tomaarsen/bert-base-uncased-gooaq-peft

BERT base uncased trained on GooAQ triplets

This is a sentence-transformers model finetuned from google-bert/bert-base-uncased on the sentence-transformers/gooaq dataset. 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: google-bert/bert-base-uncased
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity
• Training Dataset:
  • sentence-transformers/gooaq
• Language: en
• License: apache-2.0

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: PeftModelForFeatureExtraction 
  (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})
)

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("tomaarsen/bert-base-uncased-gooaq-peft")
# Run inference
sentences = [
    'what health services are covered by medicare?',
    'Medicare Part A hospital insurance covers inpatient hospital care, skilled nursing facility, hospice, lab tests, surgery, home health care.',
    "Elephants have the longest gestation period of all mammals. These gentle giants' pregnancies last for more than a year and a half. The average gestation period of an elephant is about 640 to 660 days, or roughly 95 weeks.",
]
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

• Dataset: gooaq-dev
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.576
cosine_accuracy@3	0.7295
cosine_accuracy@5	0.7824
cosine_accuracy@10	0.8462
cosine_precision@1	0.576
cosine_precision@3	0.2432
cosine_precision@5	0.1565
cosine_precision@10	0.0846
cosine_recall@1	0.576
cosine_recall@3	0.7295
cosine_recall@5	0.7824
cosine_recall@10	0.8462
cosine_ndcg@10	0.7089
cosine_mrr@10	0.6653
cosine_map@100	0.6709
dot_accuracy@1	0.5263
dot_accuracy@3	0.6922
dot_accuracy@5	0.7494
dot_accuracy@10	0.8175
dot_precision@1	0.5263
dot_precision@3	0.2307
dot_precision@5	0.1499
dot_precision@10	0.0818
dot_recall@1	0.5263
dot_recall@3	0.6922
dot_recall@5	0.7494
dot_recall@10	0.8175
dot_ndcg@10	0.6697
dot_mrr@10	0.6226
dot_map@100	0.6291

Training Details

Training Dataset

sentence-transformers/gooaq

• Dataset: sentence-transformers/gooaq at b089f72
• Size: 3,002,496 training samples
• Columns: question and answer
• Approximate statistics based on the first 1000 samples:

  	question	answer
  type	string	string
  details	min: 8 tokens mean: 11.84 tokens max: 31 tokens	min: 13 tokens mean: 60.69 tokens max: 149 tokens

• Samples:

  question	answer
  can dogs get pregnant when on their period?	2. Female dogs can only get pregnant when they're in heat. Some females will show physical signs of readiness – their discharge will lighten in color, and they will “flag,” or lift their tail up and to the side.
  are there different forms of als?	['Sporadic ALS is the most common form. It affects up to 95% of people with the disease. Sporadic means it happens sometimes without a clear cause.', 'Familial ALS (FALS) runs in families. About 5% to 10% of people with ALS have this type. FALS is caused by changes to a gene.']
  what is the difference between stayman and jacoby transfer?	1. The Stayman Convention is used only with a 4-Card Major suit looking for a 4-Card Major suit fit. Jacoby Transfer bids are used with a 5-Card suit looking for a 3-Card fit.

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

sentence-transformers/gooaq

• Dataset: sentence-transformers/gooaq at b089f72
• Size: 10,000 evaluation samples
• Columns: question and answer
• Approximate statistics based on the first 1000 samples:

  	question	answer
  type	string	string
  details	min: 8 tokens mean: 12.01 tokens max: 28 tokens	min: 19 tokens mean: 61.37 tokens max: 138 tokens

• Samples:

  question	answer
  is there a season 5 animal kingdom?	the good news for the fans is that the season five was confirmed by TNT in July, 2019. The season five of Animal Kingdom was expected to release in May, 2020.
  what are cmos voltage levels?	CMOS gate circuits have input and output signal specifications that are quite different from TTL. For a CMOS gate operating at a power supply voltage of 5 volts, the acceptable input signal voltages range from 0 volts to 1.5 volts for a “low” logic state, and 3.5 volts to 5 volts for a “high” logic state.
  dangers of drinking coke when pregnant?	Drinking it during pregnancy was linked to poorer fine motor, visual, spatial and visual motor abilities in early childhood (around age 3). By mid-childhood (age 7), kids whose moms drank diet sodas while pregnant had poorer verbal abilities, the study findings reported.

• 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: 128
• per_device_eval_batch_size: 128
• learning_rate: 2e-05
• num_train_epochs: 1
• warmup_ratio: 0.1
• bf16: 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: 128
• per_device_eval_batch_size: 128
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• learning_rate: 2e-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: 1
• 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: True
• 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: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	loss	gooaq-dev_cosine_map@100
0	0	-	-	0.2017
0.0000	1	2.584	-	-
0.0213	500	2.4164	-	-
0.0426	1000	1.1421	-	-
0.0639	1500	0.5215	-	-
0.0853	2000	0.3645	0.2763	0.6087
0.1066	2500	0.3046	-	-
0.1279	3000	0.2782	-	-
0.1492	3500	0.2601	-	-
0.1705	4000	0.2457	0.2013	0.6396
0.1918	4500	0.2363	-	-
0.2132	5000	0.2291	-	-
0.2345	5500	0.2217	-	-
0.2558	6000	0.2137	0.1770	0.6521
0.2771	6500	0.215	-	-
0.2984	7000	0.2057	-	-
0.3197	7500	0.198	-	-
0.3410	8000	0.196	0.1626	0.6594
0.3624	8500	0.1938	-	-
0.3837	9000	0.195	-	-
0.4050	9500	0.1895	-	-
0.4263	10000	0.186	0.1542	0.6628
0.4476	10500	0.1886	-	-
0.4689	11000	0.1835	-	-
0.4903	11500	0.1825	-	-
0.5116	12000	0.1804	0.1484	0.6638
0.5329	12500	0.176	-	-
0.5542	13000	0.1825	-	-
0.5755	13500	0.1785	-	-
0.5968	14000	0.1766	0.1436	0.6672
0.6182	14500	0.1718	-	-
0.6395	15000	0.1717	-	-
0.6608	15500	0.1674	-	-
0.6821	16000	0.1691	0.1406	0.6704
0.7034	16500	0.1705	-	-
0.7247	17000	0.1693	-	-
0.7460	17500	0.166	-	-
0.7674	18000	0.1676	0.1385	0.6721
0.7887	18500	0.1666	-	-
0.8100	19000	0.1658	-	-
0.8313	19500	0.1682	-	-
0.8526	20000	0.1639	0.1370	0.6705
0.8739	20500	0.1711	-	-
0.8953	21000	0.1667	-	-
0.9166	21500	0.165	-	-
0.9379	22000	0.1658	0.1356	0.6711
0.9592	22500	0.1665	-	-
0.9805	23000	0.1636	-	-
1.0	23457	-	-	0.6709

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• Energy Consumed: 1.051 kWh
• Carbon Emitted: 0.409 kg of CO2
• Hours Used: 2.832 hours

Training Hardware

• On Cloud: No
• GPU Model: 1 x NVIDIA GeForce RTX 3090
• CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
• RAM Size: 31.78 GB

Framework Versions

• Python: 3.11.6
• Sentence Transformers: 3.1.0.dev0
• Transformers: 4.41.2
• PyTorch: 2.3.0+cu121
• Accelerate: 0.31.0
• Datasets: 2.20.0
• Tokenizers: 0.19.1

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}
}