SentenceTransformer based on nreimers/MiniLM-L6-H384-uncased

This is a sentence-transformers model finetuned from nreimers/MiniLM-L6-H384-uncased. 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: nreimers/MiniLM-L6-H384-uncased
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': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

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 = [
    'Summarizing the Evidence on the International Trade in Illegal Wildlife',
    'The global trade in illegal wildlife is a multi-billion dollar industry that threatens biodiversity and acts as a potential avenue for invasive species and disease spread. Despite the broad-sweeping implications of illegal wildlife sales, scientists have yet to describe the scope and scale of the trade. Here, we provide the most thorough and current description of the illegal wildlife trade using 12 years of seizure records compiled by TRAFFIC, the wildlife trade monitoring network. These records comprise 967 seizures including massive quantities of ivory, tiger skins, live reptiles, and other endangered wildlife and wildlife products. Most seizures originate in Southeast Asia, a recently identified hotspot for future emerging infectious diseases. To date, regulation and enforcement have been insufficient to effectively control the global trade in illegal wildlife at national and international scales. Effective control will require a multi-pronged approach including community-scale education and empowering local people to value wildlife, coordinated international regulation, and a greater allocation of national resources to on-the-ground enforcement.',
    'This paper proposes a method to represent classifiers or learned regression functions using an OWL ontology. Also proposed are methods for finding an appropriate learned function to answer a simple query. The ontology standardizes variable names and dependence properties, so that feature values can be given by users or found on the semantic web.',
]
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: 730,454 training samples
Columns: sentence_0 and sentence_1
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1
type string string
details
min: 5 tokens
mean: 15.55 tokens
max: 41 tokens

min: 21 tokens
mean: 195.91 tokens
max: 512 tokens

	sentence_0	sentence_1
type	string	string
details	min: 5 tokens mean: 15.55 tokens max: 41 tokens	min: 21 tokens mean: 195.91 tokens max: 512 tokens

Samples:

sentence_0	sentence_1
`A parallel algorithm for constructing independent spanning trees in twisted cubes`	A long-standing conjecture mentions that a kk-connected graph GG admits kk independent spanning trees (ISTs for short) rooted at an arbitrary node of GG. An nn-dimensional twisted cube, denoted by TQnTQn, is a variation of hypercube with connectivity nn and has many features superior to those of hypercube. Yang (2010) first proposed an algorithm to construct nn edge-disjoint spanning trees in TQnTQn for any odd integer n⩾3n⩾3 and showed that half of them are ISTs. At a later stage, Wang et al. (2012) inferred that the above conjecture in affirmative for TQnTQn by providing an O(NlogN)O(NlogN) time algorithm to construct nn ISTs, where N=2nN=2n is the number of nodes in TQnTQn. However, this algorithm is executed in a recursive fashion and thus is hard to be parallelized. In this paper, we revisit the problem of constructing ISTs in twisted cubes and present a non-recursive algorithm. Our approach can be fully parallelized to make the use of all nodes of TQnTQn as processors for computation in such a way that each node can determine its parent in all spanning trees directly by referring its address and tree indices in O(logN)O(logN) time.
`A Novel Method for Separating and Locating Multiple Partial Discharge Sources in a Substation`	To separate and locate multi-partial discharge (PD) sources in a substation, the use of spectrum differences of ultra-high frequency signals radiated from various sources as characteristic parameters has been previously reported. However, the separation success rate was poor when signal-to-noise ratio was low, and the localization result was a coordinate on two-dimensional plane. In this paper, a novel method is proposed to improve the separation rate and the localization accuracy. A directional measuring platform is built using two directional antennas. The time delay (TD) of the signals captured by the antennas is calculated, and TD sequences are obtained by rotating the platform at different angles. The sequences are separated with the TD distribution feature, and the directions of the multi-PD sources are calculated. The PD sources are located by directions using the error probability method. To verify the method, a simulated model with three PD sources was established by XFdtd. Simulation results show that the separation rate is increased from 71% to 95% compared with the previous method, and an accurate three-dimensional localization result was obtained. A field test with two PD sources was carried out, and the sources were separated and located accurately by the proposed method.
`Every ternary permutation constraint satisfaction problem parameterized above average has a kernel with a quadratic number of variables`	`A ternary Permutation-CSP is specified by a subset @P of the symmetric group S"3. An instance of such a problem consists of a set of variables V and a multiset of constraints, which are ordered triples of distinct variables of V. The objective is to find a linear ordering @a of V that maximizes the number of triples whose rearrangement (under @a) follows a permutation in @P. We prove that every ternary Permutation-CSP parameterized above average has a kernel with a quadratic number of variables.`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

num_train_epochs: 5
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
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
num_train_epochs: 5
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: 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: batch_sampler
multi_dataset_batch_sampler: round_robin

Training Logs