amentaphd/new-snowflake-arctic-embed-m-v2.0

SentenceTransformer based on Snowflake/snowflake-arctic-embed-m-v2.0

This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-m-v2.0. 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-v2.0
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity

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: GteModel 
  (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("sentence_transformers_model_id")
# Run inference
sentences = [
    'How does the new Eurostat methodology differ in scope from the indicators used in this Directive for calculating energy consumption?',
    '(29) The methodology for calculation of primary energy consumption and final energy consumption is aligned with the new Eurostat methodology, but the indicators used for the purpose of this Directive have a different scope, in that they exclude ambient energy and include energy consumption in international aviation for the targets in primary energy consumption and final energy consumption. The use of new indicators also implies that any changes in energy consumption of blast furnaces are now only reflected in primary energy consumption.',
    '(92) InvestEU is the Union flagship programme to boost investment, especially the green and digital transition, by providing financing and technical assistance, for instance through blending mechanisms. Such an approach contributes to crowd in additional public and private capital. Moreover, Member States are encouraged to contribute to the InvestEU Member State compartment to support financial products available to net-zero technology manufacturing, without prejudice to applicable State aid rules.',
]
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

• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.7136
cosine_accuracy@3	0.9244
cosine_accuracy@5	0.9589
cosine_accuracy@10	0.9819
cosine_precision@1	0.7136
cosine_precision@3	0.3081
cosine_precision@5	0.1918
cosine_precision@10	0.0982
cosine_recall@1	0.7136
cosine_recall@3	0.9244
cosine_recall@5	0.9589
cosine_recall@10	0.9819
cosine_ndcg@10	0.8626
cosine_mrr@10	0.8228
cosine_map@100	0.8237

Training Details

Training Dataset

Unnamed Dataset

• Size: 46,338 training samples
• Columns: query_text and doc_text
• Approximate statistics based on the first 1000 samples:

  	query_text	doc_text
  type	string	string
  details	min: 9 tokens mean: 39.44 tokens max: 311 tokens	min: 7 tokens mean: 233.15 tokens max: 1900 tokens

• Samples:

  query_text	doc_text
  The regulation's applicability extends to various stakeholders involved in AI systems, including providers, deployers, importers, and manufacturers, regardless of their location. It specifically addresses high-risk AI systems and outlines the limitations of its scope, particularly concerning national security and military applications. Additionally, it clarifies that it does not interfere with the responsibilities of member states regarding national security or the operations of public authorities and international organizations in specific contexts.	(180) The European Data Protection Supervisor and the European Data Protection Board were consulted in accordance with Article 42(1) and (2) of Regulation (EU) 2018/1725 and delivered their joint opinion on 18 June 2021, HAVE ADOPTED THIS REGULATION: CHAPTER I GENERAL PROVISIONS Article 1 Subject matter` 1. The purpose of this Regulation is to improve the functioning of the internal market and promote the uptake of human-centric and trustworthy artificial intelligence (AI), while ensuring a high level of protection of health, safety, fundamental rights enshrined in the Charter, including democracy, the rule of law and environmental protection, against the harmful effects of AI systems in the Union and supporting innovation. 2. This Regulation lays down: (a) harmonised rules for the placing on the market, the putting into service, and the use of AI systems in the Union; (b) prohibitions of certain AI practices; --- --- (c) specific requirements for high-risk AI systems and oblig...
  How should loans with unknown use of proceeds be allocated in terms of sectors and alignment metrics?	instruments. For loans whose use of proceeds is known, the value shall be included for the relevant sector and alignment metric. For loans whose use of proceeds is unknown, the gross carrying amount of the exposure shall be allocated to the relevant sectors and alignment metrics based on the counterparties’ activity distribution, including by counterparties’ turnover by activity. Institutions shall add a row in the template for each relevant combination of sectors disclosed in column (b) and alignment metrics included in column (d). ---
  What measures must AIFMs implement to ensure they do not rely solely on credit ratings for assessing the creditworthiness of AIFs' assets?	▼M1 The measures specifying the risk-management systems referred to in point (a) of the first subparagraph shall ensure that the AIFMs are prevented from relying solely or mechanistically on credit ratings, as referred to in the first subparagraph of paragraph 2, for assessing the creditworthiness of the AIFs’ assets. ▼B Article 16 Liquidity management 1. AIFMs shall, for each AIF that they manage which is not an unleveraged closed- ended AIF, employ an appropriate liquidity management system and adopt procedures which enable them to monitor the liquidity risk of the AIF and to ensure that the liquidity profile of the investments of the AIF complies with its underlying obligations.

• 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: steps
• learning_rate: 2e-05
• num_train_epochs: 4
• warmup_ratio: 0.1
• fp16: True
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 8
• 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: 4
• 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: 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
• 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	cosine_ndcg@10
-1	-1	-	0.7763
0.0863	500	0.2343	-
0.1726	1000	0.1259	0.814
0.2589	1500	0.1027	-
0.3452	2000	0.0757	0.8288
0.4316	2500	0.0617	-
0.5179	3000	0.0651	0.8288
0.6042	3500	0.0863	-
0.6905	4000	0.06	0.8376
0.7768	4500	0.0579	-
0.8631	5000	0.0593	0.8342
0.9494	5500	0.0485	-
1.0357	6000	0.0465	0.8384
1.1220	6500	0.0276	-
1.2084	7000	0.0353	0.8392
1.2947	7500	0.0335	-
1.3810	8000	0.0292	0.8436
1.4673	8500	0.0276	-
1.5536	9000	0.0404	0.8485
1.6399	9500	0.0476	-
1.7262	10000	0.0265	0.8601
1.8125	10500	0.017	-
1.8988	11000	0.0217	0.8549
1.9852	11500	0.0329	-
2.0715	12000	0.0207	0.8577
2.1578	12500	0.0199	-
2.2441	13000	0.015	0.8544
2.3304	13500	0.0143	-
2.4167	14000	0.0117	0.8574
2.5030	14500	0.0204	-
2.5893	15000	0.0141	0.8595
2.6756	15500	0.0123	-
2.7620	16000	0.0211	0.8538
2.8483	16500	0.0207	-
2.9346	17000	0.0134	0.8562
3.0209	17500	0.0276	-
3.1072	18000	0.0106	0.8552
3.1935	18500	0.0129	-
3.2798	19000	0.0157	0.8582
3.3661	19500	0.0164	-
3.4524	20000	0.0192	0.8614
3.5388	20500	0.0138	-
3.6251	21000	0.0141	0.8601
3.7114	21500	0.0109	-
3.7977	22000	0.0178	0.8605
3.8840	22500	0.0088	-
3.9703	23000	0.0255	0.8626

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.10.15
• Sentence Transformers: 4.0.2
• Transformers: 4.49.0
• PyTorch: 2.6.0+cu126
• Accelerate: 0.26.0
• 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}
}