context	Query	Neg_Context
`Question: I did not receive a verification email? What should I do? Answer: Please check your spam folder and double check your registered email address. If you still dont see a verification email, contact our customer support department at support@finqalab.com.`	`My verification email didn't arrive, any ideas?`	`Question: I did not receive my instant transfer in Finqalab account within 10 minutes. What should I do? Answer: If your instant transfer hasnt been credited to your Finqalab account within 10 minutes, please email us at support@finqalab.com with a screenshot of the receipt or send it to us on Whatsapp (+923003672522). Our team will review the issue, escalate it to the bank by sending the transaction receipt, and follow up to ensure your funds are credited promptly.`
`Question: What are the applicable CGT rates for RDA Account Holders? Answer: Filer rates are applied to RDA account holders irrespective of their status (Filer or Non-filer).`	`How are capital gains taxes handled for someone with an RDA account?`	`Question: What does Minimum Lot Size mean? Answer: This means that you need to buy a minimum quantity for a share. In case of ETFs the minimum lot size is 500 or in multiples of 500 shares. Whereas, for non-ETF stocks the minimum lot size is 1 share.`
`Question: How do I receive bonus shares? Answer: Bonus shares are distributed to shareholders based on a ratio announced by the company. For example, if a company declares a 20% bonus issue, you will receive 2 additional shares for every 10 shares you already own.`	`What's the deal with getting extra shares?`	`Question: Do I have to pay for bonus shares? Answer: No, bonus shares are issued free of charge. They are typically paid for by utilizing the companys retained earnings or reserves.`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Evaluation Dataset

finqalab_embedding_finetune

Dataset: finqalab_embedding_finetune at 144dee2
Size: 21 evaluation samples
Columns: Pos_Context, Query, and Neg_Context

Approximate statistics based on the first 21 samples:

	Pos_Context	Query	Neg_Context
type	string	string	string
details	min: 23 tokens mean: 78.67 tokens max: 152 tokens	min: 12 tokens mean: 16.67 tokens max: 27 tokens	min: 27 tokens mean: 91.9 tokens max: 418 tokens

Samples:

Pos_Context	Query	Neg_Context
`Question: How will I know if my biometric verification is pending or close to the deadline? Answer: We receive reports every Monday and Friday regarding users with pending biometric verifications. If you have less than 7 days remaining, we will contact you to remind you to complete the process or provide alternate solutions if necessary.`	`What happens if my biometric verification is about to expire?`	`Question: I entered my CNIC in the Bioverify app and got CNIC not eligible for this service message. What does this mean? Answer: This means that the Biometric verification is not required.`
`Question: How long do I have to complete the biometric verification? Answer: Once you receive the OTP, you must finish the biometric process within 45 days. An automated email will remind you to complete it.`	`What's the deadline for that fingerprint thing after I get the code?`	`Question: Can I complete the biometric verification from outside Pakistan? Answer: Yes, if you are currently abroad, you can complete the biometric process either online or by visiting an NCCPL office when you return to Pakistan, provided its within 45 days of account activation.`
`Question: Is historical price data available for stocks in the app? Answer: Yes, it is.`	`Can I see stock prices from the past in this app?`	`Question: How often is the stock data updated in the app? Answer: The stock data is updated in real time.`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
learning_rate: 2e-05
num_train_epochs: 1
warmup_ratio: 0.1
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
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: 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: 1
max_steps: -1
lr_scheduler_type: linear
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: 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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	ai-job-train_cosine_accuracy	ai-job-valid_cosine_accuracy	ai-job-test_cosine_accuracy
-1	-1	0.9604	0.9524	0.9130

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.4.0
Transformers: 4.48.1
PyTorch: 2.5.1+cu121
Accelerate: 1.2.1
Datasets: 3.2.0
Tokenizers: 0.21.0

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}
}