ChenyuEcho/ai-test

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 = [
    "How were high-value business expenses reviewed for required manager approvals?",
]
documents = [
    'Subject: Reminder: Year-End Expense Submission Deadline and T&E Policy Requirements\nDate: 2025-08-25T16:44:00\nFrom: Maria Santos\nParticipants: Carlos Delgado; Ricardo Rick Mendez; Ana Lucia Vega; Miguel Torres; Carmen Ortiz\n\nBody:\nDear Team,\n\nAs we approach year-end, I want to remind everyone that all 2024 business expense reports must be submitted in Concur by December 18th. Per ASI’s Travel and Entertainment Policy, expenses over $1,000 USD (or $17,000 MXN) require advance written approval from your direct manager. Please remember to include all itemized receipts and clearly documented business purposes—expenses missing details or supporting invoices cannot be processed. Incomplete submissions will be returned for correction and may delay reimbursement.\n\nIf you have any questions or need clarification, please review the T&E Policy on the intranet or contact me directly. Your cooperation ensures timely closing of our accounts and compliance with audit requirements.\n\nThank you for your attention to these details.\n\nRegards,\nMaria Santos\nFinance Director - Mexico\n\n--\nMaria Santos\nFinance Director, Mexico Operations\nAgave Spirits International',
    "Subject: Request for Volume Discount to Support Q4 Targets\nDate: 2026-01-14T09:41:00\nFrom: Kevin O'Brien\nParticipants: Carlos Delgado\n\nBody:\nHi Carlos,\n\nI wanted to reach out regarding our ongoing discussions with Distribuidora Romero. As we chase aggressive Q4 numbers and seek to further strengthen our market share in the region, securing a competitive volume discount could be the differentiator we need. Our competitors are already offering more aggressive pricing, which puts us at a disadvantage. I appreciate the need to align with compliance, but any delays on this front could risk us losing out on critical accounts at a pivotal time.\n\nLet’s connect soon to map out the path forward and ensure we’re positioned as the preferred partner.\n\nBest,\nKevin",
    'Subject: Completed Due Diligence Package: Impresiones Jalisco – Request for Compliance Approval\nDate: 2025-12-30T17:15:00\nFrom: Miguel Torres\nParticipants: Jennifer Walsh\n\nBody:\nHi Jennifer,\n\nI’m pleased to submit the complete third-party due diligence package for Impresiones Jalisco, our proposed supplier for label printing services. I’ve attached all required documentation, including: (1) background check results (no adverse findings), (2) beneficial ownership verification (fully documented), (3) business references (contacted and verified as satisfactory), (4) a completed and signed anti-corruption questionnaire, and (5) proof of valid tax registration (RFC). Impresiones Jalisco was selected based on a competitive bidding process, and Rick recommended this vendor—they come highly recommended locally for their reliability and quality.\n\nPlease review the attached package and let me know if you need any further information. Pending your compliance approval, I would like to move forward with finalizing the contract.\n\nThanks for your attention to this. I look forward to your feedback so we can proceed.\n\nBest regards,\nMiguel\n\n--\nMiguel Torres\nProcurement Manager\nASI Mexico',
]
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([[ 0.4629, -0.0649,  0.0549]], dtype=torch.bfloat16)

Evaluation

Metrics

Information Retrieval

• Dataset: val_full_corpus
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.3977
cosine_accuracy@3	0.6165
cosine_accuracy@5	0.7102
cosine_accuracy@10	0.8352
cosine_precision@1	0.3977
cosine_precision@3	0.2055
cosine_precision@5	0.142
cosine_precision@10	0.0835
cosine_recall@1	0.3977
cosine_recall@3	0.6165
cosine_recall@5	0.7102
cosine_recall@10	0.8352
cosine_ndcg@10	0.599
cosine_mrr@10	0.5251
cosine_map@100	0.5334

Training Details

Training Dataset

Unnamed Dataset

• Size: 1,520 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: 7 tokens mean: 13.18 tokens max: 27 tokens	min: 129 tokens mean: 216.77 tokens max: 511 tokens

• Samples:

  sentence_0	sentence_1
  How did compliance procedures impact market research and account strategy discussions?	Subject: Re: Q4 Market Research Results: Strong Positioning & Opportunities Date: 2025-12-01T17:56:00 From: Thomas Tom Bradford Participants: Kevin O'Brien Body: Hi Kevin, Thank you for sharing the Q4 results and highlighting these strong gains in market share. It’s clear your team’s efforts are making a tangible impact. I agree that finding the right balance between robust compliance and commercial agility is crucial, especially with larger accounts on the horizon. Let’s schedule a session with Legal and Sales to map out where we can streamline processes, while still maintaining key controls. Patricia, could you coordinate everyone’s schedules for next week? Looking forward to optimizing our approach and building on this momentum into 2024. Best, Tom
  Communications referencing urgency or market competition potentially impacting procedural compliance	Subject: Request for Special Pricing Exception – Key Account, Q4 Impact Date: 2025-11-17T18:11:00 From: Kevin O'Brien Participants: Thomas Tom Bradford Body: Hi Tom, I'm reaching out to request approval on a special pricing exception for Rivera Distributors, one of our top targets in the Southwest region. This is a time-sensitive opportunity that could substantially boost our Q4 numbers and help us solidify our competitive positioning against Cuervo and Patron. I understand compliance needs to review these exceptions, but delays could jeopardize closing this deal before year-end, impacting both market share and our sales team's momentum. Rivera is requesting a 7% discount off standard terms, and similar concessions are being made by competitors. If we can expedite this exception, I’m confident we’ll secure the account and drive incremental value for the business. Please let me know if you need more detail or want to discuss further. Appreciate your quick consideration given the mark...
  Who approved wire transfers above threshold without complete documentation or authorization forms?	Subject: Clarification Needed: Travel Expense Reimbursements and Payment Procedures Date: 2025-10-13T20:16:00 From: Maria Santos Participants: James Cooper Body: Hi James, I’m reviewing the recent travel expense claims and noticed several wire transfers requested for reimbursement. As per our policy, all travel reimbursements above $1,500 require a completed Expense Authorization Form and copies of original receipts. Additionally, I’ve observed that the beneficiary account for one claim does not match our records for the traveler. This payment needs additional documentation before we can proceed. Please confirm account details and resubmit any missing invoices by Friday, June 14 to avoid delays in processing. Let me know if you need the updated forms or further clarification on the reimbursement workflow. Best regards, Maria Santos

• 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	95	0.5955
2.0	190	0.5966
3.0	285	0.5990

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.3
• Transformers: 5.0.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.13.0
• 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}
}