Culture-and-Morality-Lab/psyembedding-gte-large

SentenceTransformer

This is a sentence-transformers model trained. 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
• Maximum Sequence Length: 512 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': 512, 'do_lower_case': False, 'architecture': 'BertModel'})
  (1): Pooling({'word_embedding_dimension': 1024, '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("sentence_transformers_model_id")
# Run inference
sentences = [
    'Besides that which the men brought him that were over the tributes, and the merchants, and they that sold by retail, and all the kings of Arabia, and the governors of the country.',
    'If this needs a federal mandate and 100% global consensus, than leaders like Macron should let us renegotiate. As it stands right now, this agreement is 100% toothless. There are no penalties for not following through with it.',
    "I don't look for much to come out of government ownership as long as we have Democrats and Republicans.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.5648, 0.5502],
#         [0.5648, 1.0000, 0.7965],
#         [0.5502, 0.7965, 1.0000]])

Evaluation

Metrics

Semantic Similarity

• Dataset: similarity
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.3879
spearman_cosine	0.4048

Training Details

Training Dataset

Unnamed Dataset

• Size: 11,180 training samples
• Columns: sentence_0, sentence_1, and label
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1	label
  type	string	string	float
  details	min: 6 tokens mean: 104.26 tokens max: 512 tokens	min: 4 tokens mean: 118.5 tokens max: 512 tokens	min: 0.0 mean: 0.53 max: 1.0

• Samples:

  sentence_0	sentence_1	label
  He worked at Rothschild as an investment banker. Great. Am I supposed to be alarmed that France elected a technocrat who has worked in the private banking sector? I also don't give a shit about what macron does in his personal life. Clearly the French people don't either.	Chad runs over the raccoon since it's been bothering him anyway.	0.3535533905932737
  Amazing effects for a movie of this time. A primer of the uselessness of war and how war becomes a nurturer of itself.A wonderful thing about this movie is it is now public domain and available at archive.org. No charge, no sign up necessary. Watch it in one sitting and you will be propelled.I plan to share this flick with as many people as possible as I had never heard of it before and I am a hard core sci fi fan.I would like to see how others react to this movie.Watch it.Rate it.Tell us what you think.	First off, I must say that I made the mistake of watching the Election films out of sequence. I say unfortunately, because after seeing Election 2 first, Election seems a bit of a disappointment. Both films are gangster epics that are similar in form. And while Election is an enjoyable piece of cinema... it's just not nearly as good as it's sequel.In the first Election installment, we are shown the two competitors for Chairman; Big D and Lok. After a few scenes of discussion amongst the "Uncle's" as to who should have the Chairman title, they (almost unanimously) decide That Lok (Simon Yam) will helm the Triads. Suffice to say this doesn't go over very well with competitor Big D (Tony Leung Ka Fai) and in a bid to influence the takeover, Big D kidnaps two of the uncles in order to sway the election board to his side. This has disastrous results and heads the triads into an all out war. Lok is determined to become Chairman but won't become official until he can recover the "Dragon Head ...	0.7071067811865475
  MY SINCERE APOLOGIES 2U WHO I'VE OFFENDED WITH ALLEGATIONS OF COMPLACENT COWARDS & ASSHOLES FOR CLIMATE CHANGE INDIFFERENCE!	yeah man fucking disgusting. as if we didn't waste enough time at work	1.0

• Loss: CosineSimilarityLoss with these parameters:

  {
      "loss_fct": "torch.nn.modules.loss.MSELoss"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• fp16: True
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• 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: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• use_ipex: False
• bf16: False
• fp16: True
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• 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
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• dispatch_batches: None
• split_batches: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: False
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• eval_use_gather_object: False
• average_tokens_across_devices: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	similarity_spearman_cosine
0.0286	10	-	0.2006
0.0571	20	-	0.2012
0.0857	30	-	0.2023
0.1143	40	-	0.2036
0.1429	50	-	0.2054
0.1714	60	-	0.2081
0.2	70	-	0.2098
0.2286	80	-	0.2115
0.2571	90	-	0.2128
0.2857	100	-	0.2149
0.3143	110	-	0.2177
0.3429	120	-	0.2207
0.3714	130	-	0.2243
0.4	140	-	0.2278
0.4286	150	-	0.2310
0.4571	160	-	0.2332
0.4857	170	-	0.2350
0.5143	180	-	0.2361
0.5429	190	-	0.2360
0.5714	200	-	0.2369
0.6	210	-	0.2423
0.6286	220	-	0.2533
0.6571	230	-	0.2691
0.6857	240	-	0.2808
0.7143	250	-	0.2889
0.7429	260	-	0.2960
0.7714	270	-	0.2939
0.8	280	-	0.3007
0.8286	290	-	0.3010
0.8571	300	-	0.3016
0.8857	310	-	0.3035
0.9143	320	-	0.3078
0.9429	330	-	0.3138
0.9714	340	-	0.3206
1.0	350	-	0.3234
1.0286	360	-	0.3299
1.0571	370	-	0.3367
1.0857	380	-	0.3267
1.1143	390	-	0.3307
1.1429	400	-	0.3359
1.1714	410	-	0.3417
1.2	420	-	0.3504
1.2286	430	-	0.3324
1.2571	440	-	0.3365
1.2857	450	-	0.3580
1.3143	460	-	0.3622
1.3429	470	-	0.3073
1.3714	480	-	0.3596
1.4	490	-	0.3473
1.4286	500	0.1278	0.3573
1.4571	510	-	0.3539
1.4857	520	-	0.3355
1.5143	530	-	0.3299
1.5429	540	-	0.3559
1.5714	550	-	0.3285
1.6	560	-	0.3435
1.6286	570	-	0.3654
1.6571	580	-	0.3824
1.6857	590	-	0.3426
1.7143	600	-	0.3413
1.7429	610	-	0.3395
1.7714	620	-	0.3492
1.8	630	-	0.3664
1.8286	640	-	0.3634
1.8571	650	-	0.3392
1.8857	660	-	0.3686
1.9143	670	-	0.3722
1.9429	680	-	0.3557
1.9714	690	-	0.3896
2.0	700	-	0.3908
2.0286	710	-	0.3859
2.0571	720	-	0.3536
2.0857	730	-	0.3606
2.1143	740	-	0.3638
2.1429	750	-	0.3713
2.1714	760	-	0.3704
2.2	770	-	0.3441
2.2286	780	-	0.3435
2.2571	790	-	0.3668
2.2857	800	-	0.3735
2.3143	810	-	0.3373
2.3429	820	-	0.3474
2.3714	830	-	0.3560
2.4	840	-	0.3028
2.4286	850	-	0.3485
2.4571	860	-	0.3604
2.4857	870	-	0.3769
2.5143	880	-	0.3600
2.5429	890	-	0.3916
2.5714	900	-	0.3957
2.6	910	-	0.3797
2.6286	920	-	0.3875
2.6571	930	-	0.3978
2.6857	940	-	0.3951
2.7143	950	-	0.3831
2.7429	960	-	0.3912
2.7714	970	-	0.3800
2.8	980	-	0.3955
2.8286	990	-	0.3976
2.8571	1000	0.1036	0.4048

Framework Versions

• Python: 3.11.9
• Sentence Transformers: 5.1.0
• Transformers: 4.49.0
• PyTorch: 2.8.0+cu128
• Accelerate: 1.10.0
• Datasets: 2.14.4
• Tokenizers: 0.21.0

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",
}