jebish7/snowflake-arctic-embed-m-long_MNR_half

SentenceTransformer based on Snowflake/snowflake-arctic-embed-m-long

This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-m-long on the csv 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: Snowflake/snowflake-arctic-embed-m-long
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • csv

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: NomicBertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, '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("jebish7/snowflake-arctic-embed-m-long_MNR_half")
# Run inference
sentences = [
    'How should a Relevant Person ensure and demonstrate compliance with both UNSC Sanctions and U.A.E.-administered Sanctions, specifically Targeted Financial Sanctions, within the ADGM jurisdiction?',
    'Where a Relevant Person seeks to rely on a Person in (1) it may only do so if and to the extent that:\n(a)\tit immediately obtains the necessary CDD information from the third party in (1);\n(b)\tit takes adequate steps to satisfy itself that certified copies of the documents used to undertake the relevant elements of CDD will be available from the third party on request without delay;\n(c)\tthe Person in (1)(b) to (d) is subject to regulation, including AML/TFS compliance requirements, by a Non-ADGM Financial Services Regulator or other competent authority in a country with AML/TFS regulations which are equivalent to the standards set out in the FATF Recommendations and it is supervised for compliance with such regulations;\n(d)\tthe Person in (1) has not relied on any exception from the requirement to conduct any relevant elements of CDD which the Relevant Person seeks to rely on; and\n(e)\tin relation to (2), the information is up to date.',
    'REGULATORY REQUIREMENTS - SPOT COMMODITY ACTIVITIES\nRIEs operating an MTF or OTF using Accepted Spot Commodities\nAuthorised Persons that are operating an MTF or OTF wishing to also operate a RIE will be required to relinquish their FSP upon obtaining a Recognition Order (to operate the RIE).  If licensed by the FSRA to carry out both Regulated Activities (e.g., operating an MTF and operating an RIE), the Recognition Order will include a stipulation to that effect pursuant to MIR Rule 3.4.1.\n',
]
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

csv

• Dataset: csv
• Size: 29,547 training samples
• Columns: Question and positive
• Approximate statistics based on the first 1000 samples:

  	Question	positive
  type	string	string
  details	min: 18 tokens mean: 34.27 tokens max: 76 tokens	min: 13 tokens mean: 112.44 tokens max: 768 tokens

• Samples:

  Question	positive
  Regarding the assessment of the nature, ownership, and control structure of a customer, could you clarify the level of detail and due diligence expected from a Relevant Person to ensure adherence to regulatory standards?	The risk assessment under Rule ‎6.2.1(c) should identify actions to mitigate risks associated with undertaking NFTF business generally, and the use of eKYC specifically. This is because distinct risks are often likely to arise where business is conducted entirely in an NFTF manner, compared to when the business relationship includes a mix of face-to-face and NFTF interactions. The assessment should make reference to risk mitigation measures recommended by the Regulator, a competent authority of the U.A.E., FATF, and other relevant bodies.
  What specific factors should be included in our risk assessment methodology to ensure it aligns with the expectations outlined in Chapter 6 and Chapter 7 of the AML Rulebook?	The RBA should not be seen as a "tick-box" approach to AML/TFS. Instead a Relevant Person is required to assess relevant money laundering risks and adopt a proportionate response to such risks, however, even where a customer is assessed through the RBA as being low-risk a minimum of simplified CDD must be undertaken in relation to that customer.
  In the event of an investigation by the ADGM, what are the specific obligations and limitations regarding confidentiality for the entity under investigation, and what kind of disclosures are permissible under sections 197 and 198 of FSMR?	We will not normally make public the fact that we are investigating a matter. We also expect that the person who is the subject of an investigation will treat the matter as confidential. However, subject to the restrictions on disclosure of confidential information in sections 197 and 198 of FSMR, this does not stop the person under investigation from seeking professional advice or making their own enquiries into the matter, giving their auditors appropriate details of the matter or making notifications required by law.

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 4
• 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: no
• prediction_loss_only: True
• per_device_train_batch_size: 4
• per_device_eval_batch_size: 8
• 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: 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
• use_liger_kernel: False
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss
0.0271	100	0.6411
0.0541	200	0.3289
0.0812	300	0.2395
0.1083	400	0.2711
0.1354	500	0.2746
0.1624	600	0.2602
0.1895	700	0.285
0.2166	800	0.2965
0.2436	900	0.2772
0.2707	1000	0.3043
0.2978	1100	0.3059
0.3249	1200	0.316
0.3519	1300	0.2765
0.3790	1400	0.249
0.4061	1500	0.2601
0.4331	1600	0.2538
0.4602	1700	0.2443
0.4873	1800	0.2151
0.5143	1900	0.2335
0.5414	2000	0.2611
0.5685	2100	0.2557
0.5956	2200	0.2793

Framework Versions

• Python: 3.10.14
• Sentence Transformers: 3.1.1
• Transformers: 4.45.2
• PyTorch: 2.4.0
• Accelerate: 0.34.2
• Datasets: 3.0.1
• Tokenizers: 0.20.0

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}
}