SentenceTransformer based on intfloat/e5-base-v2

This is a sentence-transformers model finetuned from intfloat/e5-base-v2. 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: intfloat/e5-base-v2
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': 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})
  (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 = [
    'query: Which tool can be used to qiime2 diversity pcoa-biplot?',
    'passage: qiime2 diversity pcoa-biplot. Principal Coordinate Analysis Biplot. QIIME 2: diversity pcoa-biplot ============================== Principal Coordinate Analysis Biplot   Outputs: -------- :biplot.qza: The resulting PCoA matrix.  |    Description: ------------ Project features into a principal coordinates matrix. The features used should be the features used to compute the distance matrix. It is recommended that these variables be normalized in cases of dimensionally heterogeneous physical variables.   |',
    'passage: Manipulate loom object. Add layers, or row/column attributes to a loom file. This tool allows the user to modify an existing loom data file by adding column attributes, row attributes or additional layers via tsv files.',
]
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]

Training Details

Training Dataset

Unnamed Dataset

Size: 68,840 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: 7 tokens
mean: 14.66 tokens
max: 42 tokens

min: 21 tokens
mean: 298.3 tokens
max: 512 tokens

	sentence_0	sentence_1
type	string	string
details	min: 7 tokens mean: 14.66 tokens max: 42 tokens	min: 21 tokens mean: 298.3 tokens max: 512 tokens

Samples:

sentence_0	sentence_1
`query: Tool for vcf/bcf conversion, view, subset and filter vcf/bcf files`	passage: bcftools view. VCF/BCF conversion, view, subset and filter VCF/BCF files. ===================================== bcftools view ===================================== VCF/BCF conversion, view, subset and filter VCF/BCF files. Region Selections ----------------- Regions can be specified in a VCF, BED, or tab-delimited file (the default). The columns of the tab-delimited file are: CHROM, POS, and, optionally, POS_TO, where positions are 1-based and inclusive. Uncompressed files are stored in memory, while bgzip-compressed and tabix-indexed region files are streamed. Note that sequence names must match exactly, "chr20" is not the same as "20". Also note that chromosome ordering in FILE will be respected, the VCF will be processed in the order in which chromosomes first appear in FILE. However, within chromosomes, the VCF will always be processed in ascending genomic coordinate order no matter what order they appear in FILE. Note that overlapping regions in FILE can resul...
`query: Tool for de novo assembly of rna-seq data`	`passage: Trinity. de novo assembly of RNA-Seq data. Trinity_ assembles transcript sequences from Illumina RNA-Seq data. .. _Trinity: http://trinityrnaseq.github.io`
`query: I want to das tool in Galaxy`	passage: DAS Tool. for genome-resolved metagenomics. What it does ============ DAS Tool is an automated method that integrates the results of a flexible number of binning algorithms to calculate an optimized, non-redundant set of bins from a single assembly. Inputs ====== - Bins: Tab-separated files of contig-IDs and bin-IDs. Contigs to bin file example: :: Contig_1 bin.01 Contig_8 bin.01 Contig_42 bin.02 Contig_49 bin.03 - Contigs: Assembled contigs in fasta format: :: >Contig_1 ATCATCGTCCGCATCGACGAATTCGGCGAACGAGTACCCCTGACCATCTCCGATTA... >Contig_2 GATCGTCACGCAGGCTATCGGAGCCTCGACCCGCAAGCTCTGCGCCTTGGAGCAGG... - [Optional] Proteins: Predicted proteins in prodigal fasta format. The header contains contig-ID and gene number: :: >Contig_1_1 MPRKNKKLPRHLLVIRTSAMGDVAMLPHALRALKEAYPEVKVTVATKSLFHPFFEG... >Contig_1_2 MANKIPRVPVREQDPKVRATNFEEVCYGYNVEEATLEASRCLNCKNPRCVAACPVN... Outputs ======= - Summary of output bins including quality and c...

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 16
per_device_eval_batch_size: 16
num_train_epochs: 4
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: 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: 4
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
0.1162	500	0.0921
0.2324	1000	0.0066
0.3486	1500	0.0062
0.4648	2000	0.0081
0.5810	2500	0.0073
0.6972	3000	0.0091
0.8134	3500	0.0053
0.9296	4000	0.0083
1.0458	4500	0.0073
1.1620	5000	0.0059
1.2782	5500	0.0068
1.3944	6000	0.0047
1.5106	6500	0.0077
1.6268	7000	0.0071
1.7430	7500	0.0067
1.8592	8000	0.0069
1.9754	8500	0.0077
2.0916	9000	0.0064
2.2078	9500	0.0073
2.3240	10000	0.0075
2.4402	10500	0.0049
2.5564	11000	0.0071
2.6726	11500	0.0075
2.7888	12000	0.0078
2.9050	12500	0.0086
3.0211	13000	0.0069
3.1373	13500	0.0052
3.2535	14000	0.0065
3.3697	14500	0.0066
3.4859	15000	0.0068
3.6021	15500	0.0079
3.7183	16000	0.0077
3.8345	16500	0.0066
3.9507	17000	0.0046

Framework Versions

Python: 3.12.8
Sentence Transformers: 3.4.1
Transformers: 4.49.0
PyTorch: 2.6.0+cu124
Accelerate: 1.4.0
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",
}

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}
}