This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-l. It maps sentences & paragraphs to a 1024-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-l
Maximum Sequence Length: 512 tokens
Output Dimensionality: 1024 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': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("kamkol/ab_testing_finetuned_2_arctic_ft-711c8143-04b7-4ecb-8e01-8847274c1d9c")
# Run inference
sentences = [
    'What are the estimated monthly rates of first-party cookie deletion among U.S. Internet users according to comScore and Anirban Dasgupta et al.?',
    'Tracking on the Web: Privacy and Security Implication [23]. \nCookie deletion rates are hard to estimate.  Using a panel \nof 400,000 home PCs, comScore estimated that 31 percent of \nU.S. Internet users cleared their first -party cookies during a \nmonth [24].   Anirban Dasgupta etal. [25] showed similar \nlevels of cookie clearing (25% -33% monthly, depending on \ngeography) based on the Yahoo toolbar.  Such rates imply \nthat long studies (e.g., months), where users are identified \nbased on cookies could have a selection bi as problem – as',
    'both directions: when we show fewer ads, users increase \nengagement, and when we show more ads, users decrease \nengagement or abandon. If users abandon at different rates \nbetween Control and Treatment, the remaining surviving \npopulation is different, and the conclusions can be \ncompletely wrong.  For example, the users generating the \nmost revenue may get annoyed with more ads and abandon, \nleaving a surviving population with lower Revenue/user. \n3. Not taking into account selection bias [34].  \n In an online longitudinal study that relies on cookies for \nidentification, there is likely to be a significant attrition due',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# 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	0.5
cosine_accuracy@3	0.7419
cosine_accuracy@5	0.7903
cosine_accuracy@10	0.8629
cosine_precision@1	0.5
cosine_precision@3	0.2473
cosine_precision@5	0.1581
cosine_precision@10	0.0863
cosine_recall@1	0.5
cosine_recall@3	0.7419
cosine_recall@5	0.7903
cosine_recall@10	0.8629
cosine_ndcg@10	0.6895
cosine_mrr@10	0.6334
cosine_map@100	0.6397

Training Details

Training Dataset

Unnamed Dataset

Size: 250 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 250 samples:
sentence_0 sentence_1
type string string
details
min: 5 tokens
mean: 24.08 tokens
max: 51 tokens
min: 13 tokens
mean: 123.64 tokens
max: 151 tokens

Samples:

sentence_0	sentence_1
`What are some common pitfalls associated with long-term online controlled experiments as discussed by Dmitriev et al. in their IEEE Big Data paper?`	Appears in IEEE Big Data. Paper available at [object Object] [object Object] [object Object]Pitfalls of Long-Term Online Controlled Experiments [object Object]Pavel Dmitriev, Brian Frasca, Somit Gupta, Ron Kohavi, Garnet Vaz [object Object]Analysis and Experimentation [object Object]Microsoft Corporation [object Object]Redmond, WA 98052, USA [object Object]{padmitri,brianfra,sogupta,ronnyk,gavaz}@microsoft.com [object Object] [object Object] [object Object]Abstract—Online controlled experiments (e.g., A/B tests) are [object Object]now regularly used to guide product development and [object Object]accelerate innovation in software. Product ideas are evaluated
`What are online controlled experiments used for in product development according to the context?`	Analysis and Experimentation [object Object]Microsoft Corporation [object Object]Redmond, WA 98052, USA [object Object]{padmitri,brianfra,sogupta,ronnyk,gavaz}@microsoft.com [object Object] [object Object] [object Object]Abstract—Online controlled experiments (e.g., A/B tests) are [object Object]now regularly used to guide product development and [object Object]accelerate innovation in software. Product ideas are evaluated [object Object]as scientific hypotheses, and tested on web sit es, mobile [object Object]applications, desktop applications, services, and operating [object Object]system features. [object Object]One of the key challenges for organizations that run [object Object]controlled experiments is to select an Overall Evaluation
`What is one of the key challenges organizations face when running online controlled experiments?`	Abstract—Online controlled experiments (e.g., A/B tests) are [object Object]now regularly used to guide product development and [object Object]accelerate innovation in software. Product ideas are evaluated [object Object]as scientific hypotheses, and tested on web sit es, mobile [object Object]applications, desktop applications, services, and operating [object Object]system features. [object Object]One of the key challenges for organizations that run [object Object]controlled experiments is to select an Overall Evaluation [object Object]Criterion (OEC), i.e., the criterion by which to evaluat e the [object Object]different variants. The difficulty is that short -term changes to [object Object]metrics may not predict the long-term impact of a change. For

Loss: MatryoshkaLoss with these parameters:
json
1{ 2 "loss": "MultipleNegativesRankingLoss", 3 "matryoshka_dims": [ 4 1024, 5 768, 6 512, 7 256, 8 128 9 ], 10 "matryoshka_weights": [ 11 1, 12 1, 13 1, 14 1, 15 1 16 ], 17 "n_dims_per_step": -1 18}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
num_train_epochs: 100
multi_dataset_batch_sampler: round_robin

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: 5e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 100
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.0
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}
tp_size: 0
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
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: round_robin

Training Logs

Click to expand

Epoch	Step	Training Loss	cosine_ndcg@10
1.0	16	-	0.6381
2.0	32	-	0.6800
3.0	48	-	0.6948
3.125	50	-	0.6967
4.0	64	-	0.6999
5.0	80	-	0.7013
6.0	96	-	0.6831
6.25	100	-	0.6885
7.0	112	-	0.6893
8.0	128	-	0.7012
9.0	144	-	0.6893
9.375	150	-	0.6900
10.0	160	-	0.6860
11.0	176	-	0.6840
12.0	192	-	0.7025
12.5	200	-	0.6775
13.0	208	-	0.6786
14.0	224	-	0.7115
15.0	240	-	0.7309
15.625	250	-	0.7355
16.0	256	-	0.7339
17.0	272	-	0.7246
18.0	288	-	0.7249
18.75	300	-	0.7244
19.0	304	-	0.7183
20.0	320	-	0.6952
21.0	336	-	0.6968
21.875	350	-	0.7299
22.0	352	-	0.7278
23.0	368	-	0.7122
24.0	384	-	0.6990
25.0	400	-	0.6857
26.0	416	-	0.6824
27.0	432	-	0.6905
28.0	448	-	0.6494
28.125	450	-	0.6461
29.0	464	-	0.6747
30.0	480	-	0.6761
31.0	496	-	0.6958
31.25	500	0.8459	0.6855
32.0	512	-	0.6940
33.0	528	-	0.7156
34.0	544	-	0.7284
34.375	550	-	0.7281
35.0	560	-	0.7288
36.0	576	-	0.6940
37.0	592	-	0.6992
37.5	600	-	0.6948
38.0	608	-	0.6991
39.0	624	-	0.6931
40.0	640	-	0.6893
40.625	650	-	0.6905
41.0	656	-	0.6960
42.0	672	-	0.7133
43.0	688	-	0.7137
43.75	700	-	0.7055
44.0	704	-	0.7077
45.0	720	-	0.7215
46.0	736	-	0.7056
46.875	750	-	0.7030
47.0	752	-	0.6996
48.0	768	-	0.6985
49.0	784	-	0.6878
50.0	800	-	0.7042
51.0	816	-	0.7097
52.0	832	-	0.7067
53.0	848	-	0.6977
53.125	850	-	0.6972
54.0	864	-	0.6999
55.0	880	-	0.6963
56.0	896	-	0.7006
56.25	900	-	0.6971
57.0	912	-	0.7008
58.0	928	-	0.7051
59.0	944	-	0.7034
59.375	950	-	0.7031
60.0	960	-	0.6978
61.0	976	-	0.6923
62.0	992	-	0.6966
62.5	1000	0.2227	0.6989
63.0	1008	-	0.6941
64.0	1024	-	0.7020
65.0	1040	-	0.7024
65.625	1050	-	0.7050
66.0	1056	-	0.7079
67.0	1072	-	0.7059
68.0	1088	-	0.7200
68.75	1100	-	0.7207
69.0	1104	-	0.7177
70.0	1120	-	0.7008
71.0	1136	-	0.6948
71.875	1150	-	0.6883
72.0	1152	-	0.6858
73.0	1168	-	0.6894
74.0	1184	-	0.6961
75.0	1200	-	0.6934
76.0	1216	-	0.6894
77.0	1232	-	0.6830
78.0	1248	-	0.6892
78.125	1250	-	0.6903
79.0	1264	-	0.6903
80.0	1280	-	0.6911
81.0	1296	-	0.7021
81.25	1300	-	0.7032
82.0	1312	-	0.7032
83.0	1328	-	0.6984
84.0	1344	-	0.6872
84.375	1350	-	0.6863
85.0	1360	-	0.6863
86.0	1376	-	0.6879
87.0	1392	-	0.6909
87.5	1400	-	0.6910
88.0	1408	-	0.6911
89.0	1424	-	0.6894
90.0	1440	-	0.6892
90.625	1450	-	0.6952
91.0	1456	-	0.6954
92.0	1472	-	0.6924
93.0	1488	-	0.6895
93.75	1500	0.1783	0.6895
94.0	1504	-	0.6895
95.0	1520	-	0.6895
96.0	1536	-	0.6895
96.875	1550	-	0.6895
97.0	1552	-	0.6895
98.0	1568	-	0.6895
99.0	1584	-	0.6895
100.0	1600	-	0.6895

Framework Versions

Python: 3.11.12
Sentence Transformers: 4.1.0
Transformers: 4.51.3
PyTorch: 2.6.0+cu124
Accelerate: 1.6.0
Datasets: 3.6.0
Tokenizers: 0.21.1

Citation

BibTeX

Sentence Transformers

bibtex

@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

bibtex

@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

bibtex

@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}
}

	sentence_0	sentence_1
type	string	string
details	min: 5 tokens mean: 24.08 tokens max: 51 tokens	min: 13 tokens mean: 123.64 tokens max: 151 tokens