This is a sentence-transformers model finetuned from Alibaba-NLP/gte-multilingual-base. 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: Alibaba-NLP/gte-multilingual-base
Maximum Sequence Length: 8192 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': 8192, 'do_lower_case': False}) with Transformer model: NewModel 
  (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("sentence_transformers_model_id")
# Run inference
sentences = [
    'If I save the project after clicking "Continue," will the missing files be automatically relinked or do I need to fix the file paths manually?',
    'Title: Error: no disk in drive\nAnswer: Issue This error usually occurs when an FLP (FL Studio project) looks for a sample/file in a location that doesn\'t exist (disconnected/offline drives or renamed drives) or one that doesn\'t have a disc inserted (DVD drive for example). Solution Click "Continue" and let the FLP load until the end. Save the project and the message should no longer present for that project. Note: If the error still presents itself after following the above steps try restarting your computer. Image-Line Support Team',
    'Document_title: The Playlist \nFile_name: playlist.htm\nHeading_hierarchy: [The Playlist -> Universal Clip Editing Methods]\nAnchor_id: [none]\n• Snap settings ( [Image: Horseshoe magnet icon, gray background | Ref: img_glob/flicon_snap.png] ) - Snap controls how Clips move on the Playlist grid and many of the other editing functions below. Set this to \' line \' if you want the Clips to snap to bar boundaries or to \' none \' if you want to\nfreely position & slice Clips. • Inserting Clips - To insert a Clip, select the Clip you want to place from the Clip Source menu (indicated above) OR Right-Click the Clip focus selectors (indicated above) to choose Clips of the type selected. Next, in Draw ( [Image: Gray pencil drawing/editing tool\nicon | Ref: img_glob/flicon_pencilup.png] ) or Paint ( [Image: Paintbrush dripping with liquid icon | Ref: img_glob/flicon_paint.png] ) mode, Left-click on an empty area of a Clips track where you want to place the Clip. Paint mode allows you to paint multiple instances of the Clip at once (hold the\nMouse button & drag). • Insert time into the Playlist (shift Clips to the right) - Hold ( Ctrl ), click and drag on the time-line at the top of the Playlist starting from the point you want to insert space into the timeline AND for the duration of the insertion you want. With ( Ctrl ) still held press the ( Ins\nert ) key on your keyboard. Insert time into the Playlist (slice and move Clips to the right) - Make a time selection as above, then hold ( Ctrl+Alt ) and press the ( Insert ) key on your keyboard. The slice point will be at the start of the selection. • Delete a time selection - Hold ( Ctrl ), click and\ndrag on the time-line at the\n      top of the Playlist to select the region to be deleted. Press ( Ctrl+Del ) to delete all patterns in the current selection and move Clips to the right of the selection into the gap. • Select Clips - To make a selection of more than one Clip. Use the [Image: Dashed square crop selection icon | Ref:\nimg_glob/flicon_select.png] Select tool and click and drag a rectangle around the Clips to be selected. To add Clips to this selection hold the ( Shift ) key and make further selections. OR if you are in Draw or Paint mode, hold  ( Ctrl ) key and you will drop into Select mode while the ( Ctrl ) key is\nheld. Holding the ( Ctrl+Shift ) keys will allow you to add and remove Clips to / from an existing selection. OR use the Magic lasso feature. Move your Mouse pointer in a loop that encloses the Clips you want to select (no clicking required, it\'s \'magic\' remember). Make sure you have no Clips\nselected before you perform the lasso action (Right-Click the empty Playlist to deselect all Clips and make another selection). • Zoom to area - ( Ctrl+Right-Click ) and drag over a selection of the Playlist. Un-Zoom area - ( Ctrl+Right-Click ) anywhere\n      on the Playlist (don\'t drag). • Zoom to a Clip - ( Ctrl+Shift+Right-Click ) on a Clip will maximize the zoom\n      of the Playlist to the selected Clip. To Un-Zoom Clip - ( Ctrl+Right-Click ) anywhere\n      on the Playlist (don\'t drag in this mode since that is how to zoom to selection). • Delete/Erase a Clip - Right-Click the target Clip or use the left Mouse button if you are in erase ( [Image: Circular gray "no entry" symbol icon | Ref: img_glob/flicon_delete.png] ) mode. • Copy/Paste Clips - Make\na selection, release the Mouse and any keys held, then hold the ( Shift ) key, then click on the selection and drag. A copy of the selection will follow the Mouse cursor. Release the selection when the Clips are in the alternative location. NOTE: In \' Slice tool \' mode this will initiate a slice, not\ncopy. OR Make a selection, press  ( Ctrl+C ) to copy, then scroll the Playlist so that the starting bar for the paste location is at the left edge of the Playlist and press  ( Ctrl+V ). • Duplicate selection - Make a selection and press ( Ctrl+B ). The selection, or all Clips if no selection is made, will be\nduplicated to the right of the original selection.',
]
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	0.8225
cosine_accuracy@3	0.9548
cosine_accuracy@5	0.9764
cosine_accuracy@10	0.9919
cosine_precision@1	0.8225
cosine_precision@3	0.3183
cosine_precision@5	0.1953
cosine_precision@10	0.0992
cosine_recall@1	0.8225
cosine_recall@3	0.9548
cosine_recall@5	0.9764
cosine_recall@10	0.9919
cosine_ndcg@10	0.9157
cosine_mrr@10	0.8903
cosine_map@100	0.8908

Training Details

Training Dataset

Unnamed Dataset

Size: 5,900 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: 13 tokens
mean: 24.05 tokens
max: 45 tokens
min: 22 tokens
mean: 417.54 tokens
max: 1441 tokens

Samples:

sentence_0	sentence_1
`How long does it usually take for the payment status to update from 'unpaid' to 'processed' on the site?`	Title: My payment at my online bank page shows as processed, but your site still says 'unpaid'.[object Object]Answer: Please allow our site and servers a few minutes to process your orders payment. Communication and verification between the different servers can take some time.There are different companies involved. They include your own bank, your credit card company, our 3rd party payment processors Verifone (2Checkout) & PayPal and our own registration servers. If you have the order status page still open, were you were forwarded after your purchase, hit the 'refresh' button of your browser (or the F5 key of your keyboard) to see the latest payment processing[object Object]information.FL Studio Support Team
`If refreshing the order status page doesn't update the payment status, what should I do next?`	Title: My payment at my online bank page shows as processed, but your site still says 'unpaid'.[object Object]Answer: Please allow our site and servers a few minutes to process your orders payment. Communication and verification between the different servers can take some time.There are different companies involved. They include your own bank, your credit card company, our 3rd party payment processors Verifone (2Checkout) & PayPal and our own registration servers. If you have the order status page still open, were you were forwarded after your purchase, hit the 'refresh' button of your browser (or the F5 key of your keyboard) to see the latest payment processing[object Object]information.FL Studio Support Team
`Why do delay echoes not work properly with BooBass?`	`Document_title: BooBass[object Object]File_name: plugins/BooBass.htm[object Object]Heading_hierarchy: [BooBass -> Notes & Tips][object Object]Anchor_id: [none][object Object]Since BooBass is monophonic, delay echoes will not work properly (see [object Object] ). Plugin Credits: David Billen`

Loss: MultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 20.0, 3 "similarity_fct": "cos_sim" 4}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 10
per_device_eval_batch_size: 10
num_train_epochs: 2
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: 2
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	Training Loss	cosine_ndcg@10
0.0847	50	-	0.8699
0.1695	100	-	0.8846
0.2542	150	-	0.8902
0.3390	200	-	0.9001
0.4237	250	-	0.9052
0.5085	300	-	0.9056
0.5932	350	-	0.9073
0.6780	400	-	0.9108
0.7627	450	-	0.9097
0.8475	500	0.0591	0.9157

Framework Versions

Python: 3.11.12
Sentence Transformers: 3.4.1
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",
}

MultipleNegativesRankingLoss