smikulas/MNLP_M3_document_encoder

MNLP M3 Encoder SciQA

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2 on the json dataset. 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
• Training Dataset:
  • json
• Language: en
• License: apache-2.0

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("sentence_transformers_model_id")
# Run inference
sentences = [
    'There are only four possible bases that make up each dna nucleotide: adenine, guanine, thymine, and?',
    'The only difference between each nucleotide is the identity of the base. There are only four possible bases that make up each DNA nucleotide: adenine (A), guanine (G), thymine (T), and cytosine (C).',
    'Metamorphism. This long word means “to change form. “ A rock undergoes metamorphism if it is exposed to extreme heat and pressure within the crust. With metamorphism , the rock does not melt all the way. The rock changes due to heat and pressure. A metamorphic rock may have a new mineral composition and/or texture.',
]
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]

Evaluation

Metrics

Information Retrieval

• Datasets: dim_384, dim_256, dim_192, dim_128, dim_96 and dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	dim_384	dim_256	dim_192	dim_128	dim_96	dim_64
cosine_accuracy@1	0.612	0.5977	0.5891	0.5663	0.552	0.5167
cosine_accuracy@3	0.8017	0.7912	0.7788	0.7626	0.7417	0.7045
cosine_accuracy@5	0.8541	0.8398	0.8332	0.8265	0.8093	0.7684
cosine_accuracy@10	0.9276	0.9152	0.9037	0.8913	0.8732	0.837
cosine_precision@1	0.612	0.5977	0.5891	0.5663	0.552	0.5167
cosine_precision@3	0.2672	0.2637	0.2596	0.2542	0.2472	0.2348
cosine_precision@5	0.1708	0.168	0.1666	0.1653	0.1619	0.1537
cosine_precision@10	0.0928	0.0915	0.0904	0.0891	0.0873	0.0837
cosine_recall@1	0.612	0.5977	0.5891	0.5663	0.552	0.5167
cosine_recall@3	0.8017	0.7912	0.7788	0.7626	0.7417	0.7045
cosine_recall@5	0.8541	0.8398	0.8332	0.8265	0.8093	0.7684
cosine_recall@10	0.9276	0.9152	0.9037	0.8913	0.8732	0.837
cosine_ndcg@10	0.769	0.7559	0.7467	0.7276	0.712	0.6755
cosine_mrr@10	0.7185	0.705	0.6965	0.6752	0.6603	0.6239
cosine_map@100	0.721	0.7085	0.7004	0.6794	0.6649	0.6293

Training Details

Training Dataset

json

• Dataset: json
• Size: 9,432 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 7 tokens mean: 18.15 tokens max: 60 tokens	min: 10 tokens mean: 94.56 tokens max: 256 tokens

• Samples:

  anchor	positive
  What is the term for atherosclerosis of arteries that supply the heart muscle?	Atherosclerosis of arteries that supply the heart muscle is called coronary heart disease . This disease may or may not have symptoms, such as chest pain. As the disease progresses, there is an increased risk of heart attack. A heart attack occurs when the blood supply to part of the heart muscle is blocked and cardiac muscle fibers die. Coronary heart disease is the leading cause of death of adults in the United States.
  What term describes a drug that has an effect on the central nervous system?	Caffeine is an example of a psychoactive drug. It is found in coffee and many other products (see Table below ). Caffeine is a central nervous system stimulant . Like other stimulant drugs, it makes you feel more awake and alert. Other psychoactive drugs include alcohol, nicotine, and marijuana. Each has a different effect on the central nervous system. Alcohol, for example, is a depressant . It has the opposite effects of a stimulant like caffeine.
  What scale is used to succinctly communicate the acidity or basicity of a solution?	The pH scale is used to succinctly communicate the acidity or basicity of a solution.

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          384,
          256,
          192,
          128,
          96,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• gradient_accumulation_steps: 16
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• bf16: True
• tf32: True
• load_best_model_at_end: True
• optim: adamw_torch_fused
• batch_sampler: no_duplicates

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 16
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 16
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 4
• max_steps: -1
• lr_scheduler_type: cosine
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.1
• 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: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: True
• 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: True
• 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_fused
• 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
• 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: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_384_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_192_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_96_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.5424	10	22.4049	-	-	-	-	-	-
1.0	19	-	0.7424	0.7315	0.7263	0.7093	0.6919	0.6575
1.0542	20	16.6616	-	-	-	-	-	-
1.5966	30	16.8367	-	-	-	-	-	-
2.0	38	-	0.7612	0.7520	0.7431	0.7261	0.7097	0.6708
2.1085	40	12.8169	-	-	-	-	-	-
2.6508	50	13.7826	-	-	-	-	-	-
3.0	57	-	0.7675	0.7548	0.7477	0.7274	0.7125	0.6756
3.1627	60	12.4455	-	-	-	-	-	-
3.7051	70	12.2968	-	-	-	-	-	-
3.8136	72	-	0.769	0.7559	0.7467	0.7276	0.712	0.6755

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.12.8
• Sentence Transformers: 3.4.1
• Transformers: 4.51.3
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.6.0
• 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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}