bge-nuclear-finetuned-5 / README.md

robwhelan

Upload fine-tuned BGE embeddings model for nuclear licensing search

58c1d9a verified 6 months ago

preview code

raw

history blame contribute delete

21.1 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:3157
  - loss:TripletLoss
  - dataset_size:2525
  - loss:MultipleNegativesRankingLoss
base_model: BAAI/bge-base-en-v1.5
widget:
  - source_sentence: >-
      What is the role of the Public Information Officer during a radiological
      emergency?
    sentences:
      - >-
        The Public Information Officer (PIO), assigned duties at the Emergency
        Operations Center (EOC), and the Parish Spokesperson will be responsible
        for implementation of this procedure.
      - >-
        Be prepared to report to the Parish EOC if requested by the Director of
        the Office of Homeland Security and Emergency Preparedness.
      - >-
        HRI may proceed with its planned mining-related activities in these
        areas to the extent authorized by its NRC Materials License SUA-1508.
  - source_sentence: >-
      The NRC also regulates the licensing and reporting obligations for
      materials that are byproducts in the medical field.
    sentences:
      - Parts 30, 31, 32 and 150
      - >-
        Exemptions From Licensing, General Licenses, and Distribution of
        Byproduct Material: Licensing and Reporting Requirements
      - >-
        Containment vessel (CNV) pressure/temperature response analysis method,
        similar to method used in DCA technical report, responds to LOCA pipe
        break, secondary line breaks, IORV events, or inadvertent ECCS
        actuation.
  - source_sentence: What is the aging management program for concrete in nuclear facilities?
    sentences:
      - >-
        Further evaluation is required to determine if a plant-specific aging
        management program is needed.
      - >-
        The DCPP Structures Monitoring AMP (B.2.3.33) is credited with managing
        cracking due to reaction with aggregates (such as ASR), for DCPP group
        1, 3, 4, 5, and 7 structures, including inaccessible areas.
      - >-
        The Survey Units listed in Figure 4 measure various acreage sizes,
        detailing the land area covered by each unit within the
        Non-Industrialized section of the site.
  - source_sentence: >-
      What is the purpose of an emergency core cooling system in nuclear
      reactors?
    sentences:
      - >-
        The Commission issued Staff Requirements Memorandum (SRM) SECY-10-0113
        directing the staff to consider alternative options for resolving
        GSI-191 (Reference 5).
      - >-
        preclude the formulation or implementation of reasonable and prudent
        alternatives to avoid jeopardizing the continued existence of endangered
        or threatened species or destroying or modifying critical habitat
        [Section 7(d)].
      - >-
        ECCS must be designed so that calculated cooling performance following
        postulated loss-of-coolant accidents conforms to the criteria set forth
        in paragraph (b) of this section.
  - source_sentence: corrosion related to nuclear components
    sentences:
      - >-
        This alternative is requested for the duration of the Brunswick Steam
        Electric Plant, Units 1 & 2, Third Ten-Year Containment Inservice
        Inspection Interval, which is currently scheduled to end no later than
        May 10, 2028.
      - >-
        *Note: Initiation of Reactor Enclosure isolation starts Reactor
        Enclosure Recirculation System (RERS) and SGTS. Ref: UFSAR 6.2.3.2.3*
      - >-
        In the proposed alternative by the Owner (Duke Energy), corrosion or
        erosion that has reduced the component wall thickness to less than 145%
        of the minimum design wall thickness will be considered a relevant
        condition that will require evaluation or corrective measures to the
        extent necessary to meet the acceptance standards of IWE-3500 prior to
        continued service.
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy
model-index:
  - name: SentenceTransformer based on BAAI/bge-base-en-v1.5
    results:
      - task:
          type: triplet
          name: Triplet
        dataset:
          name: validation
          type: validation
        metrics:
          - type: cosine_accuracy
            value: 0.8797468543052673
            name: Cosine Accuracy
          - type: cosine_accuracy
            value: 0.9556962251663208
            name: Cosine Accuracy
          - type: cosine_accuracy
            value: 0.9588607549667358
            name: Cosine Accuracy

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

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. 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: BAAI/bge-base-en-v1.5
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': True}) with Transformer model: BertModel 
  (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.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("sentence_transformers_model_id")
# Run inference
sentences = [
    'corrosion related to nuclear components',
    'In the proposed alternative by the Owner (Duke Energy), corrosion or erosion that has reduced the component wall thickness to less than 145% of the minimum design wall thickness will be considered a relevant condition that will require evaluation or corrective measures to the extent necessary to meet the acceptance standards of IWE-3500 prior to continued service.',
    'This alternative is requested for the duration of the Brunswick Steam Electric Plant, Units 1 & 2, Third Ten-Year Containment Inservice Inspection Interval, which is currently scheduled to end no later than May 10, 2028.',
]
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

Triplet

Dataset: validation
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.8797

Triplet

Dataset: validation
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9557

Triplet

Dataset: validation
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9589

Training Details

Training Dataset

Unnamed Dataset

Size: 2,525 training samples
Columns: sentence_0, sentence_1, and sentence_2

Approximate statistics based on the first 1000 samples:

	sentence_0	sentence_1	sentence_2
type	string	string	string
details	min: 4 tokens mean: 15.01 tokens max: 44 tokens	min: 6 tokens mean: 45.67 tokens max: 512 tokens	min: 4 tokens mean: 33.2 tokens max: 228 tokens

Samples:

sentence_0	sentence_1	sentence_2
`What is the significance of groundwater monitoring wells in nuclear safety assessments?`	`Locations of monitoring wells B-22 and B-36 that showed groundwater “mounding” supported by maps of the stormwater drainage system near the wells (Related to Need GW-5)`	`To the extent that these components of the intake and discharge systems are accessible/viewable: - Submerged multi-port intake and intake tunnel, including depiction of location of intake - Traveling screens - Service water pumphouse - Emergency service water forebay - Cooling towers and cooling tower basin - Discharge tunnel and discharge outfall`
`An analysis of the stresses experienced by reactor vessel studs is crucial for evaluating operational safety.`	`Table 2. Calculation of Primary Stresses in Reactor Vessel Studs, Two Studs Out of Service provides detailed calculations of primary stresses for each stud in the reactor vessel.`	`Regulatory filings often require extensive documentation to demonstrate compliance with safety protocols.`
`An assessment was carried out to determine the potential for liquefaction at the EGC ESP Site.`	`An evaluation of liquefaction potential was conducted at the EGC ESP Site.`	`The static groundwater table within the Illinois till is approximately 30 ft below the ground surface, but that there are shallower perched groundwater layers closer to the surface.`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 24
per_device_eval_batch_size: 24
num_train_epochs: 5
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: 24
per_device_eval_batch_size: 24
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: 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}
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	validation_cosine_accuracy
2.5253	500	4.2171	-
0.6329	100	-	0.9177
1.0	158	-	0.8592
1.2658	200	-	0.8972
1.8987	300	-	0.875
2.0	316	-	0.8940
2.5316	400	-	0.8734
3.0	474	-	0.8956
3.1646	500	3.985	0.8813
3.7975	600	-	0.8703
4.0	632	-	0.9003
4.4304	700	-	0.8797
5.0	790	-	0.8797
0.6329	100	-	0.8228
1.0	158	-	0.9383
1.2658	200	-	0.9541
1.8987	300	-	0.9573
2.0	316	-	0.9589
2.5316	400	-	0.9541
3.0	474	-	0.9525
3.1646	500	2.0222	0.9525
3.7975	600	-	0.9541
4.0	632	-	0.9557
4.4304	700	-	0.9573
5.0	790	-	0.9557
0.9434	100	-	0.9509
1.0	106	-	0.9525
1.8868	200	-	0.9541
2.0	212	-	0.9573
2.8302	300	-	0.9557
3.0	318	-	0.9589

Framework Versions

Python: 3.10.14
Sentence Transformers: 4.0.2
Transformers: 4.51.3
PyTorch: 2.2.2
Accelerate: 1.6.0
Datasets: 3.5.0
Tokenizers: 0.21.1

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}
}