ChenyuEcho/hospital_emaillevel_newtrainmethod

SentenceTransformer based on Qwen/Qwen3-Embedding-0.6B

This is a sentence-transformers model finetuned from Qwen/Qwen3-Embedding-0.6B. It maps sentences & paragraphs to a 1024-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: Qwen/Qwen3-Embedding-0.6B
• Maximum Sequence Length: 32768 tokens
• Output Dimensionality: 1024 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': 32768, 'do_lower_case': False, 'architecture': 'Qwen3Model'})
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, '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': True, '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
queries = [
    "What is Patricia Vasquez\u0027s availability for a risk mitigation meeting today regarding the OR3 surgical incident?",
]
documents = [
    'Subject: Re: URGENT: Incident Report - OR3 Surgical Case\nDate: 2026-01-28T09:16:00\nFrom: David R. Park\nParticipants: Patricia Vasquez\n\nBody:\nDear Patricia,\n\nThank you for your prompt and thorough notification regarding the OR3 surgical incident. Given the seriousness of the situation, I strongly recommend convening a risk mitigation meeting with the core team as soon as possible today. It is essential that we conduct a careful review of all documentation related to the event and ensure our messaging to the patient, family, and regulatory bodies remains consistent and accurate. Please remind all involved staff to refrain from independent written or verbal statements about the incident; all communications should be directed through my office to preserve attorney-client privilege. Additionally, we must protect peer review privilege wherever it may apply.\n\nLet me know your availability for a meeting, and I will coordinate accordingly. I appreciate your diligence as we navigate this sensitive matter together.\n\nBest regards,\nDavid',
    'Subject: Re: Cultural Competency Training Compliance Concerns\nDate: 2025-10-23T12:42:00\nFrom: Maria C. Gonzalez\nParticipants: Carlos J. Rodriguez\n\nBody:\nHi Carlos,\n\nThank you for your prompt response and for taking immediate action to reinforce the cultural competency training requirements with your team. I appreciate your commitment to ensuring that all staff members complete the training by the outlined deadline. Please keep me updated if you encounter any challenges or need additional support, as maintaining regulatory compliance is a top priority for our department. Your attention to this matter is greatly appreciated.\n\nBest regards,\nMaria',
    'Subject: Fuera de la oficina: Karen M. Phillips\nDate: 2025-09-29T14:19:00\nFrom: Karen M. Phillips\nParticipants: Remitente\n\nBody:\nEstimado/a,\n\nGracias por su correo. Lamentablemente, hoy me encuentro fuera de la oficina debido a una enfermedad imprevista. Mi ausencia se debe a la necesidad de un monitoreo riguroso de mis síntomas y ajustes de medicación, prestando especial atención a posibles interacciones farmacológicas y la dosificación precisa para evitar complicaciones.\n\nSi su consulta es urgente, o requiere revisión inmediata de interacción de medicamentos o recomendaciones sobre dosificación, por favor contacte a mi colega Dr. Samuel Rivera al correo s.rivera@patientsafetyinstitute.org, quien podrá asistirle y revisar cualquier cuestión relativa a protocolos de seguridad farmacológica en mi ausencia.\n\nAgradezco su comprensión ante la situación. Revisaré y responderé su mensaje a mi regreso tan pronto me sea posible.\n\nAtentamente,\nKaren M. Phillips\n\n--\nKaren Phillips, PharmD | Pharmacy Services',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 1024] [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.7148, -0.0698,  0.0708]], dtype=torch.bfloat16)

Evaluation

Metrics

Information Retrieval

• Dataset: val_full_corpus
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.8173
cosine_accuracy@3	0.887
cosine_accuracy@5	0.9203
cosine_accuracy@10	0.9651
cosine_precision@1	0.8173
cosine_precision@3	0.2957
cosine_precision@5	0.1841
cosine_precision@10	0.0965
cosine_recall@1	0.8173
cosine_recall@3	0.887
cosine_recall@5	0.9203
cosine_recall@10	0.9651
cosine_ndcg@10	0.8867
cosine_mrr@10	0.8623
cosine_map@100	0.8644

Training Details

Training Dataset

Unnamed Dataset

• Size: 2,384 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 10 tokens mean: 26.23 tokens max: 73 tokens	min: 114 tokens mean: 175.45 tokens max: 391 tokens

• Samples:

  sentence_0	sentence_1
  Which specific crash cart items should be prioritized for procurement outreach during this week's supply review?	Subject: Re: Emergency Code Response Drill – Supply Issues Noted Date: 2025-10-02T18:56:00 From: Dr. Samantha L. Chen Participants: Carol A. Campbell Body: Hi Carol, Thank you for bringing these supply challenges to my attention. I will have my team conduct a thorough review of all crash cart inventories during our upcoming drills this week, and we will keep a detailed record of any additional shortages or supply concerns we encounter. Additionally, I’ll reach out to our procurement contacts to identify potential alternative suppliers and expedite any viable options. Please let me know if there are specific items you need us to prioritize or if you require immediate assistance coordinating with vendors. Best regards, Samantha
  Who is responsible for coordinating this week's biannual calibration with Biomedical Engineering for ADL lab equipment (electronic lifts and hand dynamometer) and what is the preferred time window?	Subject: Lab Equipment Calibration Reminder – Request for Coordination Date: 2026-01-04T17:47:00 From: Jordan P. Anderson Participants: Chloe R. Anderson Body: Hello Chloe, I wanted to bring to your attention that several pieces of adaptive equipment in our ADL lab are due for their biannual calibration, including the electronic lifts and the hand strength dynamometer. As maintaining accurate calibration is critical for patient safety and functional assessments, could you assist in coordinating a time for calibration with biomedical engineering this week? I’m hoping to minimize disruptions to our daily ADL training schedule, so your input on the best window would be invaluable. Let me know how you’d like to proceed, and thanks for your help in ensuring safe practice environments for our patients. Best, Jordan -- Jordan Anderson, OTR/L | Occupational Therapy
  Can the phlebotomy slot be moved to before 9:30 AM to resolve the overlap with the CT with IV contrast and blood draw?	Subject: OR Schedule Conflict: Contrast Protocol Coordination Needed Date: 2026-01-12T21:21:00 From: Emily T. O'Brien Participants: Carlos J. Rodriguez Body: Hi Carlos, I've noticed a scheduling overlap for tomorrow's 10:00 AM ortho case—the patient is listed for both a CT with IV contrast and a blood draw in the OR holding area at the same time. To avoid protocol delays and ensure contrast safety, could you shift the phlebotomy slot to before 9:30 AM? Let me know if that works or if you need me to adjust the CT start time instead. Thanks for collaborating on this. Best, Emily

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• 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: 3
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_ratio: None
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• enable_jit_checkpoint: False
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• use_cpu: False
• seed: 42
• data_seed: None
• bf16: False
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: -1
• ddp_backend: None
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• 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
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_for_metrics: []
• eval_do_concat_batches: True
• auto_find_batch_size: False
• full_determinism: False
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• use_cache: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	val_full_corpus_cosine_ndcg@10
1.0	149	0.8776
2.0	298	0.8866
3.0	447	0.8867

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.3
• Transformers: 5.0.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.12.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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}
}