This is a sentence-transformers model finetuned from BAAI/bge-large-en-v1.5. 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: 512 tokens
Output Dimensionality: 1024 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': True}) with Transformer model: 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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("ml-enthusiast13/telecom_bge_embedding_model")
# Run inference
sentences = [
    'In EVS (Enhanced Voice Services) codec, as per 3GPP standards, why is interpolation for obtaining LP parameters for each subframe performed on LSP (Line Spectral Pairs) parameters instead of directly on LP (Linear Prediction) parameters?',
    "['In the context of the EVS codec according to 3GPP specifications, why is interpolation for LP parameter retrieval in each subframe conducted on LSP parameters rather than directly on LP parameters?', 'Why does the EVS codec perform interpolation on LSP parameters instead of LP parameters when obtaining LP parameters for each subframe as per 3GPP standards?', 'For obtaining LP parameters in every subframe, why is interpolation executed on LSP parameters within the EVS codec framework defined by 3GPP?']",
    "['In the context of the EVS codec according to 3GPP specifications, why is interpolation for LP parameter retrieval in each subframe conducted directly on LP parameters instead of LSP parameters?', 'Why does the EVS codec perform interpolation on LP parameters rather than LSP parameters when obtaining LP parameters for each subframe as per 3GPP standards?', 'For obtaining LP parameters in every subframe, why is interpolation executed on both LP and LSP parameters simultaneously within the EVS codec framework defined by 3GPP?']",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Triplet

Dataset: use_triplet_all-nli-test
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9992

Training Details

Training Dataset

Unnamed Dataset

Size: 9,623 training samples
Columns: query, positive, and negative
Approximate statistics based on the first 1000 samples:
query positive negative
type string string string
details
min: 13 tokens
mean: 38.09 tokens
max: 84 tokens
min: 53 tokens
mean: 132.93 tokens
max: 313 tokens
min: 58 tokens
mean: 114.72 tokens
max: 244 tokens

Samples:

query	positive	negative
`What action does the MME take if the bearer resource allocation requested by the UE is accepted by the network?`	["Upon acceptance of the UE's requested bearer resource allocation by the network, the MME initiates either a dedicated EPS bearer context activation procedure or an EPS bearer context modification procedure.", "When the network approves the UE's request for bearer resources, the MME proceeds to start either a dedicated EPS bearer context activation process or modify an existing EPS bearer context.", 'If the requested bearer resource allocation by the UE is granted by the network, the MME begins either a dedicated EPS bearer context activation procedure or modifies an existing EPS bearer context.']	`["When the network accepts the UE's bearer resource allocation request, the MME sends a BEARER RESOURCE ALLOCATION REJECT message to the UE.", 'If the network approves the requested bearer resources from the UE, the MME stops timer T3480 and transitions to the PROCEDURE TRANSACTION INACTIVE state.', 'Upon acceptance of the bearer resource allocation by the network, the MME releases the PTI value assigned for this procedure.']`
`What is the primary purpose of conducting a single point power reference measurement at the beginning and end of the UE's OTA antenna performance testing procedure?`	['The main objective of performing a single point power reference measurement before and after testing is to track and adjust for any Tx-power drift in the DUT, which helps maintain precise sensitivity and power assessment.', 'Conducting single point power reference measurements at both the commencement and conclusion of OTA antenna performance tests serves to monitor and correct for Tx-power variations in the DUT, ensuring accurate measurement results.', 'The purpose of executing a single point power reference measurement at the start and finish of the testing procedure is to observe and adjust for any changes in the Tx-power of the DUT, thus maintaining the integrity of sensitivity and power readings.']	['The primary aim of conducting a single point power reference measurement before and after testing is to ensure that the DUT adheres to its battery usage specifications during transmission tests.', "Single point power reference measurements at both ends of the testing process are intended to confirm the DUT's compliance with maximum transmit power limits, rather than focusing on Tx-power drift.", 'The main reason for conducting a single point power reference measurement before and after testing is to calibrate the testing equipment across all frequency bands used during the procedure.']
`What is the primary purpose of the Nnef_AnalyticsExposure_Unsubscribe service operation in the 5G System (5GS)?`	['The main function of the Nnef_AnalyticsExposure_Unsubscribe service operation in the 5GS is to cancel an existing subscription for analytics data.', 'In the context of 5G System, the Nnef_AnalyticsExposure_Unsubscribe operation allows a network function consumer to terminate its current subscription to analytics information using the Subscription Correlation ID.', 'The purpose of the Nnef_AnalyticsExposure_Unsubscribe service in 5GS is to enable the NF consumer to withdraw from an existing analytics information subscription by providing the necessary Subscription Correlation ID.']	`['The Nnef_AnalyticsExposure_Unsubscribe service operation in 5G System is used to initiate a new subscription for receiving analytics data.', "In 5GS, the primary role of the Nnef_AnalyticsExposure_Unsubscribe operation is to modify an existing subscription's parameters regarding analytics information.", 'The Nnef_AnalyticsExposure_Unsubscribe service in 5GS is designed to fetch and deliver analytics information to a network function consumer.']`

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

Evaluation Dataset

Unnamed Dataset

Size: 1,203 evaluation samples
Columns: query, positive, and negative
Approximate statistics based on the first 1000 samples:
query positive negative
type string string string
details
min: 14 tokens
mean: 38.17 tokens
max: 87 tokens
min: 59 tokens
mean: 132.72 tokens
max: 316 tokens
min: 59 tokens
mean: 115.01 tokens
max: 241 tokens

Samples:

query	positive	negative
`What is the purpose of the Nnsacf_NSAC_NumOfPDUsUpdate service operation in the 5G System (5GS)?`	['The Nnsacf_NSAC_NumOfPDUsUpdate service operation in 5GS is designed to modify the count of established PDU Sessions on a specific network slice and verify if the upper limit for PDU Sessions per slice has been met.', 'In 5G System, the purpose of the Nnsacf_NSAC_NumOfPDUsUpdate service is to adjust the number of PDU Sessions on a network slice while ensuring that this count does not surpass the predefined maximum allowed sessions for that slice.', 'The function of the Nnsacf_NSAC_NumOfPDUsUpdate operation in 5GS involves updating the tally of active PDU Sessions within a network slice and checking if it has reached its maximum capacity.']	`['The Nnsacf_NSAC_NumOfPDUsUpdate service operation aims to create new PDU Sessions on a network slice without considering any limits on the number of sessions.', 'The role of the Nnsacf_NSAC_NumOfPDUsUpdate is to disconnect existing PDU Sessions from a network slice, ensuring no further sessions are active.', 'In 5GS, the Nnsacf_NSAC_NumOfPDUsUpdate service updates UE ID and PDU Session ID parameters within the NSACF without adjusting session counts.']`
`When the Discovery-Type in a Match-Report AVP is set to 'MONITORING_REQUEST_FOR_OPEN_PROSE_DIRECT_DISCOVERY', which of the following elements are MANDATORY to be included in the Match-Report?`	["In a Match-Report AVP, when Discovery-Type is 'MONITORING_REQUEST_FOR_OPEN_PROSE_DIRECT_DISCOVERY', it is mandatory to include ProSe-App-Code, ProSe-App-Id, ProSe-Validity-Timer, and ProSe-Match-Refresh-Timer.", "For the discovery type 'MONITORING_REQUEST_FOR_OPEN_PROSE_DIRECT_DISCOVERY' in a Match-Report AVP, required elements are ProSe-App-Code, ProSe-App-Id, ProSe-Validity-Timer, and ProSe-Match-Refresh-Timer.", "When the Discovery-Type is set to 'MONITORING_REQUEST_FOR_OPEN_PROSE_DIRECT_DISCOVERY' in a Match-Report AVP, inclusion of ProSe-App-Code, ProSe-App-Id, ProSe-Validity-Timer, and ProSe-Match-Refresh-Timer is mandatory."]	["In the case where Discovery-Type in a Match-Report AVP is 'MONITORING_REQUEST_FOR_OPEN_PROSE_DIRECT_DISCOVERY', only ProSe-App-Code and PC5-tech are required.", "For a Match-Report with Discovery-Type as 'MONITORING_REQUEST_FOR_OPEN_PROSE_DIRECT_DISCOVERY', it's mandatory to include ProSe-Application-Metadata, ProSe-Validity-Timer, and Discovery-Type itself.", "When the Discovery-Type in a Match-Report AVP is set to 'PROTECTED_SERVICE_ENABLING', elements like ProSe-App-Code, ProSe-App-Id, and ProSe-Match-Refresh-Timer are mandatory."]
`In a 3GPP-compliant FDD UE supporting UL OLTD and HSDPA, when calculating the Nominal CDP ratio for 16QAM modulation, what adjustment is made to the βed value?`	['In a 3GPP-compliant FDD user equipment (UE) that supports UL OLTD and HSDPA, what is the adjustment made to the βed value when calculating the Nominal CDP ratio for 16QAM modulation?', 'For 16QAM modulation in an FDD UE with UL OLTD and HSDPA capabilities, how is the βed value adjusted during the calculation of the Nominal CDP ratio?', 'When determining the Nominal CDP ratio for a 3GPP-compliant FDD UE supporting both UL OLTD and HSDPA using 16QAM modulation, what formula adjustment is applied to the βed value?']	['In a 3GPP-compliant FDD UE with UL OLTD and HSDPA, when calculating the Nominal CDP ratio for 16QAM modulation, the βed value is increased by a fixed correction factor of 0.5 dB.', 'For 16QAM modulation in an FDD UE supporting UL OLTD and HSDPA, the adjustment involves multiplying the βed value directly by the sum of all nominal CDPs without any additional operations.', 'When calculating the Nominal CDP ratio for a 3GPP-compliant FDD UE using 16QAM modulation with both UL OLTD and HSDPA, no formula involving symbol fractions is applied to adjust the βed value.']

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: 16
per_device_eval_batch_size: 16
num_train_epochs: 5
warmup_ratio: 0.1
fp16: True
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
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: 1.0
num_train_epochs: 5
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
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: True
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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	use_triplet_all-nli-test_cosine_accuracy
0.1661	100	0.3396	0.0357	-
0.3322	200	0.0571	0.0240	-
0.4983	300	0.061	0.0232	-
0.6645	400	0.0402	0.0178	-
0.8306	500	0.0438	0.0194	-
0.9967	600	0.0346	0.0144	-
1.1628	700	0.0205	0.0110	-
1.3289	800	0.0142	0.0193	-
1.4950	900	0.0098	0.0135	-
1.6611	1000	0.0053	0.0238	-
1.8272	1100	0.0054	0.0121	-
1.9934	1200	0.0058	0.0120	-
2.1595	1300	0.0028	0.0145	-
2.3256	1400	0.0047	0.0104	-
2.4917	1500	0.0023	0.0113	-
2.6578	1600	0.001	0.0086	-
2.8239	1700	0.0014	0.0072	-
2.9900	1800	0.0011	0.0070	-
3.1561	1900	0.0007	0.0095	-
3.3223	2000	0.0013	0.0069	-
3.4884	2100	0.0003	0.0099	-
3.6545	2200	0.0003	0.0103	-
3.8206	2300	0.0002	0.0097	-
3.9867	2400	0.0001	0.0117	-
4.1528	2500	0.0001	0.0115	-
4.3189	2600	0.0008	0.0113	-
4.4850	2700	0.0001	0.0117	-
4.6512	2800	0.0001	0.0114	-
4.8173	2900	0.0001	0.0117	-
4.9834	3000	0.0001	0.0117	-
5.0	3010	-	-	0.9992

Framework Versions

Python: 3.10.14
Sentence Transformers: 3.3.1
Transformers: 4.49.0
PyTorch: 2.6.0+cu118
Accelerate: 1.4.0
Datasets: 2.21.0
Tokenizers: 0.21.0

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

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}
}

	query	positive	negative
type	string	string	string
details	min: 13 tokens mean: 38.09 tokens max: 84 tokens	min: 53 tokens mean: 132.93 tokens max: 313 tokens	min: 58 tokens mean: 114.72 tokens max: 244 tokens