SentenceTransformer based on BAAI/bge-large-en-v1.5

This is a sentence-transformers model finetuned from BAAI/bge-large-en-v1.5 on the json dataset. 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: BAAI/bge-large-en-v1.5
Maximum Sequence Length: 384 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json

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': 384, 'do_lower_case': True, 'architecture': '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("sentence_transformers_model_id")
# Run inference
sentences = [
    'Is there a specific location where I can find workspace filters?',
    '## Windows\n\nYou can access the filters of a workspace in the grid of filters.\n\nThe filter window has most properties of the filter.',
    'Find/Replace Panel\n\nSee\n\nchoosers and panels\n\nfor information on\n    displaying the Find/Replace Panel.\n\nThe Find/Replace Panel allows you to search for specific text in the\n\nproperties\n\nof the\n\ncomponents\n\nof the open\n\nworkspaces\n\nand\n\nresults workspaces\n\n.\n\nYou should enter the text for which you wish to search in the\n\nFind what\n\nedit field.\n\nYou should select the parts of the open workspaces and results workspaces within which you\n    wish to search in the tree under\n\nWithin\n\n.\n\nYou can select multiple items discontinuously by holding down the\n\nCtrl\n\nkey while clicking with the mouse.\n\nYou can use the\n\nName\n\n,\n\nFormula\n\nand\n\nAll\n     fields\n\ncheckboxes to specify whether the search should include the\n\nName\n\nproperty, the\n\nFormula\n\nproperty or all properties, respectively. You\n    must check at least one of these checkboxes so that there are some properties in which to\n    search.\n\nYou can also select further search options:\n\n* Match case- check this checkbox to perform a case-sensitive search\n* Match whole- check this checkbox to exclude matches with parts of words, including\n     names of variables and components\n* Ignore spaces- check this checkbox to ignore all white space in the properties being\n     searched\n* Ignore info fields- check this checkbox to exclude theDescription,Documentation,Last modified,Modified by,Path,Protected byandReserved byproperties.\n\nYou should press the\n\nFind\n\nbutton to start the search.\n\nAfter searching the lower pane will display the number of occurrences of the text that have\n    been found and provide a tree showing where these are. You can double-click on any of the\n    results to open that component in the Central Window, with the found item selected.\n\nYou can select items in the tree if you wish to replace the found text in these items. You\n    should then type the text to replace the found text in the\n\nReplace with\n\nedit field and click the\n\nReplace\n\nbutton.\n\nThe read-only icon\n\nnext to a tree\n    item indicates that it has been\n\nprotected\n\nand so none of its\n    text can be replaced using this feature.\n\nYou can drag or copy tree items from the Find/Replace Panel into the\n\nCentral Window\n\n.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000, -0.9967, -0.9964],
#         [-0.9967,  1.0000,  0.9994],
#         [-0.9964,  0.9994,  1.0000]])

Training Details

Training Dataset

json

Dataset: json
Size: 16,909 training samples
Columns: anchor, positive, and negative

Approximate statistics based on the first 1000 samples:

	anchor	positive	negative
type	string	string	string
details	min: 7 tokens mean: 18.63 tokens max: 53 tokens	min: 4 tokens mean: 188.63 tokens max: 384 tokens	min: 3 tokens mean: 150.13 tokens max: 384 tokens

Samples:

anchor	positive	negative
`What is the purpose of the Analyzer tab in a results workspace?`	`Analyzer The Analyzer tab of a results workspace shows how the variables in the results workspace depend on each other. If the results workspace contains sample output, the Analyzer shows these calculated results.`	Analyzer The Analyzer tool for a component shows how variables in the component depend on each other. Most components that contain variables with formulas have an Analyzer tab at the bottom of their component window. The Analyzer tab gives access to the Analyzer tool. Components with an Analyzer tab include assumption sets , data views , database views , initialization modules , layer modules , modules , MtF views , programs , projection processes , stochastic processes , and results workspaces . The Analyzer tab of a results workspace differs from the Analyzer tab of the other components and is covered separately.
`What kind of output is displayed in the Analyzer if available?`	`Analyzer The Analyzer tab of a results workspace shows how the variables in the results workspace depend on each other. If the results workspace contains sample output, the Analyzer shows these calculated results.`	`Accessing output You can view and use the output from R³S Modeler in a variety of different ways.`
`Where can I find the dependency relationships between variables in my results?`	`Analyzer The Analyzer tab of a results workspace shows how the variables in the results workspace depend on each other. If the results workspace contains sample output, the Analyzer shows these calculated results.`	Analyzer dependency diagram The dependency diagram of the Analyzer tab of a results workspace shows which variable you are currently analyzing with the variables that it depends on and the variables that depend upon it. You can double-click another variable in the dependency diagram to analyze that variable. The dependency diagram shows the value of each variable if this is available in sample output. The dependency diagram is divided into three strips of variables: * The top strip shows variables whose value depends on the value of the current variable (its dependants). * The middle strip contains the variable currently being analyzed. * The bottom strip shows variables on which the value of the current variable depends (its precedents). Each variable has a box that shows: * An icon representing the data type of the variable * A name bar that shows the name of the variable * A value box that shows the value of the variable The variable boxes are linked by arrows that show the ...

Loss: TripletLoss with these parameters:

{
    "distance_metric": "TripletDistanceMetric.EUCLIDEAN",
    "triplet_margin": 5
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 16
gradient_accumulation_steps: 2
learning_rate: 2e-05
num_train_epochs: 2
warmup_ratio: 0.05
bf16: True
dataloader_num_workers: 2
remove_unused_columns: False

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 16
per_device_eval_batch_size: 8
per_gpu_train_batch_size: None
per_gpu_eval_batch_size: None
gradient_accumulation_steps: 2
eval_accumulation_steps: None
torch_empty_cache_steps: None
learning_rate: 2e-05
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1.0
num_train_epochs: 2
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.05
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: 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: 2
dataloader_prefetch_factor: None
past_index: -1
disable_tqdm: False
remove_unused_columns: False
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: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss
0.0946	50	9.7648
0.1892	100	9.3037
0.2838	150	9.1803
0.3784	200	9.2374
0.4730	250	9.1815
0.5676	300	9.2019
0.6623	350	9.2085
0.7569	400	9.0603
0.8515	450	9.1276
0.9461	500	9.1794
1.0397	550	9.0348
1.1343	600	9.1246
1.2289	650	9.1251
1.3236	700	9.1681
1.4182	750	8.907
1.5128	800	9.0067
1.6074	850	9.1056
1.7020	900	9.0715
1.7966	950	8.9425
1.8912	1000	9.0148
1.9858	1050	9.0477

Framework Versions

Python: 3.11.11
Sentence Transformers: 5.1.1
Transformers: 4.49.0
PyTorch: 2.5.1+cu124
Accelerate: 1.10.1
Datasets: 4.1.1
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",
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}