antonkirk/retrieval-mpnet-dot-finetuned-gpt-4o-mini-synthetic-dataset

SentenceTransformer based on sentence-transformers/multi-qa-mpnet-base-dot-v1

This is a sentence-transformers model finetuned from sentence-transformers/multi-qa-mpnet-base-dot-v1. 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: sentence-transformers/multi-qa-mpnet-base-dot-v1
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Dot Product

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: MPNetModel 
  (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})
)

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 = [
    'nerve cell dysfunction, riboflavin deficiency',
    'Riboflavin transporter deficiency neuronopathy is a disorder that affects nerve cells (neurons). Affected individuals typically have hearing loss caused by nerve damage in the inner ear (sensorineural hearing loss) and signs of damage to other nerves.',
    'A number sign (#) is used with this entry because hyperprolinemia type I (HYRPRO1) is caused by homozygous or compound heterozygous mutation in the proline dehydrogenase gene (PRODH; 606810) on chromosome 22q11.\n\nThe PRODH gene falls within the region deleted in the 22q11 deletion syndrome, including DiGeorge syndrome (188400) and velocardiofacial syndrome (192430).\n\nDescription\n\nPhang et al. (2001) noted that prospective studies of HPI probands identified through newborn screening as well as reports of several families have suggested that it is a metabolic disorder not clearly associated with clinical manifestations. Phang et al. (2001) concluded that HPI is a relatively benign condition in most individuals under most circumstances. However, other reports have suggested that some patients have a severe phenotype with neurologic manifestations, including epilepsy and mental retardation (Jacquet et al., 2003).\n\n### Genetic Heterogeneity of Hyperprolinemia',
]
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

• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.1933
cosine_accuracy@3	0.5626
cosine_accuracy@5	0.7512
cosine_accuracy@10	0.841
cosine_precision@1	0.1933
cosine_precision@3	0.1875
cosine_precision@5	0.1502
cosine_precision@10	0.0841
cosine_recall@1	0.1933
cosine_recall@3	0.5626
cosine_recall@5	0.7512
cosine_recall@10	0.841
cosine_ndcg@10	0.512
cosine_mrr@10	0.4059
cosine_map@100	0.411
dot_accuracy@1	0.1949
dot_accuracy@3	0.5673
dot_accuracy@5	0.7571
dot_accuracy@10	0.8415
dot_precision@1	0.1949
dot_precision@3	0.1891
dot_precision@5	0.1514
dot_precision@10	0.0842
dot_recall@1	0.1949
dot_recall@3	0.5673
dot_recall@5	0.7571
dot_recall@10	0.8415
dot_ndcg@10	0.5141
dot_mrr@10	0.4084
dot_map@100	0.4136

Training Details

Training Dataset

Unnamed Dataset

• Size: 95,159 training samples
• Columns: queries and chunks
• Approximate statistics based on the first 1000 samples:

  	queries	chunks
  type	string	string
  details	min: 5 tokens mean: 15.01 tokens max: 30 tokens	min: 5 tokens mean: 158.91 tokens max: 319 tokens

• Samples:

  queries	chunks
  hypotrichosis, wiry hair, onycholysis	Green et al. (2003) reported an Australian family in which 22 members over 4 generations had progressive patterned scalp hypotrichosis and wiry hair similar to that seen in Marie Unna hereditary hypotrichosis (MUHH; 146550). Features differing from those of MUHH included absence of signs of abnormality at birth, relative sparing of body hair, distal onycholysis, and intermittent cosegregation with autosomal dominant cleft lip and palate. Five individuals had associated cleft lip and palate. Green et al. (2003) excluded linkage of the disorder in the Australian family to the MUHH locus on chromosome 8p21.
  cleft lip, cleft palate, hair loss	Green et al. (2003) reported an Australian family in which 22 members over 4 generations had progressive patterned scalp hypotrichosis and wiry hair similar to that seen in Marie Unna hereditary hypotrichosis (MUHH; 146550). Features differing from those of MUHH included absence of signs of abnormality at birth, relative sparing of body hair, distal onycholysis, and intermittent cosegregation with autosomal dominant cleft lip and palate. Five individuals had associated cleft lip and palate. Green et al. (2003) excluded linkage of the disorder in the Australian family to the MUHH locus on chromosome 8p21.
  progressive patterned scalp, autosomal dominant inheritance	Green et al. (2003) reported an Australian family in which 22 members over 4 generations had progressive patterned scalp hypotrichosis and wiry hair similar to that seen in Marie Unna hereditary hypotrichosis (MUHH; 146550). Features differing from those of MUHH included absence of signs of abnormality at birth, relative sparing of body hair, distal onycholysis, and intermittent cosegregation with autosomal dominant cleft lip and palate. Five individuals had associated cleft lip and palate. Green et al. (2003) excluded linkage of the disorder in the Australian family to the MUHH locus on chromosome 8p21.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 1,
      "similarity_fct": "dot_score"
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 8,747 evaluation samples
• Columns: queries and chunks
• Approximate statistics based on the first 1000 samples:

  	queries	chunks
  type	string	string
  details	min: 6 tokens mean: 14.71 tokens max: 31 tokens	min: 4 tokens mean: 155.81 tokens max: 305 tokens

• Samples:

  queries	chunks
  white patches, corrugated tongue, immunocompromised, Epstein-Barr virus	Not to be confused with Hairy tongue. Hairy leukoplakia Other namesOral hairy leukoplakia,[1]:385 OHL, or HIV-associated hairy leukoplakia[2] SpecialtyGastroenterology Hairy leukoplakia is a white patch on the side of the tongue with a corrugated or hairy appearance. It is caused by Epstein-Barr virus (EBV) and occurs usually in persons who are immunocompromised, especially those with human immunodeficiency virus infection/acquired immunodeficiency syndrome (HIV/AIDS). The white lesion, which cannot be scraped off, is benign and does not require any treatment, although its appearance may have diagnostic and prognostic implications for the underlying condition. Depending upon what definition of leukoplakia is used, hairy leukoplakia is sometimes considered a subtype of leukoplakia, or a distinct diagnosis. ## Contents
  HIV-associated lesions, oral hairy leukoplakia, benign white lesions, tongue appearance	Not to be confused with Hairy tongue. Hairy leukoplakia Other namesOral hairy leukoplakia,[1]:385 OHL, or HIV-associated hairy leukoplakia[2] SpecialtyGastroenterology Hairy leukoplakia is a white patch on the side of the tongue with a corrugated or hairy appearance. It is caused by Epstein-Barr virus (EBV) and occurs usually in persons who are immunocompromised, especially those with human immunodeficiency virus infection/acquired immunodeficiency syndrome (HIV/AIDS). The white lesion, which cannot be scraped off, is benign and does not require any treatment, although its appearance may have diagnostic and prognostic implications for the underlying condition. Depending upon what definition of leukoplakia is used, hairy leukoplakia is sometimes considered a subtype of leukoplakia, or a distinct diagnosis. ## Contents
  hairy leukoplakia symptoms, non-scrapable lesions, HIV/AIDS, oral lesions	Not to be confused with Hairy tongue. Hairy leukoplakia Other namesOral hairy leukoplakia,[1]:385 OHL, or HIV-associated hairy leukoplakia[2] SpecialtyGastroenterology Hairy leukoplakia is a white patch on the side of the tongue with a corrugated or hairy appearance. It is caused by Epstein-Barr virus (EBV) and occurs usually in persons who are immunocompromised, especially those with human immunodeficiency virus infection/acquired immunodeficiency syndrome (HIV/AIDS). The white lesion, which cannot be scraped off, is benign and does not require any treatment, although its appearance may have diagnostic and prognostic implications for the underlying condition. Depending upon what definition of leukoplakia is used, hairy leukoplakia is sometimes considered a subtype of leukoplakia, or a distinct diagnosis. ## Contents

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 1,
      "similarity_fct": "dot_score"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• learning_rate: 2e-05
• num_train_epochs: 15
• warmup_ratio: 0.1
• fp16: True
• load_best_model_at_end: True
• eval_on_start: 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: 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: 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: 15
• 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: True
• 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: True
• 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
• eval_on_start: True
• eval_use_gather_object: False
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Click to expand

Epoch	Step	Training Loss	loss	dot_map@100
0	0	-	1.4355	0.2271
0.1346	100	1.2599	-	-
0.2692	200	0.7627	-	-
0.4038	300	0.6061	-	-
0.5384	400	0.5632	-	-
0.6729	500	0.3965	0.4589	0.3852
0.8075	600	0.3104	-	-
0.9421	700	0.446	-	-
1.0767	800	0.4426	-	-
1.2113	900	0.4518	-	-
1.3459	1000	0.4145	0.3726	0.3964
1.4805	1100	0.4296	-	-
1.6151	1200	0.4144	-	-
1.7497	1300	0.1536	-	-
1.8843	1400	0.3425	-	-
2.0188	1500	0.3225	0.3433	0.3930
2.1534	1600	0.3529	-	-
2.2880	1700	0.3382	-	-
2.4226	1800	0.3092	-	-
2.5572	1900	0.339	-	-
2.6918	2000	0.1681	0.3633	0.4032
2.8264	2100	0.1753	-	-
2.9610	2200	0.2552	-	-
3.0956	2300	0.2549	-	-
3.2301	2400	0.2759	-	-
3.3647	2500	0.2513	0.3338	0.4066
3.4993	2600	0.258	-	-
3.6339	2700	0.2222	-	-
3.7685	2800	0.0541	-	-
3.9031	2900	0.2275	-	-
4.0377	3000	0.1919	0.3529	0.4026
4.1723	3100	0.215	-	-
4.3069	3200	0.2114	-	-
4.4415	3300	0.2153	-	-
4.5760	3400	0.2164	-	-
4.7106	3500	0.0773	0.3509	0.4090
4.8452	3600	0.1211	-	-
4.9798	3700	0.1553	-	-
5.1144	3800	0.1764	-	-
5.2490	3900	0.1953	-	-
5.3836	4000	0.1559	0.3474	0.4089
5.5182	4100	0.1686	-	-
5.6528	4200	0.1327	-	-
5.7873	4300	0.0514	-	-
5.9219	4400	0.1381	-	-
6.0565	4500	0.1445	0.3521	0.4056
6.1911	4600	0.1621	-	-
6.3257	4700	0.1365	-	-
6.4603	4800	0.1579	-	-
6.5949	4900	0.1547	-	-
6.7295	5000	0.0316	0.3895	0.4094
6.8641	5100	0.0958	-	-
6.9987	5200	0.1082	-	-
7.1332	5300	0.1379	-	-
7.2678	5400	0.1348	-	-
7.4024	5500	0.1322	0.3552	0.4100
7.5370	5600	0.1321	-	-
7.6716	5700	0.0763	-	-
7.8062	5800	0.0472	-	-
7.9408	5900	0.0989	-	-
8.0754	6000	0.1045	0.3631	0.3967
8.2100	6100	0.122	-	-
8.3445	6200	0.1057	-	-
8.4791	6300	0.1194	-	-
8.6137	6400	0.113	-	-
8.7483	6500	0.0126	0.3944	0.4116
8.8829	6600	0.089	-	-
9.0175	6700	0.0849	-	-
9.1521	6800	0.1052	-	-
9.2867	6900	0.111	-	-
9.4213	7000	0.1026	0.3665	0.4133
9.5559	7100	0.1165	-	-
9.6904	7200	0.0394	-	-
9.8250	7300	0.0443	-	-
9.9596	7400	0.0756	-	-
10.0942	7500	0.0806	0.3785	0.4090
10.2288	7600	0.103	-	-
10.3634	7700	0.0875	-	-
10.4980	7800	0.0959	-	-
10.6326	7900	0.0851	-	-
10.7672	8000	0.0073	0.3902	0.4136
10.9017	8100	0.079	-	-
11.0363	8200	0.0664	-	-
11.1709	8300	0.0766	-	-
11.3055	8400	0.084	-	-
11.4401	8500	0.0947	0.3733	0.4099
11.5747	8600	0.0906	-	-
11.7093	8700	0.0224	-	-
11.8439	8800	0.0424	-	-
11.9785	8900	0.0569	-	-
12.1131	9000	0.0697	0.3824	0.4071
12.2476	9100	0.095	-	-
12.3822	9200	0.0651	-	-
12.5168	9300	0.0756	-	-
12.6514	9400	0.065	-	-
12.7860	9500	0.0194	0.3876	0.4110
12.9206	9600	0.0595	-	-
13.0552	9700	0.0629	-	-
13.1898	9800	0.0808	-	-
13.3244	9900	0.0652	-	-
13.4590	10000	0.0802	0.3783	0.4091
13.5935	10100	0.0809	-	-
13.7281	10200	0.0111	-	-
13.8627	10300	0.0465	-	-
13.9973	10400	0.0504	-	-
14.1319	10500	0.068	0.3831	0.4071
14.2665	10600	0.0739	-	-
14.4011	10700	0.0734	-	-
14.5357	10800	0.0737	-	-
14.6703	10900	0.0379	-	-
14.8048	11000	0.0231	0.3841	0.4112
14.9394	11100	0.0493	-	-
15.0	11145	-	0.3902	0.4136

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.9
• Sentence Transformers: 3.0.1
• Transformers: 4.43.3
• PyTorch: 2.3.1+cu121
• Accelerate: 0.30.1
• Datasets: 2.19.2
• 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}
}