kwondw/distilbert-base-uncased-msmarco-margin-mse

distilbert-base-uncased finetuned on MSMARCO with MarginMSELoss (BERT-CAT ensemble teacher)

This is a sentence-transformers model finetuned from distilbert/distilbert-base-uncased on the msmarco dataset. It maps sentences & paragraphs to a 768-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: distilbert/distilbert-base-uncased
• Maximum Sequence Length: 300 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Dot Product
• Supported Modality: Text
• Training Dataset:
  • msmarco
• 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({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'forward', 'method_output_name': 'last_hidden_state'}}, 'module_output_name': 'token_embeddings', 'architecture': 'DistilBertModel'})
  (1): Pooling({'embedding_dimension': 768, 'pooling_mode': 'mean', '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("kwondw/distilbert-base-uncased-msmarco-margin-mse")
# Run inference
queries = [
    'average pay of it analyst in austin texas',
]
documents = [
    'How much does an IT Analyst in Austin, TX make? The average IT Analyst salary in Austin, TX is approximately $26.79 per hour, which is 26% below the national average. Salary information comes from 76 data points collected directly from employees, users, and past and present job advertisements on Indeed in the past 24 months.',
    'The best city between Austin, TX and Santa Fe, NM to meet is Lubbock, Texas which is about 34 miles from the exact midpoint. The town that marks the exact halfway point is actually Close City, Texas.',
    'Food and cosmetic use. Arrowroot powder is a natural powdered root starch that is white in color and very lightweight. It is derived from the tropical South American plant Maranta arundinacea. The plant was given the name â\x80\x9cArrowrootâ\x80\x9d because it was once used to treat those injured with wounds from poisoned arrows.',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[46.6321, 42.2351, 35.4387]])

Evaluation

Metrics

Information Retrieval

• Dataset: NanoMSMARCO
• Evaluated with InformationRetrievalEvaluator

Metric	Value
dot_accuracy@1	0.02
dot_accuracy@3	0.04
dot_accuracy@5	0.08
dot_accuracy@10	0.14
dot_precision@1	0.02
dot_precision@3	0.0133
dot_precision@5	0.016
dot_precision@10	0.014
dot_recall@1	0.02
dot_recall@3	0.04
dot_recall@5	0.08
dot_recall@10	0.14
dot_ndcg@10	0.0683
dot_mrr@10	0.0469
dot_map@100	0.0689

Training Details

Training Dataset

msmarco

• Dataset: msmarco
• Size: 100 training samples
• Columns: score, query, positive, and negative
• Approximate statistics based on the first 100 samples:

  	score	query	positive	negative
  type	float	string	string	string
  details	min: 0.52 mean: 14.12 max: 22.35	min: 5 tokens mean: 9.28 tokens max: 20 tokens	min: 27 tokens mean: 86.96 tokens max: 282 tokens	min: 23 tokens mean: 71.84 tokens max: 146 tokens

• Samples:

  score	query	positive	negative
  18.26768692334493	weisskopf syndrome	Dinno-Shearer-Weisskopf syndrome: Introduction. Dinno-Shearer-Weisskopf syndrome: A very rare syndrome characterized mainly by long limbs, tall stature, large head, ataxia and facial anomalies. More detailed information about the symptoms, causes, and treatments of Dinno-Shearer-Weisskopf syndrome is available below.	Related posts: 1 Charles Bonnet Syndrome - Treatment, Symptoms and Causes. 2 Refeeding Syndrome - Symptoms, Causes, Diagnosis, Treatment Guidelines. 3 Dubowitz Syndrome.
  21.609440167744957	weather in beirut, lebanon	Beirut: Annual Weather Averages. August is the hottest month in Beirut with an average temperature of 28°C (82°F) and the coldest is February at 13°C (55°F) with the most daily sunshine hours at 14 in July. The wettest month is January with an average of 130mm of rain.	When Brant and his Mohawk warriors joined the fortunes. of the Johnson family in Canada, the influence of President. Whelock, of Lebanon Institute, where Brant had received his. early education, was brought to bear upon the crisis.
  15.881475607554119	what are the clots in my period	Blood Clots during Period. It is normal to pass blood clots during menses, as along with the blood the thickened lining of the uterus is also being shed and expelled outside the body. In case the abnormal blood clots during period occur, visit your gynecologist as it may then be an indication of other serious medical illnesses.	Heparin lock flush is used for: Preventing blood clots from forming in intravenous (IV) lines or catheters. Heparin lock flush is an anticoagulant. It works by preventing blood clots from forming in intravenous lines when they are not in use. Do NOT use heparin lock flush if:

• Loss: MarginMSELoss with these parameters:

  {
      "similarity_fct": "pairwise_dot_score"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• num_train_epochs: 1
• warmup_steps: 0.1
• fp16: True

All Hyperparameters

Click to expand

• do_predict: False
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• 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: 1.0
• num_train_epochs: 1
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_ratio: None
• warmup_steps: 0.1
• 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: True
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: -1
• ddp_backend: None
• debug: []
• dataloader_drop_last: True
• 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: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	NanoMSMARCO_dot_ndcg@10
-1	-1	-	0.0683
1.0	1	186.3345	0.0683
-1	-1	-	0.0683

Training Time

• Training: 8.7 seconds
• Evaluation: 3.5 seconds
• Total: 12.2 seconds

Framework Versions

• Python: 3.12.13
• Sentence Transformers: 5.4.1
• Transformers: 5.0.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.13.0
• Datasets: 5.0.0
• 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",
}

MarginMSELoss

@misc{hofstätter2021improving,
    title={Improving Efficient Neural Ranking Models with Cross-Architecture Knowledge Distillation},
    author={Sebastian Hofstätter and Sophia Althammer and Michael Schröder and Mete Sertkan and Allan Hanbury},
    year={2021},
    eprint={2010.02666},
    archivePrefix={arXiv},
    primaryClass={cs.IR}
}