tomaarsen/TinyBERT_L-4_H-312_v2-distilled-from-stsb-roberta-base-v2-cosine-projection-dim

SentenceTransformer based on nreimers/TinyBERT_L-4_H-312_v2

This is a sentence-transformers model finetuned from nreimers/TinyBERT_L-4_H-312_v2. It maps sentences & paragraphs to a 312-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: nreimers/TinyBERT_L-4_H-312_v2
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 312 dimensions
• Similarity Function: Cosine Similarity
• Supported Modality: Text

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': 'BertModel'})
  (1): Pooling({'embedding_dimension': 312, '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("tomaarsen/TinyBERT_L-4_H-312_v2-distilled-from-stsb-roberta-base-v2-cosine-projection-dim")
# Run inference
sentences = [
    'A black dog is drinking next to a brown and white dog that is looking at an orange ball in the lake, whilst a horse and rider passes behind.',
    'There are two people running around a track in lane three and the one wearing a blue shirt with a green thing over the eyes is just barely ahead of the guy wearing an orange shirt and sunglasses.',
    'the guy is dead',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 312]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.2242, 0.0591],
#         [0.2242, 1.0000, 0.0078],
#         [0.0591, 0.0078, 1.0000]])

Evaluation

Metrics

Semantic Similarity

• Datasets: sts-dev and sts-test
• Evaluated with EmbeddingSimilarityEvaluator

Metric	sts-dev	sts-test
pearson_cosine	0.8016	0.7529
spearman_cosine	0.8196	0.7601

Training Details

Training Dataset

Unnamed Dataset

• Size: 200,000 training samples
• Columns: sentence and label
• Approximate statistics based on the first 100 samples:

  	sentence	label
  type	string	list
  modality	text
  details	min: 8 tokens mean: 17.05 tokens max: 52 tokens	size: 768 elements

• Samples:

  sentence	label
  A person on a horse jumps over a broken down airplane.	[-0.477225124835968, -0.027898235246539116, 0.6169318556785583, -1.6224359273910522, 0.7474681735038757, ...]
  Children smiling and waving at camera	[-0.1697935163974762, 0.9077808856964111, -0.8368250727653503, -0.47047966718673706, -0.5604732036590576, ...]
  A boy is jumping on skateboard in the middle of a red bridge.	[0.6267533898353577, 0.011438215151429176, 0.47103747725486755, 0.4887479841709137, -0.3095979690551758, ...]

• Loss: EmbedDistillLoss with these parameters:

  {
      "distance_metric": "cosine",
      "projection_dim": 768
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 10,000 evaluation samples
• Columns: sentence and label
• Approximate statistics based on the first 100 samples:

  	sentence	label
  type	string	list
  modality	text
  details	min: 7 tokens mean: 17.97 tokens max: 45 tokens	size: 768 elements

• Samples:

  sentence	label
  Two women are embracing while holding to go packages.	[1.398005485534668, 0.6596574783325195, -0.6711955070495605, -0.35688260197639465, 0.08937469869852066, ...]
  Two young children in blue jerseys, one with the number 9 and one with the number 2 are standing on wooden steps in a bathroom and washing their hands in a sink.	[0.08953701704740524, -0.16486810147762299, -0.5275247097015381, -0.13387243449687958, 0.3173069953918457, ...]
  A man selling donuts to a customer during a world exhibition event held in the city of Angeles	[-0.18134371936321259, -0.2724460959434509, 0.605330228805542, 0.48794686794281006, -0.472873330116272, ...]

• Loss: EmbedDistillLoss with these parameters:

  {
      "distance_metric": "cosine",
      "projection_dim": 768
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 64
• num_train_epochs: 1
• learning_rate: 0.0001
• warmup_steps: 0.1
• fp16: True
• per_device_eval_batch_size: 64
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• per_device_train_batch_size: 64
• num_train_epochs: 1
• max_steps: -1
• learning_rate: 0.0001
• 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: True
• 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: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• per_device_eval_batch_size: 64
• 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: True
• 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_static_graph: None
• 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: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	sts-dev_spearman_cosine	sts-test_spearman_cosine
0.032	100	0.9106	-	-	-
0.064	200	0.7165	-	-	-
0.096	300	0.5754	-	-	-
0.128	400	0.4974	-	-	-
0.16	500	0.4469	0.5623	0.7548	-
0.192	600	0.4125	-	-	-
0.224	700	0.3872	-	-	-
0.256	800	0.3697	-	-	-
0.288	900	0.3531	-	-	-
0.32	1000	0.3358	0.4818	0.7909	-
0.352	1100	0.3267	-	-	-
0.384	1200	0.3171	-	-	-
0.416	1300	0.3102	-	-	-
0.448	1400	0.3016	-	-	-
0.48	1500	0.2960	0.4439	0.8069	-
0.512	1600	0.2878	-	-	-
0.544	1700	0.2838	-	-	-
0.576	1800	0.2785	-	-	-
0.608	1900	0.2739	-	-	-
0.64	2000	0.2704	0.4245	0.8131	-
0.672	2100	0.2687	-	-	-
0.704	2200	0.2659	-	-	-
0.736	2300	0.2622	-	-	-
0.768	2400	0.2589	-	-	-
0.8	2500	0.2577	0.4149	0.8181	-
0.832	2600	0.2552	-	-	-
0.864	2700	0.2555	-	-	-
0.896	2800	0.2524	-	-	-
0.928	2900	0.2524	-	-	-
0.96	3000	0.2516	0.4083	0.8190	-
0.992	3100	0.2504	-	-	-
1.0	3125	-	0.4077	0.8196	-
-1	-1	-	-	-	0.7601

• The bold row denotes the saved checkpoint.

Training Time

• Training: 3.3 minutes
• Evaluation: 1.0 minutes
• Total: 4.3 minutes

Framework Versions

• Python: 3.11.6
• Sentence Transformers: 5.5.0.dev0
• Transformers: 5.6.2
• PyTorch: 2.10.0+cu128
• Accelerate: 1.13.0.dev0
• Datasets: 4.8.4
• 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",
}

EmbedDistillLoss

@article{kim2023embeddistill,
    title={EmbedDistill: A Geometric Knowledge Distillation for Information Retrieval},
    author={Kim, Seungyeon and Rawat, Ankit Singh and Zaheer, Manzil and Jayasumana, Sadeep and Sadhanala, Veeranjaneyulu and Jitkrittum, Wittawat and Menon, Aditya Krishna and Fergus, Rob and Kumar, Sanjiv},
    year={2023},
    eprint={2301.12005},
    archivePrefix={arXiv},
    primaryClass={cs.IR}
}