wwydmanski/specter2_pubmed-v0.7-full

SentenceTransformer based on allenai/specter2_base

This is a sentence-transformers model finetuned from allenai/specter2_base on the json 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: allenai/specter2_base
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • json

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})
)

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 = [
    'Algesimetric study of hypoalgesic effect',
    '[Experimental algesimetric study of the hypoalgesic effect of body acupuncture]. ',
    '[Pain analysis is basis for correct choice of therapeutic method]. ',
]
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

• Datasets: NanoNQ and NanoMSMARCO
• Evaluated with InformationRetrievalEvaluator

Metric	NanoNQ	NanoMSMARCO
cosine_accuracy@1	0.02	0.12
cosine_accuracy@3	0.06	0.3
cosine_accuracy@5	0.08	0.34
cosine_accuracy@10	0.22	0.44
cosine_precision@1	0.02	0.12
cosine_precision@3	0.02	0.1
cosine_precision@5	0.016	0.068
cosine_precision@10	0.022	0.044
cosine_recall@1	0.01	0.12
cosine_recall@3	0.05	0.3
cosine_recall@5	0.07	0.34
cosine_recall@10	0.19	0.44
cosine_ndcg@10	0.0836	0.2718
cosine_mrr@10	0.06	0.2189
cosine_map@100	0.057	0.2299

Nano BEIR

• Dataset: NanoBEIR_mean
• Evaluated with NanoBEIREvaluator

Metric	Value
cosine_accuracy@1	0.07
cosine_accuracy@3	0.18
cosine_accuracy@5	0.21
cosine_accuracy@10	0.33
cosine_precision@1	0.07
cosine_precision@3	0.06
cosine_precision@5	0.042
cosine_precision@10	0.033
cosine_recall@1	0.065
cosine_recall@3	0.175
cosine_recall@5	0.205
cosine_recall@10	0.315
cosine_ndcg@10	0.1777
cosine_mrr@10	0.1395
cosine_map@100	0.1435

Training Details

Training Dataset

json

• Dataset: json
• Size: 57,306 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 4 tokens mean: 7.57 tokens max: 25 tokens	min: 4 tokens mean: 20.36 tokens max: 78 tokens	min: 3 tokens mean: 12.38 tokens max: 49 tokens

• Samples:

  anchor	positive	negative
  Intramedullary Hemangioblastoma	Hydrocephalus: a rare initial manifestation of sporadic intramedullary hemangioblastoma : Intramedullary hemangioblastoma presenting as hydrocephalus.	Intramedullary capillary haemangioma.
  Density-based load estimation algorithm	A contact algorithm for density-based load estimation.	Density propagation based adaptive multi-density clustering algorithm.
  Herbicide Adjuvant Efficacy	The efficiency of adjuvants combined with flupyrsulfuron-methyl plus metsulfuron-methyl (Lexus XPE) on weed control.	Are herbicides a once in a century method of weed control?

• 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: 64
• per_device_eval_batch_size: 64
• gradient_accumulation_steps: 4
• learning_rate: 2e-07
• num_train_epochs: 1
• lr_scheduler_type: cosine_with_restarts
• 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: 64
• per_device_eval_batch_size: 64
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 4
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-07
• 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: cosine_with_restarts
• 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: 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	NanoNQ_cosine_ndcg@10	NanoMSMARCO_cosine_ndcg@10	NanoBEIR_mean_cosine_ndcg@10
0	0	-	0.0682	0.2560	0.1621
0.0134	1	14.8664	-	-	-
0.0268	2	14.6017	-	-	-
0.0401	3	14.8474	-	-	-
0.0535	4	14.7156	-	-	-
0.0669	5	14.5967	-	-	-
0.0803	6	14.8373	-	-	-
0.0936	7	14.7819	-	-	-
0.1070	8	14.5891	-	-	-
0.1204	9	14.5531	-	-	-
0.1338	10	14.5441	-	-	-
0.1472	11	14.5516	-	-	-
0.1605	12	14.5739	-	-	-
0.1739	13	14.5974	-	-	-
0.1873	14	14.4102	-	-	-
0.2007	15	14.3615	-	-	-
0.2140	16	14.2877	-	-	-
0.2274	17	14.2774	-	-	-
0.2408	18	14.4985	-	-	-
0.2542	19	14.2307	-	-	-
0.2676	20	14.3657	-	-	-
0.2809	21	14.3261	-	-	-
0.2943	22	14.2946	-	-	-
0.3077	23	14.2311	-	-	-
0.3211	24	14.0789	-	-	-
0.3344	25	13.9392	0.0764	0.2652	0.1708
0.3478	26	14.0972	-	-	-
0.3612	27	14.0966	-	-	-
0.3746	28	13.9205	-	-	-
0.3880	29	13.8919	-	-	-
0.4013	30	14.1233	-	-	-
0.4147	31	14.1351	-	-	-
0.4281	32	14.1106	-	-	-
0.4415	33	14.166	-	-	-
0.4548	34	13.7817	-	-	-
0.4682	35	14.0178	-	-	-
0.4816	36	13.8457	-	-	-
0.4950	37	14.074	-	-	-
0.5084	38	13.9665	-	-	-
0.5217	39	13.9726	-	-	-
0.5351	40	13.8546	-	-	-
0.5485	41	13.9037	-	-	-
0.5619	42	13.6977	-	-	-
0.5753	43	14.0445	-	-	-
0.5886	44	13.93	-	-	-
0.6020	45	13.7835	-	-	-
0.6154	46	13.819	-	-	-
0.6288	47	13.6248	-	-	-
0.6421	48	13.846	-	-	-
0.6555	49	13.6079	-	-	-
0.6689	50	13.6848	0.0836	0.2724	0.1780
0.6823	51	13.668	-	-	-
0.6957	52	13.5784	-	-	-
0.7090	53	13.7519	-	-	-
0.7224	54	13.6455	-	-	-
0.7358	55	13.6757	-	-	-
0.7492	56	13.5647	-	-	-
0.7625	57	13.7072	-	-	-
0.7759	58	13.5603	-	-	-
0.7893	59	13.6437	-	-	-
0.8027	60	13.6656	-	-	-
0.8161	61	13.479	-	-	-
0.8294	62	13.5965	-	-	-
0.8428	63	13.6793	-	-	-
0.8562	64	13.6121	-	-	-
0.8696	65	13.841	-	-	-
0.8829	66	13.4793	-	-	-
0.8963	67	13.5875	-	-	-
0.9097	68	13.4063	-	-	-
0.9231	69	13.6365	-	-	-
0.9365	70	13.4696	-	-	-
0.9498	71	13.5018	-	-	-
0.9632	72	13.5956	-	-	-
0.9766	73	13.3945	-	-	-
0.9900	74	13.5684	0.0836	0.2718	0.1777

Framework Versions

• Python: 3.12.3
• Sentence Transformers: 3.3.1
• Transformers: 4.49.0
• PyTorch: 2.5.1
• Accelerate: 1.2.1
• Datasets: 2.19.0
• 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}
}