plaguss/bge-base-argilla-sdk-matryoshka

BGE base ArgillaSDK Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-base-en-v1.5. It maps sentences & paragraphs to a 768-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: BAAI/bge-base-en-v1.5
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 tokens
• Similarity Function: Cosine Similarity
• 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': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, '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("plaguss/bge-base-argilla-sdk-matryoshka")
# Run inference
sentences = [
    'hide: footer\n\nrg.Argilla\n\nTo interact with the Argilla server from python you can use the Argilla class. The Argilla client is used to create, get, update, and delete all Argilla resources, such as workspaces, users, datasets, and records.\n\nUsage Examples\n\nConnecting to an Argilla server\n\nTo connect to an Argilla server, instantiate the Argilla class and pass the api_url of the server and the api_key to authenticate.\n\n```python\nimport argilla_sdk as rg',
    'Can the Argilla class be employed to streamline dataset administration tasks in my Argilla server setup?',
    'The Argilla flowers were blooming beautifully in the garden.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Information Retrieval

• Dataset: dim_768
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.1327
cosine_accuracy@3	0.2857
cosine_accuracy@5	0.3878
cosine_accuracy@10	0.5204
cosine_precision@1	0.1327
cosine_precision@3	0.0952
cosine_precision@5	0.0776
cosine_precision@10	0.052
cosine_recall@1	0.1327
cosine_recall@3	0.2857
cosine_recall@5	0.3878
cosine_recall@10	0.5204
cosine_ndcg@10	0.3086
cosine_mrr@10	0.2432
cosine_map@100	0.2604

Information Retrieval

• Dataset: dim_512
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.102
cosine_accuracy@3	0.2755
cosine_accuracy@5	0.3878
cosine_accuracy@10	0.5102
cosine_precision@1	0.102
cosine_precision@3	0.0918
cosine_precision@5	0.0776
cosine_precision@10	0.051
cosine_recall@1	0.102
cosine_recall@3	0.2755
cosine_recall@5	0.3878
cosine_recall@10	0.5102
cosine_ndcg@10	0.2942
cosine_mrr@10	0.2264
cosine_map@100	0.2426

Information Retrieval

• Dataset: dim_256
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.1224
cosine_accuracy@3	0.2755
cosine_accuracy@5	0.3878
cosine_accuracy@10	0.5
cosine_precision@1	0.1224
cosine_precision@3	0.0918
cosine_precision@5	0.0776
cosine_precision@10	0.05
cosine_recall@1	0.1224
cosine_recall@3	0.2755
cosine_recall@5	0.3878
cosine_recall@10	0.5
cosine_ndcg@10	0.2931
cosine_mrr@10	0.2291
cosine_map@100	0.2445

Information Retrieval

• Dataset: dim_128
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0918
cosine_accuracy@3	0.2551
cosine_accuracy@5	0.3163
cosine_accuracy@10	0.4694
cosine_precision@1	0.0918
cosine_precision@3	0.085
cosine_precision@5	0.0633
cosine_precision@10	0.0469
cosine_recall@1	0.0918
cosine_recall@3	0.2551
cosine_recall@5	0.3163
cosine_recall@10	0.4694
cosine_ndcg@10	0.2629
cosine_mrr@10	0.1992
cosine_map@100	0.2165

Information Retrieval

• Dataset: dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0816
cosine_accuracy@3	0.2551
cosine_accuracy@5	0.3163
cosine_accuracy@10	0.4796
cosine_precision@1	0.0816
cosine_precision@3	0.085
cosine_precision@5	0.0633
cosine_precision@10	0.048
cosine_recall@1	0.0816
cosine_recall@3	0.2551
cosine_recall@5	0.3163
cosine_recall@10	0.4796
cosine_ndcg@10	0.2611
cosine_mrr@10	0.194
cosine_map@100	0.2059

Training Details

Training Dataset

Unnamed Dataset

• Size: 882 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 1000 samples:

  	anchor	positive	negative
  type	string	string	string
  details	min: 6 tokens mean: 90.85 tokens max: 198 tokens	min: 8 tokens mean: 25.44 tokens max: 91 tokens	min: 10 tokens mean: 22.33 tokens max: 61 tokens

• Samples:

  anchor	positive	negative
  `` !!! note "Update the metadata" ThemetadataofRecordobject is a python dictionary. So to update the metadata of a record, you can iterate over the records and update the metadata by key or usingmetadata.update`. After that, you should update the records in the dataset.	Can I use Argilla to annotate the metadata of Record objects and update them in the dataset?	The beautiful scenery of the Argilla valley in Italy is perfect for a relaxing summer vacation.
  git checkout [branch-name] git rebase [default-branch] <br><br>If everything is right, we need to commit and push the changes to your fork. For that, run the following commands:<br><br>sh Add the changes to the staging area git add filename Commit the changes by writing a proper message git commit -m "commit-message" Push the changes to your fork	Can I commit Argilla's annotation changes and push them to a forked project repository after rebasing from the default branch?	The beautiful beach in Argilla, Spain, is a popular spot for surfers to catch a wave and enjoy the sunny weather.
  Accessing Record Attributes The Record object has suggestions, responses, metadata, and vectors attributes that can be accessed directly whilst iterating over records in a dataset. python for record in dataset.records( with_suggestions=True, with_responses=True, with_metadata=True, with_vectors=True ): print(record.suggestions) print(record.responses) print(record.metadata) print(record.vectors)	Is it possible to retrieve the suggestions, responses, metadata, and vectors of a Record object at the same time when iterating over a dataset in Argilla?	The new hiking trail offered breathtaking suggestions for scenic views, responses to environmental concerns, and metadata about the surrounding ecosystem, but it lacked vectors for navigation.

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "TripletLoss",
      "matryoshka_dims": [
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_eval_batch_size: 4
• gradient_accumulation_steps: 4
• learning_rate: 2e-05
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 8
• per_device_eval_batch_size: 4
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 4
• eval_accumulation_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: 3
• 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: 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: 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}
• 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: False
• hub_always_push: False
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• 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
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_128_cosine_map@100	dim_256_cosine_map@100	dim_512_cosine_map@100	dim_64_cosine_map@100	dim_768_cosine_map@100
0.1802	5	21.701	-	-	-	-	-
0.3604	10	21.7449	-	-	-	-	-
0.5405	15	21.7453	-	-	-	-	-
0.7207	20	21.7168	-	-	-	-	-
0.9009	25	21.6945	-	-	-	-	-
0.973	27	-	0.2165	0.2445	0.2426	0.2059	0.2604
1.0811	30	21.7248	-	-	-	-	-
1.2613	35	21.7322	-	-	-	-	-
1.4414	40	21.7367	-	-	-	-	-
1.6216	45	21.6821	-	-	-	-	-
1.8018	50	21.8392	-	-	-	-	-
1.9820	55	21.6441	0.2165	0.2445	0.2426	0.2059	0.2604
2.1622	60	21.8154	-	-	-	-	-
2.3423	65	21.7098	-	-	-	-	-
2.5225	70	21.6447	-	-	-	-	-
2.7027	75	21.6033	-	-	-	-	-
2.8829	80	21.8271	-	-	-	-	-
2.9189	81	-	0.2165	0.2445	0.2426	0.2059	0.2604

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.8
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.1.2
• Accelerate: 0.31.0
• Datasets: 2.19.2
• Tokenizers: 0.19.1

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",
}

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}
}

TripletLoss

@misc{hermans2017defense,
    title={In Defense of the Triplet Loss for Person Re-Identification}, 
    author={Alexander Hermans and Lucas Beyer and Bastian Leibe},
    year={2017},
    eprint={1703.07737},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}