austinpatrickm/multilingual-e5-small-finetuned

SentenceTransformer based on intfloat/multilingual-e5-small

This is a sentence-transformers model finetuned from intfloat/multilingual-e5-small. 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: intfloat/multilingual-e5-small
• Maximum Sequence Length: 512 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': 512, 'do_lower_case': False, 'architecture': '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 = [
    'Can I automate the Gain image over time to create dynamic volume changes?',
    'Document_title: Harmor \nFile_name: plugins/Harmor.htm\nHeading_hierarchy: [Harmor -> About images and planes]\nAnchor_id: [none]\nThere are independent images that control the Pitch/Frequency and Gain of partials. Together these can create any sound, just as sampler can. In the image window the vertical dimension\n  is frequency (each line of pixels is a single partial), while the horizontal dimension is time.',
    'Document_title: PoiZone V2 \nFile_name: plugins/PoiZone.htm\nHeading_hierarchy: [PoiZone V2 -> Voicing]\nAnchor_id: [none]\n• 2 main oscillators for subtractive synthesis: SAW and PULSE shapes, pulse width adjustable. • 1 NOISE Oscillator. • Variable polyphony (1 to 32 voices).',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities)
# tensor([[1.0000, 0.4301, 0.1394],
#         [0.4301, 1.0000, 0.1051],
#         [0.1394, 0.1051, 1.0000]])

Evaluation

Metrics

Information Retrieval

• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.6911
cosine_accuracy@3	0.878
cosine_accuracy@5	0.9252
cosine_accuracy@10	0.9643
cosine_precision@1	0.6911
cosine_precision@3	0.2927
cosine_precision@5	0.185
cosine_precision@10	0.0964
cosine_recall@1	0.6911
cosine_recall@3	0.878
cosine_recall@5	0.9252
cosine_recall@10	0.9643
cosine_ndcg@10	0.834
cosine_mrr@10	0.7914
cosine_map@100	0.7931

Training Details

Training Dataset

Unnamed Dataset

• Size: 29,840 training samples
• Columns: sentence_0 and sentence_1
• Approximate statistics based on the first 1000 samples:

  	sentence_0	sentence_1
  type	string	string
  details	min: 9 tokens mean: 19.39 tokens max: 34 tokens	min: 29 tokens mean: 324.18 tokens max: 512 tokens

• Samples:

  sentence_0	sentence_1
  How do I load a *.SPEECH preset into the Sampler plugin?	Document_title: Speech Preset (.SPEECH) File_name: fformats_sample_speech.htm Heading_hierarchy: [Speech Preset (.SPEECH)] Anchor_id: [none] The Speech synthesizer processes text to create computerized or Vocoder-like vocals to your projects. The *.SPEECH presets are supported by all native FL Studio plugins that use custom samples for synthesizing, i.e. Sampler , Granulizer , Fruity Slicer and Fruity Scratcher .
  Can I use Speech presets with the Fruity Granulizer to create vocal effects?	Document_title: Speech Preset (.SPEECH) File_name: fformats_sample_speech.htm Heading_hierarchy: [Speech Preset (.SPEECH)] Anchor_id: [none] The Speech synthesizer processes text to create computerized or Vocoder-like vocals to your projects. The *.SPEECH presets are supported by all native FL Studio plugins that use custom samples for synthesizing, i.e. Sampler , Granulizer , Fruity Slicer and Fruity Scratcher .
  What kind of vocals can the Speech synthesizer create from text?	Document_title: Speech Preset (.SPEECH) File_name: fformats_sample_speech.htm Heading_hierarchy: [Speech Preset (.SPEECH)] Anchor_id: [none] The Speech synthesizer processes text to create computerized or Vocoder-like vocals to your projects. The *.SPEECH presets are supported by all native FL Studio plugins that use custom samples for synthesizing, i.e. Sampler , Granulizer , Fruity Slicer and Fruity Scratcher .

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 10
• per_device_eval_batch_size: 10
• num_train_epochs: 2
• multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 10
• per_device_eval_batch_size: 10
• 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: 2
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: None
• warmup_ratio: None
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• enable_jit_checkpoint: False
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• use_cpu: False
• seed: 42
• data_seed: None
• bf16: False
• fp16: False
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: -1
• ddp_backend: None
• debug: []
• dataloader_drop_last: False
• dataloader_num_workers: 0
• dataloader_prefetch_factor: None
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• 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
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_for_metrics: []
• eval_do_concat_batches: True
• auto_find_batch_size: False
• full_determinism: False
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• use_cache: False
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: round_robin
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Click to expand

Epoch	Step	Training Loss	cosine_ndcg@10
0.0168	50	-	0.7103
0.0335	100	-	0.7267
0.0503	150	-	0.7500
0.0670	200	-	0.7715
0.0838	250	-	0.7892
0.1005	300	-	0.7921
0.1173	350	-	0.7940
0.1340	400	-	0.7958
0.1508	450	-	0.7889
0.1676	500	0.3978	0.7999
0.1843	550	-	0.7861
0.2011	600	-	0.7848
0.2178	650	-	0.7780
0.2346	700	-	0.7885
0.2513	750	-	0.7926
0.2681	800	-	0.7914
0.2849	850	-	0.8043
0.3016	900	-	0.7939
0.3184	950	-	0.8057
0.3351	1000	0.1115	0.8093
0.3519	1050	-	0.8056
0.3686	1100	-	0.7941
0.3854	1150	-	0.8042
0.4021	1200	-	0.8007
0.4189	1250	-	0.8071
0.4357	1300	-	0.8121
0.4524	1350	-	0.8037
0.4692	1400	-	0.7958
0.4859	1450	-	0.8052
0.5027	1500	0.0989	0.8028
0.5194	1550	-	0.7989
0.5362	1600	-	0.8078
0.5529	1650	-	0.8117
0.5697	1700	-	0.8108
0.5865	1750	-	0.8101
0.6032	1800	-	0.8102
0.6200	1850	-	0.8080
0.6367	1900	-	0.8150
0.6535	1950	-	0.8156
0.6702	2000	0.0901	0.8138
0.6870	2050	-	0.8127
0.7038	2100	-	0.8123
0.7205	2150	-	0.8128
0.7373	2200	-	0.8141
0.7540	2250	-	0.8108
0.7708	2300	-	0.8108
0.7875	2350	-	0.8164
0.8043	2400	-	0.8159
0.8210	2450	-	0.8175
0.8378	2500	0.0908	0.8206
0.8546	2550	-	0.8223
0.8713	2600	-	0.8238
0.8881	2650	-	0.8264
0.9048	2700	-	0.8212
0.9216	2750	-	0.8204
0.9383	2800	-	0.8236
0.9551	2850	-	0.8170
0.9718	2900	-	0.8217
0.9886	2950	-	0.8246
1.0	2984	-	0.8222
1.0054	3000	0.0868	0.8207
1.0221	3050	-	0.8173
1.0389	3100	-	0.8165
1.0556	3150	-	0.8211
1.0724	3200	-	0.8236
1.0891	3250	-	0.8207
1.1059	3300	-	0.8173
1.1227	3350	-	0.8197
1.1394	3400	-	0.8164
1.1562	3450	-	0.8212
1.1729	3500	0.0611	0.8225
1.1897	3550	-	0.8250
1.2064	3600	-	0.8256
1.2232	3650	-	0.8253
1.2399	3700	-	0.8254
1.2567	3750	-	0.8254
1.2735	3800	-	0.8284
1.2902	3850	-	0.8324
1.3070	3900	-	0.8311
1.3237	3950	-	0.8272
1.3405	4000	0.0581	0.8245
1.3572	4050	-	0.8227
1.3740	4100	-	0.8235
1.3908	4150	-	0.8211
1.4075	4200	-	0.8199
1.4243	4250	-	0.8230
1.4410	4300	-	0.8248
1.4578	4350	-	0.8266
1.4745	4400	-	0.8268
1.4913	4450	-	0.8273
1.5080	4500	0.0499	0.8305
1.5248	4550	-	0.8293
1.5416	4600	-	0.8291
1.5583	4650	-	0.8287
1.5751	4700	-	0.8285
1.5918	4750	-	0.8286
1.6086	4800	-	0.8289
1.6253	4850	-	0.8277
1.6421	4900	-	0.8283
1.6588	4950	-	0.8287
1.6756	5000	0.0595	0.8285
1.6924	5050	-	0.8289
1.7091	5100	-	0.8274
1.7259	5150	-	0.8277
1.7426	5200	-	0.8296
1.7594	5250	-	0.8326
1.7761	5300	-	0.8323
1.7929	5350	-	0.8308
1.8097	5400	-	0.8312
1.8264	5450	-	0.8314
1.8432	5500	0.0544	0.8328
1.8599	5550	-	0.8331
1.8767	5600	-	0.8327
1.8934	5650	-	0.8335
1.9102	5700	-	0.8340

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.3
• Transformers: 5.0.0
• PyTorch: 2.10.0+cu128
• Accelerate: 1.12.0
• Datasets: 4.0.0
• Tokenizers: 0.22.2

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",
}

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}
}