This is a sentence-transformers model finetuned from Snowflake/snowflake-arctic-embed-m. 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
Maximum Sequence Length: 512 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': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("bsmith3715/legal-ft-demo_final")
# Run inference
sentences = [
    'What muscle groups are primarily engaged during the lunge exercise described in the context?',
    "hips and we're gonna lunge it down the\nweight is going to feel a little bit\nlight we're working more stabilizers to\nstart here\nstabilizers through the hips ankles\nknees\nall the things okay lunge it down I like\nto put my foot right up against that\nedge\nto help me have a nice grip toes are off\nof the carriage\nand then coming up to squeeze up on a\nstraight standing leg squeeze up through\nthe glute\ndown\nand\nsqueeze and lift good\nthe slower you move here the more work\nyou're going to feel through those quads\nand glutes as well\nslow back\nslow up\nI know sometimes we feel like we want to\nget that heart rate going\nbut sometimes we need this slow movement\nis going to give us even more benefits\nto support us for those fast movements\nlater",
    "one\nfind that lengthened position you're\ngoing to lift up slide those shoulder\nblades down the back lift up towards the\nsky big inhale\nand exhale back and away\nleft hand to Center\nturn back towards me and lift I\napologize if you're not on the same side\nwhen you're facing me\nother arm to Center bottom arm all right\nwe're gonna flip around\nto do the same thing on the other side\nso lifting up tall\nplace that hand in front of your\nshoulder we're going up and over long\nspine and then lift shoulders down again\nup and over\nand then you use that oblique to lift up\nexhale lift\nand lengthen\nopen the spine Flex\noblique to come up\nand three\ntwo\nand one\ngood full mermaid now up and over turn\nto face the ground separate those arms",
]
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	0.45
cosine_accuracy@3	0.69
cosine_accuracy@5	0.73
cosine_accuracy@10	0.86
cosine_precision@1	0.45
cosine_precision@3	0.23
cosine_precision@5	0.146
cosine_precision@10	0.086
cosine_recall@1	0.45
cosine_recall@3	0.69
cosine_recall@5	0.73
cosine_recall@10	0.86
cosine_ndcg@10	0.6469
cosine_mrr@10	0.5798
cosine_map@100	0.5877

Training Details

Training Dataset

Unnamed Dataset

Size: 200 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 200 samples:
sentence_0 sentence_1
type string string
details
min: 12 tokens
mean: 16.91 tokens
max: 24 tokens
min: 96 tokens
mean: 159.61 tokens
max: 176 tokens

Samples:

sentence_0	sentence_1
`What type of spring is the instructor using for the workout?`	hi guys thanks for joining me today we[object Object]have a really fun really challenging[object Object]workout for you today before we get[object Object]started don't forget to like share[object Object]subscribe feel free to leave me those[object Object]super likes I really appreciate you guys[object Object]joining me for these workouts we're[object Object]going to get started setting up foot[object Object]bars all the way down I'm gonna go on to[object Object]one red Spring today which is going to[object Object]be one heavy spring on my reformer again[object Object]I'm gonna go really heavy for my arms[object Object]today if this is way too much for you[object Object]guys you can do a blue instead of a red[object Object]or a medium instead of a heavy spring[object Object]again it is going to be very heavy for[object Object]arms so feel free to change as needed we[object Object]are going to start first by straddling[object Object]your reformers your feet are on the
`What should participants do if the red spring is too heavy for them?`	hi guys thanks for joining me today we[object Object]have a really fun really challenging[object Object]workout for you today before we get[object Object]started don't forget to like share[object Object]subscribe feel free to leave me those[object Object]super likes I really appreciate you guys[object Object]joining me for these workouts we're[object Object]going to get started setting up foot[object Object]bars all the way down I'm gonna go on to[object Object]one red Spring today which is going to[object Object]be one heavy spring on my reformer again[object Object]I'm gonna go really heavy for my arms[object Object]today if this is way too much for you[object Object]guys you can do a blue instead of a red[object Object]or a medium instead of a heavy spring[object Object]again it is going to be very heavy for[object Object]arms so feel free to change as needed we[object Object]are going to start first by straddling[object Object]your reformers your feet are on the
`What is the initial position described for starting the workout on the reformers?`	are going to start first by straddling[object Object]your reformers your feet are on the[object Object]floor we're going to take our hands to[object Object]our shoulder blocks we're just going to[object Object]do a quick stretch before we get moving[object Object]so I'm going to bend my knees slightly[object Object]I'm going to inhale press my Carriage[object Object]out let my chest drop down in between my[object Object]arms and then my exhale I'm going to[object Object]tuck my pelvis around through my spine[object Object]to come back in inhale press out let[object Object]your chest drop down X tail tuck around[object Object]to come in we have two more again after[object Object]this we're going to get a really good[object Object]workout in today last one[object Object]and then round and come in all right now[object Object]once we bring it back in we're going to[object Object]take our knees onto our carriages and[object Object]then our hands are going to go into our

Loss: MatryoshkaLoss with these parameters:
json
1{ 2 "loss": "MultipleNegativesRankingLoss", 3 "matryoshka_dims": [ 4 768, 5 512, 6 256, 7 128, 8 64 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: 10
per_device_eval_batch_size: 10
num_train_epochs: 10
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: 10
per_device_eval_batch_size: 10
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: 10
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

Epoch	Step	cosine_ndcg@10
1.0	20	0.6130
2.0	40	0.6454
2.5	50	0.6445
3.0	60	0.6498
4.0	80	0.6507
5.0	100	0.6463
6.0	120	0.6433
7.0	140	0.6461
7.5	150	0.6409
8.0	160	0.6417
9.0	180	0.6425
10.0	200	0.6469

Framework Versions

Python: 3.13.2
Sentence Transformers: 4.1.0
Transformers: 4.51.3
PyTorch: 2.7.0+cpu
Accelerate: 1.7.0
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: 12 tokens mean: 16.91 tokens max: 24 tokens	min: 96 tokens mean: 159.61 tokens max: 176 tokens