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

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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("tabesink92/mg_alloy-snowflake-arctic-embed-l-ft-v2")
# Run inference
sentences = [
    "content='1. What does MSE stand for in the context provided?\\n2. Which technique is abbreviated as OES?' additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 24, 'prompt_tokens': 196, 'total_tokens': 220, 'completion_tokens_details': {'accepted_prediction_tokens': 0, 'audio_tokens': 0, 'reasoning_tokens': 0, 'rejected_prediction_tokens': 0}, 'prompt_tokens_details': {'audio_tokens': 0, 'cached_tokens': 0}}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_709714d124', 'finish_reason': 'stop', 'logprobs': None} id='run-798f8f65-4b52-4198-a267-dc6f61a192d0-0' usage_metadata={'input_tokens': 196, 'output_tokens': 24, 'total_tokens': 220, 'input_token_details': {'audio': 0, 'cache_read': 0}, 'output_token_details': {'audio': 0, 'reasoning': 0}}",
    '# MSE – Mean Squared Error\n\n# NRMSE – Normalized Root Mean Squared Error\n\n# OES – Optical Emission Spectrometry\n\n# PSN – Particle Stimulated Nucleation\n\n# ReLU – Rectified linear unit\n\n# RT – Room Temperature\n\n# SDAS – Secondary Dendrite Arm Spacing\n\n# SEM – Scanning Electron Microscopy\n\n# SGD – Stochastic Gradient Descent\n\n# SVM – Support Vector Machine\n\n# TD – Transverse Direction\n\n# UTS – Ultimate Tensile Strength\n\n# VAE – Variational Autoencoder\n\n# XCT – X-ray Computed Tomography\n\n# YS – Yield Strength\n\n# xxi',
    'analytical or even empirical model that could relate a specific process parameter to a selected property. Also, compared to steel or aluminum alloys, there are still not enough experimental results that guide researchers and manufacturers in selecting proper processing routes and finding the optimum combination of process parameters.\n\nMachine learning techniques, as computational tools that use data gathered from experience to improve performance, have shown to be successful in establishing such relationships between interacting variables. Machine learning methods can be used to connect sparse findings on production of magnesium alloys to provide some insight into missing pieces of information. Such a link between different parameters not only help researchers to predict desired properties or select optimum parameters, but also can be used in theoretical studies of controlling mechanisms, which can finally give rise to more robust analytical or empirical models.\n\nIn the current study, the feasibility of using machine learning approaches in establishing process-microstructure-property relationship would be put to the test. For this purpose, cast-forging process of AZ80 magnesium alloy, as a cost-effective hybrid manufacturing method, is going to be studied through advanced characterization methods. The effect of different process parameters of casting, forging, and intermediate thermo-mechanical processes and microstructural and mechanical properties of material in various stages of manufacturing are parameters of interest which will also be linked through machine learning models.\n\n# 1.2 Challenges and Opportunities\n\nTo establish a link between process, microstructure, and property, a complete study of all the controlling parameters is required. Previous studies on the viability of production of magnesium-based structural components cover a variety of parameters and properties [3,4,7,8], but since these studies have focused on different aspects of production, a complete experimental study on the effect of parameters of casting and forging processes for AZ80 magnesium alloy is selected as a basis for the establishment of this relationship. It also contributes to broadening the knowledge of cast-forging manufacturing of magnesium alloys, where thorough experimental work is unavailable.\n\nSuch an experimental study allows us to measure all properties in a controlled, consistent approach, while it also allows us to incorporate different processing parameters and, if necessary, append new variables based on the performance of machine learning models.\n\nAs the experimental part progresses through different processing steps like wedge casting, cylinder casting, homogenization, I-beam forging, optimized preform casting, and forging the final structural components, data-driven models are developed to link process parameters to microstructural features.',
]
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	0.9949
cosine_accuracy@3	1.0
cosine_accuracy@5	1.0
cosine_accuracy@10	1.0
cosine_precision@1	0.9949
cosine_precision@3	0.3333
cosine_precision@5	0.2
cosine_precision@10	0.1
cosine_recall@1	0.9949
cosine_recall@3	1.0
cosine_recall@5	1.0
cosine_recall@10	1.0
cosine_ndcg@10	0.9981
cosine_mrr@10	0.9974
cosine_map@100	0.9974

Training Details

Training Dataset

Unnamed Dataset

Size: 892 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 892 samples:
sentence_0 sentence_1
type string string
details
min: 340 tokens
mean: 371.01 tokens
max: 410 tokens

min: 29 tokens
mean: 439.26 tokens
max: 512 tokens

	sentence_0	sentence_1
type	string	string
details	min: 340 tokens mean: 371.01 tokens max: 410 tokens	min: 29 tokens mean: 439.26 tokens max: 512 tokens

Samples:

sentence_0	sentence_1
content='1. What experimental method was used to investigate the fatigue properties of extruded AZ80 magnesium alloys in the study? \n2. How does the crack initiation mechanism change with varying stress amplitudes in the AZ80 magnesium alloys?' additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 46, 'prompt_tokens': 731, 'total_tokens': 777, 'completion_tokens_details': {'accepted_prediction_tokens': 0, 'audio_tokens': 0, 'reasoning_tokens': 0, 'rejected_prediction_tokens': 0}, 'prompt_tokens_details': {'audio_tokens': 0, 'cached_tokens': 0}}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_7fcd609668', 'finish_reason': 'stop', 'logprobs': None} id='run-d649e508-ccc4-420e-b812-67719c92f2cb-0' usage_metadata={'input_tokens': 731, 'output_tokens': 46, 'total_tokens': 777, 'input_token_details': {'audio': 0, 'cache_read': 0}, 'output_token_details': {'audio': 0, 'reasoning': 0}}	# Fatigue behaviour and fractography of extruded AZ80 magnesium alloys in very high cycle regime # Kazuaki Shiozawaa *, Tomoki Kashiwagi b, Tutomu Murai c, Tooru Takahashi c b Tohoku Electric Power Co.Inc., Sendai980-8550, Japan c Sankyo-Tateyama Aluminum Industry Co. Ltd., Imizu934-8577, Japan Received 26 February 2010; revised 11 March 2010; accepted 15 March 2010 # Abstract In order to investigate the fatigue properties of extruded magnesium alloy in very high-cycle regime, rotary bending fatigue test was performed in ambient atmosphere at room temperature using the hourglass shaped specimens of AZ80 alloys extruded (F-specimen) and treated by an artificial aging after extrusion (T5-specimen). From the experimental results, both specimens show a clear step-wise S-N curve on which two knees exists. Specific stress amplitude formed the knee corresponded to the 0.2% offset proof stress of 160MPa in compression. From the detailed observation of fracture surface, small facet-like s...
content='1. What experimental method was used to investigate the fatigue properties of extruded AZ80 magnesium alloys in the study? \n2. How does the crack initiation mechanism change with varying stress amplitudes in the AZ80 magnesium alloys?' additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 46, 'prompt_tokens': 731, 'total_tokens': 777, 'completion_tokens_details': {'accepted_prediction_tokens': 0, 'audio_tokens': 0, 'reasoning_tokens': 0, 'rejected_prediction_tokens': 0}, 'prompt_tokens_details': {'audio_tokens': 0, 'cached_tokens': 0}}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_7fcd609668', 'finish_reason': 'stop', 'logprobs': None} id='run-5e3c339c-30e6-4dc7-bacb-24b9143312ad-0' usage_metadata={'input_tokens': 731, 'output_tokens': 46, 'total_tokens': 777, 'input_token_details': {'audio': 0, 'cache_read': 0}, 'output_token_details': {'audio': 0, 'reasoning': 0}}	# Fatigue behaviour and fractography of extruded AZ80 magnesium alloys in very high cycle regime # Kazuaki Shiozawaa *, Tomoki Kashiwagi b, Tutomu Murai c, Tooru Takahashi c b Tohoku Electric Power Co.Inc., Sendai980-8550, Japan c Sankyo-Tateyama Aluminum Industry Co. Ltd., Imizu934-8577, Japan Received 26 February 2010; revised 11 March 2010; accepted 15 March 2010 # Abstract In order to investigate the fatigue properties of extruded magnesium alloy in very high-cycle regime, rotary bending fatigue test was performed in ambient atmosphere at room temperature using the hourglass shaped specimens of AZ80 alloys extruded (F-specimen) and treated by an artificial aging after extrusion (T5-specimen). From the experimental results, both specimens show a clear step-wise S-N curve on which two knees exists. Specific stress amplitude formed the knee corresponded to the 0.2% offset proof stress of 160MPa in compression. From the detailed observation of fracture surface, small facet-like s...
content='1. What is the chemical composition of the AZ80 magnesium alloy used in the study?\n2. How was the T5 treatment applied to the specimens, and what was its purpose in the experiment?' additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 41, 'prompt_tokens': 617, 'total_tokens': 658, 'completion_tokens_details': {'accepted_prediction_tokens': 0, 'audio_tokens': 0, 'reasoning_tokens': 0, 'rejected_prediction_tokens': 0}, 'prompt_tokens_details': {'audio_tokens': 0, 'cached_tokens': 0}}, 'model_name': 'gpt-4o-mini-2024-07-18', 'system_fingerprint': 'fp_7fcd609668', 'finish_reason': 'stop', 'logprobs': None} id='run-1ea8b034-255f-4994-b459-394267605838-0' usage_metadata={'input_tokens': 617, 'output_tokens': 41, 'total_tokens': 658, 'input_token_details': {'audio': 0, 'cache_read': 0}, 'output_token_details': {'audio': 0, 'reasoning': 0}}	# 2. Experimental Procedures # 2.1. Testing materials and specimen The material used in this study was commercial Mg-Al-Zn magnesium alloy, AZ80. The chemical composition of these materials (in mass percentage) is 8.24Al, 0.67Zn, 0.20Mn, 0.005Fe, 0.012Si, 0.0008Cu, 0.0007Ni and balanced Mg. The bar with 16mm in diameter was extruded from a billet of 160mm diameter (an extrusion ratio of 99.4:1) under the extrusion ram speed of 1.5m/min at temperature of 623K. Hour-glass shaped specimens with a grip diameter of 10 mm and minimum diameter of 5 mm (Fig. 1) was machined from the extruded bar with the loading axis parallel to their extrusion directions. The elastic stress concentration factor, Kt, of these specimens was 1.065. On the other hand, to investigate the effect of an aging treatment on fatigue behavior, specimens were prepared from the extruded bar which was heated at 473K for 32h and then air-cooled (T5 treatment). From now on, the specimen extruded and no heat-treated is refer...

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
per_device_train_batch_size: 16
per_device_eval_batch_size: 16
num_train_epochs: 25
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: 25
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}
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: round_robin

Training Logs

Epoch	Step	Training Loss	cosine_ndcg@10
0.8929	50	-	0.9525
1.0	56	-	0.9638
1.7857	100	-	0.9801
2.0	112	-	0.9743
2.6786	150	-	0.9849
3.0	168	-	0.9812
3.5714	200	-	0.9868
4.0	224	-	0.9962
4.4643	250	-	0.9925
5.0	280	-	0.9944
5.3571	300	-	0.9944
6.0	336	-	0.9944
6.25	350	-	0.9944
7.0	392	-	0.9981
7.1429	400	-	0.9962
8.0	448	-	0.9981
8.0357	450	-	0.9981
8.9286	500	0.3646	0.9981
9.0	504	-	0.9981
9.8214	550	-	0.9981
10.0	560	-	0.9981
10.7143	600	-	0.9981
11.0	616	-	0.9981
11.6071	650	-	0.9981
12.0	672	-	0.9944
12.5	700	-	0.9981
13.0	728	-	0.9981
13.3929	750	-	0.9981
14.0	784	-	0.9981
14.2857	800	-	0.9981
15.0	840	-	0.9981
15.1786	850	-	0.9981
16.0	896	-	0.9981
16.0714	900	-	0.9981
16.9643	950	-	0.9981
17.0	952	-	0.9981
17.8571	1000	0.064	0.9981
18.0	1008	-	0.9981
18.75	1050	-	0.9981
19.0	1064	-	0.9981
19.6429	1100	-	0.9981
20.0	1120	-	0.9981
20.5357	1150	-	0.9981
21.0	1176	-	0.9981
21.4286	1200	-	0.9981
22.0	1232	-	0.9981
22.3214	1250	-	0.9981
23.0	1288	-	0.9981
23.2143	1300	-	0.9981
24.0	1344	-	0.9981
24.1071	1350	-	0.9981
25.0	1400	-	0.9981

Framework Versions

Python: 3.11.11
Sentence Transformers: 3.4.1
Transformers: 4.48.3
PyTorch: 2.5.1+cu124
Accelerate: 1.3.0
Datasets: 3.3.2
Tokenizers: 0.21.0

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",
}

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}
}

tabesink92
/

mg_alloy-snowflake-arctic-embed-l-ft-v2