krumeto/snowflake-arctic-embed-xs-ms-marco-triplet

Fine-tuned snowflake actic xs based on MS-Marco triplets

This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-xs on the sentence-transformers/msmarco-msmarco-mini_lm-l-6-v3 dataset. It maps sentences & paragraphs to a 384-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-xs
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • sentence-transformers/msmarco-msmarco-mini_lm-l-6-v3
• 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': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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("krumeto/snowflake-arctic-embed-xs-ms-marco-triplet")
# Run inference
sentences = [
    'Represent this sentence for searching relevant passages: how many pitchers used per game',
    'In modern day baseball, teams generally have five starting pitchers, and they take it in turn to start a game every fifth day (hence the phrase rotation). Sometimes, if the schedule pans out, a team can get away with a four man rotation, and in the distant past some teams managed a three man rotation.',
    'Accroding to the statistics in the Baseball Reference page showing 2014 Major League Baseball Pitching Pitches, 745 pitchers threw 704,983 pitches in 2430 games for an average of 290 pitches per game.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Triplet

• Dataset: xs-msmarco-triplet
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.571
dot_accuracy	0.4286
manhattan_accuracy	0.5728
euclidean_accuracy	0.571
max_accuracy	0.5728

Triplet

• Dataset: xs-msmarco-triplet-train
• Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.5696
dot_accuracy	0.43
manhattan_accuracy	0.5674
euclidean_accuracy	0.5696
max_accuracy	0.5696

Training Details

Training Dataset

sentence-transformers/msmarco-msmarco-mini_lm-l-6-v3

• Dataset: sentence-transformers/msmarco-msmarco-mini_lm-l-6-v3
• Size: 100,000 training samples
• Columns: query, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative
  type	string	string	string
  details	min: 12 tokens mean: 17.05 tokens max: 36 tokens	min: 19 tokens mean: 78.68 tokens max: 212 tokens	min: 16 tokens mean: 73.73 tokens max: 205 tokens

• Samples:

  query	positive	negative
  Represent this sentence for searching relevant passages: what are the liberal arts?	liberal arts. 1. the academic course of instruction at a college intended to provide general knowledge and comprising the arts, humanities, natural sciences, and social sciences, as opposed to professional or technical subjects.	Liberal Arts Defined. The liberal arts are a set of academic disciplines that include the sciences and the humanities. When you study a liberal arts curriculum, you don't have to have one specific career goal, although you might. Instead, you'll assemble a broad foundation of knowledge that can be used in a wide spectrum of careers.
  Represent this sentence for searching relevant passages: what is the mechanism of action of fibrinolytic or thrombolytic drugs?	Baillière's Clinical Haematology. 6 Mechanism of action of the thrombolytic agents. 6 Mechanism of action of the thrombolytic agents JEFFREY I. WEITZ Fibrin formed during the haemostatic, inflammatory or tissue repair process serves a temporary role, and must be degraded to restore normal tissue function and structure.	Fibrinolytic drug. Fibrinolytic drug, also called thrombolytic drug, any agent that is capable of stimulating the dissolution of a blood clot (thrombus). Fibrinolytic drugs work by activating the so-called fibrinolytic pathway.
  Represent this sentence for searching relevant passages: what is normal plat count	78 Followers. A. Platelets are the tiny blood cells that help stop bleeding by binding together to form a clump or plug at sites of injury inside blood vessels. A normal platelet count is between 150,000 and 450,000 platelets per microliter (one-millionth of a liter, abbreviated mcL).The average platelet count is 237,000 per mcL in men and 266,000 per mcL in women.8 Followers. A. Platelets are the tiny blood cells that help stop bleeding by binding together to form a clump or plug at sites of injury inside blood vessels. A normal platelet count is between 150,000 and 450,000 platelets per microliter (one-millionth of a liter, abbreviated mcL).	In the context of blood work, PLT refers to the platelet count. Platelets are the cells that cause blood clotting and control bleeding. The normal range of platelets for adults is 3.5 to 10.5 billion cells per liter of blood, according to the Mayo Clinic. Continue Reading.

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

sentence-transformers/msmarco-msmarco-mini_lm-l-6-v3

• Dataset: sentence-transformers/msmarco-msmarco-mini_lm-l-6-v3
• Size: 5,000 evaluation samples
• Columns: query, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative
  type	string	string	string
  details	min: 12 tokens mean: 17.09 tokens max: 35 tokens	min: 21 tokens mean: 80.07 tokens max: 250 tokens	min: 16 tokens mean: 76.73 tokens max: 341 tokens

• Samples:

  query	positive	negative
  Represent this sentence for searching relevant passages: what is the name of the material harder than diamonds	Nano-material is harder than diamonds By Will Knight A material that is harder than diamond has been created in the lab, by packing together tiny “nanorods” of carbon. The new material, known as aggregated carbon nanorods (ACNR), was created by compressing and heating super-strong carbon molecules called buckyballs or carbon-60.	What material is stronger than diamond? Diamonds are famous for their hardness, but not particularly for strength,since they are easily cleaved to create the facets that make the different cuts of diamonds for setting in a ring or necklace. Many materials are stronger than diamonds, only a couple of synthesized materials are harder than diamonds.
  Represent this sentence for searching relevant passages: is pink impression a perennial tulip?	Tulip Pink Impression. close video. VIDEO. Tulip Pink Impression. The rich pink blooms of this hybrid are bound to make an impression, whether used in the landscape or as a cut flower. Robust stems and giant blooms characterise the range, and this hybrid is no exception. ‘Pink Impression’ will continue to impress thanks to the perennial potential of this range.	Tulip Pink Impression. The lustrous petals are a deep, rich rose at the center, shading to a delicate pale pink at the edge, while doing amazing things in between that include shades of both melon and sunset. Tall, strong, long-lasting and reliable, like most Darwin hybrids. An absolutely first-class Tulip.
  Represent this sentence for searching relevant passages: define: colonization	Colonization. the settlement and economic development of the uninhabited borderlands of a country (internal colonization) or the establishment of settlements (engaging primarily in agricultural activity) beyond the frontiers of a country (external colonization).	Colonization is a process by which a central system of power dominates the surrounding land and its components. The term is derived from the Latin word colere, which means to inhabit. Also, colonization refers strictly to migration, for example, to settler colonies in America or Australia, trading posts, and plantations, while colonialism deals with this, along with ruling the existing indigenous peoples of styled new territories. Colonization was linked to the spread of tens of millions fro

• 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
• num_train_epochs: 1
• warmup_ratio: 0.1
• fp16: True
• 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
• learning_rate: 5e-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: 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: 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	xs-msmarco-triplet-train_max_accuracy	xs-msmarco-triplet_max_accuracy
0	0	-	-	-	0.3648
0.016	100	1.0213	1.0078	-	0.3672
0.032	200	0.9785	0.9630	-	0.3764
0.048	300	0.9591	0.9190	-	0.4014
0.064	400	0.9088	0.8906	-	0.4392
0.08	500	0.9215	0.8707	-	0.4598
0.096	600	0.8692	0.8681	-	0.4874
0.112	700	0.8806	0.8538	-	0.4964
0.128	800	0.8801	0.8477	-	0.5106
0.144	900	0.8692	0.8414	-	0.5228
0.16	1000	0.8624	0.8391	-	0.5194
0.176	1100	0.8737	0.8397	-	0.5264
0.192	1200	0.8505	0.8344	-	0.5214
0.208	1300	0.8818	0.8358	-	0.5164
0.224	1400	0.8464	0.8269	-	0.5326
0.24	1500	0.8623	0.8291	-	0.5232
0.256	1600	0.8203	0.8407	-	0.5328
0.272	1700	0.8566	0.8257	-	0.5302
0.288	1800	0.8386	0.8198	-	0.5364
0.304	1900	0.8587	0.8172	-	0.5388
0.32	2000	0.8472	0.8233	-	0.5568
0.336	2100	0.8466	0.8188	-	0.5468
0.352	2200	0.8273	0.8190	-	0.5386
0.368	2300	0.8356	0.8183	-	0.5396
0.384	2400	0.8294	0.8156	-	0.5538
0.4	2500	0.8274	0.8168	-	0.5448
0.416	2600	0.8392	0.8093	-	0.5422
0.432	2700	0.8541	0.8087	-	0.5426
0.448	2800	0.8218	0.8086	-	0.5474
0.464	2900	0.8446	0.8062	-	0.554
0.48	3000	0.8405	0.8076	-	0.548
0.496	3100	0.8447	0.8087	-	0.553
0.512	3200	0.8453	0.8073	-	0.5536
0.528	3300	0.8371	0.8089	-	0.5504
0.544	3400	0.8548	0.8005	-	0.5516
0.56	3500	0.8162	0.8026	-	0.5572
0.576	3600	0.8577	0.7994	-	0.5558
0.592	3700	0.8289	0.7990	-	0.5526
0.608	3800	0.824	0.7967	-	0.562
0.624	3900	0.833	0.7959	-	0.5608
0.64	4000	0.8362	0.7958	-	0.5554
0.656	4100	0.8057	0.7966	-	0.5578
0.672	4200	0.8001	0.7943	-	0.5646
0.688	4300	0.8215	0.7937	-	0.5602
0.704	4400	0.8257	0.7933	-	0.5614
0.72	4500	0.8173	0.7942	-	0.5648
0.736	4600	0.8002	0.7922	-	0.5698
0.752	4700	0.8445	0.7899	-	0.5626
0.768	4800	0.825	0.7897	-	0.5592
0.784	4900	0.8151	0.7870	-	0.5696
0.8	5000	0.8223	0.7895	-	0.5676
0.816	5100	0.8235	0.7877	-	0.5656
0.832	5200	0.8355	0.7866	-	0.5688
0.848	5300	0.8218	0.7864	-	0.5672
0.864	5400	0.8384	0.7866	-	0.5652
0.88	5500	0.7988	0.7860	-	0.569
0.896	5600	0.8117	0.7867	-	0.5684
0.912	5700	0.8113	0.7861	-	0.5734
0.928	5800	0.8129	0.7860	-	0.5698
0.944	5900	0.799	0.7863	-	0.5688
0.96	6000	0.8269	0.7858	-	0.5708
0.976	6100	0.8066	0.7857	-	0.572
0.992	6200	0.8302	0.7856	-	0.5728
1.0	6250	-	-	0.5696	-

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.42.4
• PyTorch: 2.3.1+cu121
• Accelerate: 0.32.1
• Datasets: 2.20.0
• Tokenizers: 0.19.1

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",
}

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}
}