bge-nuclear-finetuned-4-2 / README.md

robwhelan

Upload fine-tuned BGE embeddings model for nuclear licensing search

d065c14 verified 6 months ago

preview code

raw

history blame contribute delete

18.8 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:8622
  - loss:MultipleNegativesRankingLoss
base_model: BAAI/bge-base-en-v1.5
widget:
  - source_sentence: What is the purpose of geotechnical exploration at the PSEG Site?
    sentences:
      - >-
        The purposes of the PSEG Site geotechnical exploration and testing were
        to: - Obtain new data to meet current NRC and vendor design control
        document Tier 1 site characteristics requirements as appropriate for an
        ESPA - Confirm and demonstrate the applicability of the existing field
        data from the previous site exploration work for the existing nuclear
        plants
      - >-
        Geotechnical evaluations at the PSEG Site included assessing soil
        stratigraphy and groundwater conditions to identify potential risks and
        the suitability of the site for construction, focusing on the mechanical
        properties of subsurface materials.
      - >-
        Table 3.8-3 Illinois Inventory of Archaeological Sites Entries within
        6-miles of DNPS (Sheet 2 of 28) lists various archaeological sites and
        their statuses relevant to the regulatory considerations for the plant.
  - source_sentence: >-
      The analysis of the identified nuclides can greatly aid in determining the
      safety measures necessary for nuclear facilities.
    sentences:
      - IDENTIFIED NUCLIDES
      - 'Peak Analysis Performed on: 5/29/2019 6:14:38 AM'
      - >-
        10 CFR Part 50, Appendix H, “Reactor Vessel Material Surveillance
        Program Requirements,” requires that peak neutron fluence at the end of
        the design life of the vessel will not exceed 1.0 x 10¹⁷ n/cm² (E > 1.0
        MeV), or that reactor vessel beltline materials be monitored by a
        surveillance program.
  - source_sentence: >-
      The NRC assessment includes evaluations to determine the impact of
      specific events on safety measures.
    sentences:
      - >-
        The staff noted that the licensee performed a root cause evaluation with
        an extent of condition and extent of cause evaluation following the May
        25 scram.
      - >-
        In assessing operational events, it is crucial to differentiate between
        various types of occurrences to ensure comprehensive safety evaluations
        encompass all relevant aspects, including human factors and procedural
        adherence.
      - >-
        The reactor trip breaker indicating lights provide crucial information
        on the status of the reactor trip system during an Anticipated Transient
        Without Scram (ATWS).
  - source_sentence: >-
      Each reactor building isolation valve must remain effective during various
      operational modes.
    sentences:
      - >-
        The RHRSW System functions to remove heat from the RHR System and
        Emergency Equipment Cooling Water (EECW) System components by pumping
        water from Wheeler Reservoir through the Residual Heat Removal (RHR)
        heat exchangers and Emergency Equipment Cooling Water (EECW) System
        components and discharges back to Wheeler Reservoir.
      - Each reactor building isolation valve shall be OPERABLE.
      - Separate Condition entry is allowed for each penetration flow path.
  - source_sentence: What is the purpose of the Rapid Borate Stop Valve in Reactor Control?
    sentences:
      - >-
        CLOSE the Air Supply Isolation Valve, 12CV160 A/S, AIR SUPPLY FOR
        12CV160.
      - >-
        The NRC staff is reviewing Westinghouse’s license renewal application
        and preparing an environmental impact statement (EIS) in accordance with
        the National Environmental Policy Act of 1969.
      - >-
        Locates and discusses opening 1CV175, Rapid Borate Stop Valve by
        disengaging clutch and rotating handwheel (counterclockwise).
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: 0.9397031664848328
            name: Cosine Accuracy
          - type: cosine_accuracy
            value: 0.9387755393981934
            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 = [
    'What is the purpose of the Rapid Borate Stop Valve in Reactor Control?',
    'Locates and discusses opening 1CV175, Rapid Borate Stop Valve by disengaging clutch and rotating handwheel (counterclockwise).',
    'CLOSE the Air Supply Isolation Valve, 12CV160 A/S, AIR SUPPLY FOR 12CV160.',
]
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	0.9397

Triplet

Dataset: validation
Evaluated with TripletEvaluator

Metric	Value
cosine_accuracy	0.9388

Training Details

Training Dataset

Unnamed Dataset

Size: 8,622 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: 5 tokens mean: 14.64 tokens max: 41 tokens	min: 5 tokens mean: 43.24 tokens max: 512 tokens	min: 3 tokens mean: 31.29 tokens max: 512 tokens

Samples:

sentence_0	sentence_1	sentence_2
`What is the concentration of H-3 in µCi/ml?`	`H-3 has a concentration of 8.5E-10 µCi/ml.`	`The isotope Rb-89 has a release rate of 4.7E-05 Ci/yr.`
`gamma calibration procedures`	`Gamma Calibration: GM detectors positioned perpendicular to source for M-44-9 in which the front of probe faces source.`	`Effective calibration of GM detectors is crucial for accurate measurement. Procedures often involve using a consistent radiation source and monitoring the response of various detector models across multiple energy levels.`
`What is the function of the TAP-A program in thermal analysis?`	`The TAP-A program is applicable to both “transient and steady-state heat transfer in multidimensional systems having arbitrary geometric configurations, boundary conditions, initial conditions, and physical properties.`	`The wall panel model for the crane wall is 48 ft long with 8 axial stations each 6 ft in length.`

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: 32
per_device_eval_batch_size: 32
num_train_epochs: 5
fp16: True
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: 32
per_device_eval_batch_size: 32
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: 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}
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
0.5556	200	-	0.9272
1.0	360	-	0.9318
1.1111	400	-	0.9309
1.3889	500	0.5286	-
1.6667	600	-	0.9355
2.0	720	-	0.9378
2.2222	800	-	0.9374
2.7778	1000	0.2751	0.9397
3.0	1080	-	0.9397
0.7407	200	-	0.9374
1.0	270	-	0.9369
1.4815	400	-	0.9374
1.8519	500	0.2128	-
2.0	540	-	0.9383
2.2222	600	-	0.9388

Framework Versions

Python: 3.10.14
Sentence Transformers: 4.1.0
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}
}