tomaarsen/all-distilroberta-v1-trec-balanced-sampler

SentenceTransformer based on sentence-transformers/all-distilroberta-v1

This is a sentence-transformers model finetuned from sentence-transformers/all-distilroberta-v1 on the trec 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: sentence-transformers/all-distilroberta-v1
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • trec

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, 'architecture': 'RobertaModel'})
  (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})
  (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("tomaarsen/all-distilroberta-v1-trec-balanced-sampler")
# Run inference
sentences = [
    "What country contains Africa 's northernmost point ?",
    'What is the difference between classical conditioning and operant conditioning ?',
    'Where can stocks be traded on-line ?',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[ 1.0000, -0.6694,  0.9470],
#         [-0.6694,  1.0000, -0.6573],
#         [ 0.9470, -0.6573,  1.0000]])

Evaluation

Metrics

Triplet

• Datasets: trec-dev and trec-test
• Evaluated with TripletEvaluator

Metric	trec-dev	trec-test
cosine_accuracy	0.9451	0.9434

Training Details

Training Dataset

trec

• Dataset: trec at a073d2e
• Size: 4,952 training samples
• Columns: text and label
• Approximate statistics based on the first 1000 samples:

  	text	label
  type	string	int
  details	min: 6 tokens mean: 13.29 tokens max: 39 tokens	0: ~0.30% 1: ~1.50% 2: ~3.10% 3: ~0.60% 4: ~0.80% 5: ~3.90% 7: ~1.80% 8: ~1.20% 9: ~1.70% 10: ~0.20% 11: ~0.40% 12: ~0.20% 13: ~4.30% 14: ~0.40% 15: ~0.90% 16: ~0.10% 17: ~1.20% 18: ~0.70% 19: ~0.10% 20: ~0.70% 21: ~1.40% 22: ~0.20% 23: ~0.50% 24: ~7.90% 25: ~4.40% 26: ~4.90% 27: ~3.90% 28: ~3.30% 29: ~17.50% 30: ~0.50% 31: ~0.70% 32: ~2.70% 33: ~2.80% 34: ~0.80% 35: ~7.90% 36: ~1.40% 37: ~0.10% 38: ~6.30% 39: ~4.80% 40: ~0.40% 41: ~1.10% 42: ~0.10% 43: ~0.70% 44: ~0.40% 45: ~0.40% 46: ~0.50% 47: ~0.10% 48: ~0.20%

• Samples:

  text	label
  How did serfdom develop in and then leave Russia ?	26
  What films featured the character Popeye Doyle ?	5
  How can I find a list of celebrities ' real names ?	26

• Loss: BatchAllTripletLoss

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 32
• num_train_epochs: 1
• warmup_steps: 0.1
• eval_strategy: steps
• batch_sampler: group_by_label

All Hyperparameters

Click to expand

• per_device_train_batch_size: 32
• num_train_epochs: 1
• max_steps: -1
• learning_rate: 5e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0.1
• optim: adamw_torch_fused
• optim_args: None
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• optim_target_modules: None
• gradient_accumulation_steps: 1
• average_tokens_across_devices: True
• max_grad_norm: 1.0
• label_smoothing_factor: 0.0
• bf16: False
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• use_liger_kernel: False
• liger_kernel_config: None
• use_cache: False
• neftune_noise_alpha: None
• torch_empty_cache_steps: None
• auto_find_batch_size: False
• log_on_each_node: True
• logging_nan_inf_filter: True
• include_num_input_tokens_seen: no
• log_level: passive
• log_level_replica: warning
• disable_tqdm: False
• project: huggingface
• trackio_space_id: trackio
• eval_strategy: steps
• per_device_eval_batch_size: 8
• prediction_loss_only: True
• eval_on_start: False
• eval_do_concat_batches: True
• eval_use_gather_object: False
• eval_accumulation_steps: None
• include_for_metrics: []
• batch_eval_metrics: False
• save_only_model: False
• save_on_each_node: False
• enable_jit_checkpoint: False
• push_to_hub: False
• hub_private_repo: None
• hub_model_id: None
• hub_strategy: every_save
• hub_always_push: False
• hub_revision: None
• load_best_model_at_end: False
• ignore_data_skip: False
• restore_callback_states_from_checkpoint: False
• full_determinism: False
• seed: 42
• data_seed: None
• use_cpu: 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
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• dataloader_prefetch_factor: None
• remove_unused_columns: True
• label_names: None
• train_sampling_strategy: random
• length_column_name: length
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• ddp_backend: None
• ddp_timeout: 1800
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• deepspeed: None
• debug: []
• skip_memory_metrics: True
• do_predict: False
• resume_from_checkpoint: None
• warmup_ratio: None
• local_rank: -1
• prompts: None
• batch_sampler: group_by_label
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	trec-dev_cosine_accuracy	trec-test_cosine_accuracy
-1	-1	-	0.7683	-
0.1066	13	4.8492	-	-
0.2049	25	-	0.9309	-
0.2131	26	4.5356	-	-
0.3197	39	4.2800	-	-
0.4098	50	-	0.9289	-
0.4262	52	4.2069	-	-
0.5328	65	4.1544	-	-
0.6148	75	-	0.9390	-
0.6393	78	4.1101	-	-
0.7459	91	4.0749	-	-
0.8197	100	-	0.9451	-
0.8525	104	4.0868	-	-
0.9590	117	4.0453	-	-
-1	-1	-	-	0.9434

Framework Versions

• Python: 3.11.6
• Sentence Transformers: 5.3.0.dev0
• Transformers: 5.3.0.dev0
• PyTorch: 2.10.0+cu126
• Accelerate: 1.12.0
• Datasets: 4.3.0
• 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",
}

BatchAllTripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification},
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}