khoj-ai/timely-arctic-small

Technical Report and Model Pipeline

To access our technical report and model pipeline scripts visit our github

SentenceTransformer based on Snowflake/snowflake-arctic-embed-s

This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-s. 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-s
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 384 tokens
• 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': 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("sentence_transformers_model_id")
# Run inference
sentences = [
    'Represent this sentence for searching relevant passages: Philadelphia Business Journal 01/30/83',
    'The Philadelphia Business Journal is a diversified business media company in Philadelphia, Pennsylvania, publishing daily stories on its website and social networks, and a weekly edition available in print and online. It is published by the American City Business Journals. See also List of newspapers in Pennsylvania References External links Business newspapers published in the United States Newspapers published in Philadelphia Jan 30 1983\n',
    "Have you just bought your brand new Nintendo Wii console? Are you gutted that you can't get Wii Connect 24 in your country of residence? This article will resolve this problem, so you can surf the Internet on your Wii!\n",
]
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]

Training Details

Training Dataset

Unnamed Dataset

• Size: 55,736 training samples
• Columns: anchors and positive
• Approximate statistics based on the first 1000 samples:

  	anchors	positive
  type	string	string
  details	min: 14 tokens mean: 20.25 tokens max: 33 tokens	min: 15 tokens mean: 47.2 tokens max: 75 tokens

• Samples:

  anchors	positive
  Represent this sentence for searching relevant passages: are bugs attracted to citronella November 10?	Citronella is naturally occurring oil that repels insects. ... “Citronella oil is repellent to mosquitoes to a degree, but the amount being put out by a candle isn't going to be very effective,” Eric Hoffer, president of Hoffer Pest, told TODAY Home. 11/10
  Represent this sentence for searching relevant passages: are bugs attracted to citronella 11/10/09?	Citronella is naturally occurring oil that repels insects. ... “Citronella oil is repellent to mosquitoes to a degree, but the amount being put out by a candle isn't going to be very effective,” Eric Hoffer, president of Hoffer Pest, told TODAY Home. 11/10/09
  Represent this sentence for searching relevant passages: are bugs attracted to citronella Jan 15?	Citronella is naturally occurring oil that repels insects. ... “Citronella oil is repellent to mosquitoes to a degree, but the amount being put out by a candle isn't going to be very effective,” Eric Hoffer, president of Hoffer Pest, told TODAY Home. 01/15

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

• Size: 1,000 evaluation samples
• Columns: anchors and positive
• Approximate statistics based on the first 1000 samples:

  	anchors	positive
  type	string	string
  details	min: 11 tokens mean: 21.67 tokens max: 39 tokens	min: 15 tokens mean: 67.0 tokens max: 512 tokens

• Samples:

  anchors	positive
  Represent this sentence for searching relevant passages: 4/23 Socialize With Someone Who Is Losing Interest in You	04/23 It can hurt to realize that someone you care about is losing interest in you. If a friend has stopped calling you and no longer makes plans to hang out, your first instinct might be to contact them more frequently or to ignore them in return.
  Represent this sentence for searching relevant passages: Alathur taluk, Perambalur 04/19/29	Alathur taluk is a taluk in Perambalur district in the Indian state of Tamil Nadu. It was created by former chief minister J.Jayalalithaa for issues of population increase. Kunnam taluk was bifurcated to form this new taluk. Villages There are 39 villages in Alathur taluk excluding the headquarters Alathur. References Perambalur district Taluks of Perambalur district 2029 Apr 19
  Represent this sentence for searching relevant passages: 01/04 how much weight does a baby gain in the first month?	01/04 During their first month, most newborns gain weight at a rate of about 1 ounce (30 grams) per day. They generally grow in height about 1 to 1½ inches (2.54 to 3.81 centimeters) during the first month. Many newborns go through a period of rapid growth when they are 7 to 10 days old and again at 3 and 6 weeks.

• 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: 128
• per_device_eval_batch_size: 128
• learning_rate: 2e-05
• weight_decay: 0.01
• num_train_epochs: 1
• warmup_ratio: 0.1
• warmup_steps: 400
• bf16: True
• torch_compile: True
• torch_compile_backend: inductor
• 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: 128
• per_device_eval_batch_size: 128
• 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: 0.01
• 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: 400
• 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: True
• 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: True
• torch_compile_backend: inductor
• 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
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss
0.0023	1	2.3154	-
0.0229	10	2.3237	-
0.0459	20	2.4036	-
0.0688	30	2.3314	-
0.0917	40	2.3171	-
0.1147	50	2.2891	-
0.0023	1	2.2343	-
0.0229	10	2.2256	-
0.0459	20	2.2924	-
0.0688	30	2.2354	-
0.0917	40	2.2281	-
0.1147	50	2.2018	-
0.1376	60	2.2377	-
0.1606	70	2.2001	-
0.1835	80	2.158	-
0.2064	90	2.1405	-
0.2294	100	2.0916	-
0.2523	110	2.0374	-
0.2752	120	2.0492	-
0.2982	130	1.9824	-
0.3211	140	1.9571	-
0.3440	150	1.8317	-
0.3670	160	1.7183	-
0.3899	170	1.5928	-
0.4128	180	1.5695	-
0.4358	190	1.4592	-
0.4587	200	1.2667	0.2031
0.4817	210	1.3865	-
0.5046	220	1.2924	-
0.5275	230	1.3042	-
0.5505	240	1.4393	-
0.5734	250	1.3402	-
0.5963	260	1.1939	-
0.6193	270	1.1795	-
0.6422	280	1.1012	-
0.6651	290	1.0379	-
0.6881	300	0.9865	-
0.7110	310	0.9088	-
0.7339	320	0.9132	-
0.7569	330	0.8819	-
0.7798	340	0.8631	-
0.8028	350	1.4084	-
0.8257	360	1.325	-
0.8486	370	1.2373	-
0.8716	380	1.1881	-
0.8945	390	1.1656	-
0.9174	400	0.7767	0.0607
0.9404	410	0.1511	-
0.9633	420	0.1439	-
0.9862	430	0.1216	-

Framework Versions

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.43.3
• PyTorch: 2.4.0+cu121
• Accelerate: 0.33.0
• 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}
}