fitlemon/bge-m3-ru-ostap

BGE m3 for Ostap project

This is a sentence-transformers model finetuned from BAAI/bge-m3. 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
• Base model: BAAI/bge-m3
• Maximum Sequence Length: 8192 tokens
• Output Dimensionality: 1024 dimensions
• Similarity Function: Cosine Similarity
• Language: ru
• 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': 8192, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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("fitlemon/bge-m3-ru-ostap")
# Run inference
sentences = [
    'Какой у тебя любимый фильм?',
    'У нас хотя и не Париж, но кино у нас всегда с интригой!',
    'Фильм? Знойная женщина, мечта поэта — вот мой любимый сюжет!',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Evaluation

Metrics

Information Retrieval

• Datasets: dim_1024, dim_768, dim_512, dim_256, dim_128 and dim_64
• Evaluated with InformationRetrievalEvaluator

Metric	dim_1024	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.1493	0.146	0.1482	0.1449	0.1471	0.146
cosine_accuracy@3	0.2666	0.2655	0.2655	0.2688	0.2666	0.2699
cosine_accuracy@5	0.3429	0.3473	0.3451	0.3485	0.3462	0.3473
cosine_accuracy@10	0.4856	0.4867	0.4945	0.4912	0.4945	0.4978
cosine_precision@1	0.1493	0.146	0.1482	0.1449	0.1471	0.146
cosine_precision@3	0.0889	0.0885	0.0885	0.0896	0.0889	0.09
cosine_precision@5	0.0686	0.0695	0.069	0.0697	0.0692	0.0695
cosine_precision@10	0.0486	0.0487	0.0494	0.0491	0.0494	0.0498
cosine_recall@1	0.1493	0.146	0.1482	0.1449	0.1471	0.146
cosine_recall@3	0.2666	0.2655	0.2655	0.2688	0.2666	0.2699
cosine_recall@5	0.3429	0.3473	0.3451	0.3485	0.3462	0.3473
cosine_recall@10	0.4856	0.4867	0.4945	0.4912	0.4945	0.4978
cosine_ndcg@10	0.2943	0.2932	0.2966	0.2943	0.2964	0.2968
cosine_mrr@10	0.2362	0.2344	0.2365	0.2343	0.2362	0.2359
cosine_map@100	0.2601	0.2582	0.2598	0.258	0.2598	0.2592

Training Details

Training Dataset

Unnamed Dataset

• Size: 904 training samples
• Columns: question and answer
• Approximate statistics based on the first 904 samples:

  	question	answer
  type	string	string
  details	min: 6 tokens mean: 10.16 tokens max: 14 tokens	min: 8 tokens mean: 20.91 tokens max: 43 tokens

• Samples:

  question	answer
  Как ты проводишь свободное время?	Любителя бьют, а время — не ждет!
  Какой у тебя план на будущее?	План на будущее? Широкие массы миллиардеров уже составили его за меня.
  Какой у тебя любимый цвет?	Вы мне в конце концов не художник, не дизайнер и не стилист.

• Loss: MatryoshkaLoss with these parameters:

  {
      "loss": "MultipleNegativesRankingLoss",
      "matryoshka_dims": [
          1024,
          768,
          512,
          256,
          128,
          64
      ],
      "matryoshka_weights": [
          1,
          1,
          1,
          1,
          1,
          1
      ],
      "n_dims_per_step": -1
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• learning_rate: 2e-05
• num_train_epochs: 4
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• fp16: True
• tf32: False
• 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: 8
• per_device_eval_batch_size: 8
• 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: 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: False
• fp16: True
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: False
• 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_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
• 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: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	dim_1024_cosine_ndcg@10	dim_768_cosine_ndcg@10	dim_512_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.0885	10	6.8669	-	-	-	-	-	-
0.1770	20	4.9384	-	-	-	-	-	-
0.2655	30	3.1491	-	-	-	-	-	-
0.3540	40	2.5456	-	-	-	-	-	-
0.4425	50	3.6943	-	-	-	-	-	-
0.5310	60	1.8947	-	-	-	-	-	-
0.6195	70	2.1762	-	-	-	-	-	-
0.7080	80	1.9446	-	-	-	-	-	-
0.7965	90	1.5278	-	-	-	-	-	-
0.8850	100	2.0417	-	-	-	-	-	-
0.9735	110	3.7804	-	-	-	-	-	-
1.0	113	-	0.2751	0.2747	0.2761	0.2786	0.2764	0.2715
1.0619	120	1.9706	-	-	-	-	-	-
1.1504	130	1.7073	-	-	-	-	-	-
1.2389	140	1.3279	-	-	-	-	-	-
1.3274	150	1.2724	-	-	-	-	-	-
1.4159	160	2.4455	-	-	-	-	-	-
1.5044	170	0.5255	-	-	-	-	-	-
1.5929	180	2.5764	-	-	-	-	-	-
1.6814	190	1.56	-	-	-	-	-	-
1.7699	200	0.9105	-	-	-	-	-	-
1.8584	210	1.9859	-	-	-	-	-	-
1.9469	220	1.6355	-	-	-	-	-	-
2.0088	227	-	0.2837	0.2852	0.2880	0.2899	0.2926	0.2902
2.0265	230	0.6769	-	-	-	-	-	-
2.1150	240	0.764	-	-	-	-	-	-
2.2035	250	1.0598	-	-	-	-	-	-
2.2920	260	0.9267	-	-	-	-	-	-
2.3805	270	0.9687	-	-	-	-	-	-
2.4690	280	0.7875	-	-	-	-	-	-
2.5575	290	1.3853	-	-	-	-	-	-
2.6460	300	0.8114	-	-	-	-	-	-
2.7345	310	1.6069	-	-	-	-	-	-
2.8230	320	0.8149	-	-	-	-	-	-
2.9115	330	0.8858	-	-	-	-	-	-
3.0	340	0.7858	0.2920	0.2917	0.2929	0.2927	0.2967	0.2969
3.0885	350	0.5889	-	-	-	-	-	-
3.1770	360	0.3542	-	-	-	-	-	-
3.2655	370	0.5868	-	-	-	-	-	-
3.3540	380	0.4988	-	-	-	-	-	-
3.4425	390	0.4577	-	-	-	-	-	-
3.5310	400	0.4735	-	-	-	-	-	-
3.6195	410	1.2588	-	-	-	-	-	-
3.7080	420	0.6346	-	-	-	-	-	-
3.7965	430	0.3013	-	-	-	-	-	-
3.8850	440	0.6734	-	-	-	-	-	-
3.9735	450	0.3469	-	-	-	-	-	-
3.9912	452	-	0.2943	0.2932	0.2966	0.2943	0.2964	0.2968

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 3.4.1
• Transformers: 4.48.3
• PyTorch: 2.5.1+cu124
• Accelerate: 1.3.0
• Datasets: 3.3.2
• 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}
}