dwulff/minilm-brl

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. It maps sentences & paragraphs to a 384-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-MiniLM-L6-v2
• Maximum Sequence Length: 256 tokens
• Output Dimensionality: 384 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': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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("dwulff/minilm-brl")
# Run inference
sentences = [
    'An article on behavioral reinforcement learning:\n\nTitle: Confidence and the description–experience distinction.\nAbstract: In this paper, we extend the literature on the description–experience gap in risky choices by focusing on how the mode of learning—through description or experience—affects confidence. Specifically, we explore how learning through description or experience affects confidence in (1) the information gathered to make a decision and (2) the resulting choice. In two preregistered experiments we tested whether there was a description–experience gap in both dimensions of confidence. Learning from description was associated with higher confidence—both in the information gathered and in the choice made—than was learning from experience. In a third preregistered experiment, we examined the effect of sample size on confidence in decisions from experience. Contrary to the normative view that larger samples foster confidence in statistical inference, we observed that more experience led to less confidence. This observation is reminiscent of recent theories of deliberate ignorance, which highlight the adaptive benefits of deliberately limiting information search.',
    "An article on behavioral reinforcement learning:\n\nTitle: How (in)variant are subjective representations of described and experienced risk and rewards?.\nAbstract: Decisions under risk have been shown to differ depending on whether information on outcomes and probabilities is gleaned from symbolic descriptions or gathered through experience. To some extent, this description–experience gap is due to sampling error in experience-based choice. Analyses with cumulative prospect theory (CPT), investigating to what extent the gap is also driven by differences in people's subjective representations of outcome and probability information (taking into account sampling error), have produced mixed results. We improve on previous analyses of description-based and experience-based choices by taking advantage of both a within-subjects design and a hierarchical Bayesian implementation of CPT. This approach allows us to capture both the differences and the within-person stability of individuals’ subjective representations across the two modes of learning about choice options. Relative to decisions from description, decisions from experience showed reduced sensitivity to probabilities and increased sensitivity to outcomes. For some CPT parameters, individual differences were relatively stable across modes of learning. Our results suggest that outcome and probability information translate into systematically different subjective representations in description- versus experience-based choice. At the same time, both types of decisions seem to tap into the same individual-level regularities.",
    "An article on behavioral reinforcement learning:\n\nTitle: Do narcissists make better decisions? An investigation of narcissism and dynamic decision-making performance.\nAbstract: We investigated whether narcissism affected dynamic decision-making performance in the presence and absence of misleading information. Performance was examined in a two-choice dynamic decision-making task where the optimal strategy was to forego an option providing larger immediate rewards in favor of an option that led to larger delayed rewards. Information regarding foregone rewards from the alternate option was presented or withheld to bias participants toward the sub-optimal choice. The results demonstrated that individuals high in narcissistic traits performed comparably to low narcissism individuals when foregone reward information was absent, but high narcissism individuals outperformed individuals low in narcissistic traits when misleading information was presented. The advantage for participants high in narcissistic traits was strongest within males, and, overall, males outperformed females when foregone rewards were present. While prior research emphasizes narcissists' decision-making deficits, our findings provide evidence that individuals high in narcissistic traits excel at decision-making tasks that involve disregarding ambiguous information and focusing on the long-term utility of each option. Their superior ability at filtering out misleading information may reflect an effort to maintain their self-view or avoid ego threat.",
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Training Details

Training Dataset

Unnamed Dataset

• Size: 50,000 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: 102 tokens mean: 237.66 tokens max: 256 tokens	min: 61 tokens mean: 227.84 tokens max: 256 tokens	min: 0.0 mean: 0.17 max: 0.9

• Samples:

  sentence_0	sentence_1	label
  An article on behavioral reinforcement learning: Title: Working memory and response selection: A computational account of interactions among cortico-basalganglio-thalamic loops. Abstract: Cortico-basalganglio-thalamic loops are involved in both cognitive processes and motor control. We present a biologically meaningful computational model of how these loops contribute to the organization of working memory and the development of response behavior. Via reinforcement learning in basal ganglia, the model develops flexible control of working memory within prefrontal loops and achieves selection of appropriate responses based on working memory content and visual stimulation within a motor loop. We show that both working memory control and response selection can evolve within parallel and interacting cortico-basalganglio-thalamic loops by Hebbian and three-factor learning rules. Furthermore, the model gives a coherent explanation for how complex strategies of working memory control and respo...	An article on behavioral reinforcement learning: Title: The role of basal ganglia in reinforcement learning and imprinting in domestic chicks. Abstract: Effects of bilateral kainate lesions of telencephalic basal ganglia (lobus parolfactorius, LPO) were examined in domestic chicks. In the imprinting paradigm, where chicks learned to selectively approach a moving object without any explicitly associated reward, both the pre- and post-training lesions were without effects. On the other hand, in the water-reinforced pecking task, pre-training lesions of LPO severely impaired immediate reinforcement as well as formation of the association memory. However, post-training LPO lesions did not cause amnesia, and chicks selectively pecked at the reinforced color. The LPO could thus be involved specifically in the evaluation of present rewards and the instantaneous reinforcement of pecking, but not in the execution of selective behavior based on a memorized color cue.	0.5
  An article on behavioral reinforcement learning: Title: Exploration Disrupts Choice-Predictive Signals and Alters Dynamics in Prefrontal Cortex. Abstract: In uncertain environments, decision-makers must balance two goals: they must “exploit” rewarding options but also “explore” in order to discover rewarding alternatives. Exploring and exploiting necessarily change how the brain responds to identical stimuli, but little is known about how these states, and transitions between them, change how the brain transforms sensory information into action. To address this question, we recorded neural activity in a prefrontal sensorimotor area while monkeys naturally switched between exploring and exploiting rewarding options. We found that exploration profoundly reduced spatially selective, choice-predictive activity in single neurons and delayed choice-predictive population dynamics. At the same time, reward learning was increased in brain and behavior. These results indicate that exploration i...	An article on behavioral reinforcement learning: Title: Counterfactual choice and learning in a Neural Network centered on human lateral frontopolar cortex. Abstract: Decision making and learning in a real-world context require organisms to track not only the choices they make and the outcomes that follow but also other untaken, or counterfactual, choices and their outcomes. Although the neural system responsible for tracking the value of choices actually taken is increasingly well understood, whether a neural system tracks counterfactual information is currently unclear. Using a three-alternative decision-making task, a Bayesian reinforcement-learning algorithm, and fMRI, we investigated the coding of counterfactual choices and prediction errors in the human brain. Rather than representing evidence favoring multiple counterfactual choices, lateral frontal polar cortex (lFPC), dorsomedial frontal cortex (DMFC), and posteromedial cortex (PMC) encode the reward-based evidence favoring t...	0.5
  An article on behavioral reinforcement learning: Title: Electrophysiological signatures of visual statistical learning in 3-month-old infants at familial and low risk for autism spectrum disorder. Abstract: Visual statistical learning (VSL) refers to the ability to extract associations and conditional probabilities within the visual environment. It may serve as a precursor to cognitive and social communication development. Quantifying VSL in infants at familial risk (FR) for Autism Spectrum Disorder (ASD) provides opportunities to understand how genetic predisposition can influence early learning processes which may, in turn, lay a foundation for cognitive and social communication delays. We examined electroencephalography (EEG) signatures of VSL in 3-month-old infants, examining whether EEG correlates of VSL differentiated FR from low-risk (LR) infants. In an exploratory analysis, we then examined whether EEG correlates of VSL at 3 months relate to cognitive function and ASD symptoms...	An article on behavioral reinforcement learning: Title: Reduced nucleus accumbens reactivity and adolescent depression following early-life stress. Abstract: Depression is a common outcome for those having experienced early-life stress (ELS). For those individuals, depression typically increases during adolescence and appears to endure into adulthood, suggesting alterations in the development of brain systems involved in depression. Developmentally, the nucleus accumbens (NAcc), a limbic structure associated with reward learning and motivation, typically undergoes dramatic functional change during adolescence; therefore, age-related changes in NAcc function may underlie increases in depression in adolescence following ELS. The current study examined the effects of ELS in 38 previously institutionalized children and adolescents in comparison to a group of 31 youths without a history of ELS. Consistent with previous research, the findings showed that depression was higher in adolescents...	0.0

• Loss: CosineSimilarityLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• num_train_epochs: 5
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• 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: 5
• 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: False
• 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}
• tp_size: 0
• 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

Training Logs

Epoch	Step	Training Loss
0.6394	500	0.0179
1.2788	1000	0.0124
1.9182	1500	0.0107
2.5575	2000	0.0092
3.1969	2500	0.0086
3.8363	3000	0.0078
4.4757	3500	0.0073

Framework Versions

• Python: 3.13.2
• Sentence Transformers: 4.0.2
• Transformers: 4.50.0.dev0
• PyTorch: 2.6.0
• Accelerate: 1.5.2
• Datasets: 3.5.0
• Tokenizers: 0.21.1

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",
}