swardiantara/bert-tiny-amazon_reviews-k10-fixed-cosine

SentenceTransformer based on google/bert_uncased_L-2_H-128_A-2

This is a sentence-transformers model finetuned from google/bert_uncased_L-2_H-128_A-2. It maps sentences & paragraphs to a 128-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: google/bert_uncased_L-2_H-128_A-2
• Maximum Sequence Length: 128 tokens
• Output Dimensionality: 128 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': 128, '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("swardiantara/bert-tiny-amazon_reviews-k10-fixed-cosine")
# Run inference
sentences = [
    "Arrived broken. Manufacturer defect. Two of the legs of the base were not completely formed, so there was no way to insert the casters. I unpackaged the entire chair and hardware before noticing this. So, I'll spend twice the amount of time boxing up the whole useless thing and send it back with a 1-star review of part of a chair I never got to sit in. I will go so far as to include a picture of what their injection molding and quality assurance process missed though. I will be hesitant to buy again. It makes me wonder if there aren't missing structures and supports that don't impede the assembly process.",
    'I like the design of the bag, though it came a little crumpled up. Hoping everything smoothes out with use. Also very upset that I paid more for this bag and my key chain was never sent.',
    'I bought these plates because I liked the way they looked. They are good enough for a single use but do not hold up too well with multiple uses. Once they are put in the microwave they start to crack quite easily.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 128]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.2434, 0.1683],
#         [0.2434, 1.0000, 0.4315],
#         [0.1683, 0.4315, 1.0000]])

Training Details

Training Dataset

Unnamed Dataset

• Size: 8,201,450 training samples
• Columns: text_a, text_b, and label
• Approximate statistics based on the first 100 samples:

  	text_a	text_b	label
  type	string	string	list
  modality	text	text
  details	min: 15 tokens mean: 66.74 tokens max: 128 tokens	min: 8 tokens mean: 46.15 tokens max: 128 tokens	size: 2 elements

• Samples:

  text_a	text_b	label
  Arrived broken. Manufacturer defect. Two of the legs of the base were not completely formed, so there was no way to insert the casters. I unpackaged the entire chair and hardware before noticing this. So, I'll spend twice the amount of time boxing up the whole useless thing and send it back with a 1-star review of part of a chair I never got to sit in. I will go so far as to include a picture of what their injection molding and quality assurance process missed though. I will be hesitant to buy again. It makes me wonder if there aren't missing structures and supports that don't impede the assembly process.	I ordered this a while ago, even using prime it took over a month to get here, now that I finally got it, I only receved 6 of the 12! Im a little upset. It's still funny, but would not recomend for someone who needs foam quick and in bulk like i do.	[1.0, 0.0]
  Arrived broken. Manufacturer defect. Two of the legs of the base were not completely formed, so there was no way to insert the casters. I unpackaged the entire chair and hardware before noticing this. So, I'll spend twice the amount of time boxing up the whole useless thing and send it back with a 1-star review of part of a chair I never got to sit in. I will go so far as to include a picture of what their injection molding and quality assurance process missed though. I will be hesitant to buy again. It makes me wonder if there aren't missing structures and supports that don't impede the assembly process.	I was happy to find this piece to replace on my broken Shark. Price was good and had it in a few days. When I installed it however, I was no longer satisfied. It wouldn't stay attached and kept popping of with any pull on the handle. Returned the same day it areived.	[0.0, 0.25]
  Arrived broken. Manufacturer defect. Two of the legs of the base were not completely formed, so there was no way to insert the casters. I unpackaged the entire chair and hardware before noticing this. So, I'll spend twice the amount of time boxing up the whole useless thing and send it back with a 1-star review of part of a chair I never got to sit in. I will go so far as to include a picture of what their injection molding and quality assurance process missed though. I will be hesitant to buy again. It makes me wonder if there aren't missing structures and supports that don't impede the assembly process.	Had to return. This unit is tiny and very cheap quality.	[0.0, 0.25]

• Loss: main.OrdinalProxyContrastiveLoss

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 1024
• num_train_epochs: 10
• learning_rate: 2e-05
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• per_device_train_batch_size: 1024
• num_train_epochs: 10
• max_steps: -1
• learning_rate: 2e-05
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_steps: 0
• optim: adamw_torch
• 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: None
• trackio_bucket_id: None
• trackio_static_space_id: None
• 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: 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: None
• fsdp_config: None
• 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

Click to expand

Epoch	Step	Training Loss
0.0624	500	0.0096
0.1248	1000	0.0055
0.1873	1500	0.0053
0.2497	2000	0.0052
0.3121	2500	0.0051
0.3745	3000	0.0051
0.4370	3500	0.0050
0.4994	4000	0.0050
0.5618	4500	0.0050
0.6242	5000	0.0049
0.6866	5500	0.0049
0.7491	6000	0.0049
0.8115	6500	0.0048
0.8739	7000	0.0049
0.9363	7500	0.0048
0.9988	8000	0.0048
1.0	8010	-
1.0612	8500	0.0047
1.1236	9000	0.0047
1.1860	9500	0.0047
1.2484	10000	0.0047
1.3109	10500	0.0046
1.3733	11000	0.0047
1.4357	11500	0.0047
1.4981	12000	0.0047
1.5605	12500	0.0046
1.6230	13000	0.0046
1.6854	13500	0.0047
1.7478	14000	0.0046
1.8102	14500	0.0046
1.8727	15000	0.0046
1.9351	15500	0.0046
1.9975	16000	0.0046
2.0	16020	-
2.0599	16500	0.0045
2.1223	17000	0.0045
2.1848	17500	0.0045
2.2472	18000	0.0045
2.3096	18500	0.0045
2.3720	19000	0.0044
2.4345	19500	0.0045
2.4969	20000	0.0045
2.5593	20500	0.0045
2.6217	21000	0.0045
2.6841	21500	0.0045
2.7466	22000	0.0045
2.8090	22500	0.0045
2.8714	23000	0.0044
2.9338	23500	0.0045
2.9963	24000	0.0044
3.0	24030	-
3.0587	24500	0.0044
3.1211	25000	0.0044
3.1835	25500	0.0044
3.2459	26000	0.0044
3.3084	26500	0.0044
3.3708	27000	0.0043
3.4332	27500	0.0043
3.4956	28000	0.0044
3.5581	28500	0.0044
3.6205	29000	0.0044
3.6829	29500	0.0043
3.7453	30000	0.0044
3.8077	30500	0.0044
3.8702	31000	0.0044
3.9326	31500	0.0043
3.9950	32000	0.0043
4.0	32040	-
4.0574	32500	0.0043
4.1199	33000	0.0043
4.1823	33500	0.0043
4.2447	34000	0.0043
4.3071	34500	0.0043
4.3695	35000	0.0043
4.4320	35500	0.0043
4.4944	36000	0.0043
4.5568	36500	0.0042
4.6192	37000	0.0042
4.6816	37500	0.0043
4.7441	38000	0.0043
4.8065	38500	0.0043
4.8689	39000	0.0042
4.9313	39500	0.0043
4.9938	40000	0.0043
5.0	40050	-
5.0562	40500	0.0042
5.1186	41000	0.0042
5.1810	41500	0.0042
5.2434	42000	0.0042
5.3059	42500	0.0043
5.3683	43000	0.0042
5.4307	43500	0.0043
5.4931	44000	0.0042
5.5556	44500	0.0042
5.6180	45000	0.0042
5.6804	45500	0.0042
5.7428	46000	0.0042
5.8052	46500	0.0043
5.8677	47000	0.0042
5.9301	47500	0.0042
5.9925	48000	0.0042
6.0	48060	-
6.0549	48500	0.0042
6.1174	49000	0.0042
6.1798	49500	0.0042
6.2422	50000	0.0042
6.3046	50500	0.0042
6.3670	51000	0.0042
6.4295	51500	0.0042
6.4919	52000	0.0042
6.5543	52500	0.0042
6.6167	53000	0.0042
6.6792	53500	0.0041
6.7416	54000	0.0041
6.8040	54500	0.0041
6.8664	55000	0.0042
6.9288	55500	0.0042
6.9913	56000	0.0041
7.0	56070	-
7.0537	56500	0.0042
7.1161	57000	0.0042
7.1785	57500	0.0041
7.2409	58000	0.0041
7.3034	58500	0.0041
7.3658	59000	0.0041
7.4282	59500	0.0041
7.4906	60000	0.0042
7.5531	60500	0.0041
7.6155	61000	0.0042
7.6779	61500	0.0041
7.7403	62000	0.0041
7.8027	62500	0.0042
7.8652	63000	0.0041
7.9276	63500	0.0041
7.9900	64000	0.0041
8.0	64080	-
8.0524	64500	0.0041
8.1149	65000	0.0041
8.1773	65500	0.0041
8.2397	66000	0.0041
8.3021	66500	0.0041
8.3645	67000	0.0041
8.4270	67500	0.0041
8.4894	68000	0.0041
8.5518	68500	0.0041
8.6142	69000	0.0041
8.6767	69500	0.0040
8.7391	70000	0.0041
8.8015	70500	0.0042
8.8639	71000	0.0042
8.9263	71500	0.0040
8.9888	72000	0.0041
9.0	72090	-
9.0512	72500	0.0041
9.1136	73000	0.0042
9.1760	73500	0.0041
9.2385	74000	0.0040
9.3009	74500	0.0041
9.3633	75000	0.0041
9.4257	75500	0.0041
9.4881	76000	0.0041
9.5506	76500	0.0041
9.6130	77000	0.0041
9.6754	77500	0.0042
9.7378	78000	0.0041
9.8002	78500	0.0041
9.8627	79000	0.0041
9.9251	79500	0.0041
9.9875	80000	0.0041
10.0	80100	-

• The bold row denotes the saved checkpoint.

Training Time

• Training: 6.8 hours
• Evaluation: 44.3 seconds
• Total: 6.8 hours

Framework Versions

• Python: 3.12.4
• Sentence Transformers: 5.5.1
• Transformers: 5.11.0
• PyTorch: 2.5.1+cu121
• Accelerate: 1.13.0
• Datasets: 2.21.0
• 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",
}