adugeen/phrases-reranker-e5-base

SentenceTransformer based on intfloat/multilingual-e5-base

This is a sentence-transformers model finetuned from intfloat/multilingual-e5-base. 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: intfloat/multilingual-e5-base
• Maximum Sequence Length: 512 tokens
• Output Dimensionality: 768 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}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, '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 = [
    "Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context\nDialogue Context: bot_0: Hello, I just got back from class. What are you doing?\nbot_1: I just got done working out at the gym.\nbot_0: Cool, what is your favorite exercise?\nbot_1: Do you have your own vehicle?\nbot_0: No, I am a student. I walk everywhere or I take the bus.\nbot_1: Oh wow, that must get tiring. Do you have a significant other?\nbot_0: It's not, I even have energy to play baseball. I do not, I am single.\nbot_1: Thats awesome that you have the energy. My significant other is a lawyer. We're married..\nbot_0: Awe, I hope to have a job designing ads one day.\nbot_1: That sounds neat. Are you a vegetarian?\nbot_0: No, but have thought about it!",
    'Followup phrase: I do not. My husband wants a boy, he is in the army.',
    'Followup phrase: I am amazing, except I found out I am allergic to fish!',
]
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

Binary Classification

• Evaluated with BinaryClassificationEvaluator

Metric	Value
cosine_accuracy	0.9325
cosine_accuracy_threshold	0.6963
cosine_f1	0.7933
cosine_f1_threshold	0.6896
cosine_precision	0.7918
cosine_recall	0.7948
cosine_ap	0.8752
cosine_mcc	0.7518

Training Details

Training Dataset

Unnamed Dataset

• Size: 98,660 training samples
• Columns: sentence1 and sentence2
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2
  type	string	string
  details	min: 35 tokens mean: 144.27 tokens max: 319 tokens	min: 10 tokens mean: 22.54 tokens max: 41 tokens

• Samples:

  sentence1	sentence2
  Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: What kind of car do you own? I have a jeep.	Followup phrase: I don't own my own car! I actually really enjoying walking and running, but then again, I live in a small town and semi-close to work.
  Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: What kind of car do you own? I have a jeep. bot_1: I don't own my own car! I actually really enjoying walking and running, but then again, I live in a small town and semi-close to work.	Followup phrase: Ah I see! I like going to the gym to work out.
  Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: What kind of car do you own? I have a jeep. bot_1: I don't own my own car! I actually really enjoying walking and running, but then again, I live in a small town and semi-close to work. bot_0: Ah I see! I like going to the gym to work out.	Followup phrase: I'm a computer programmer. What do you do for work.

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 100,
      "similarity_fct": "cos_sim"
  }
  

Evaluation Dataset

Unnamed Dataset

• Size: 67,104 evaluation samples
• Columns: sentence1, sentence2, and label
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	label
  type	string	string	int
  details	min: 38 tokens mean: 137.57 tokens max: 290 tokens	min: 11 tokens mean: 31.57 tokens max: 106 tokens	0: ~83.30% 1: ~16.70%

• Samples:

  sentence1	sentence2	label
  Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: Do you like music?	Followup phrase: Yes, you could say it is a great source of joy for me.	1
  Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: Do you like music?	Followup phrase: That sounds amazing! But I was thinking of going to mexico this summer and was going to ask if you were going to be there? Would your timeshare be available?	0
  Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: Do you like music?	Followup phrase: Mostly just authentic mexican food, with lots of spice.	0

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 100,
      "similarity_fct": "cos_sim"
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 100
• per_device_eval_batch_size: 100
• weight_decay: 0.01
• num_train_epochs: 5
• bf16: True
• load_best_model_at_end: True
• prompts: {'sentence1': 'Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context\nDialogue Context: ', 'sentence2': 'Followup phrase: '}
• 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: 100
• per_device_eval_batch_size: 100
• 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: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• 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: True
• 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: 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: {'sentence1': 'Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context\nDialogue Context: ', 'sentence2': 'Followup phrase: '}
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	cosine_ap
0.1013	100	1.8292	-	-
0.2026	200	1.4433	-	-
0.3040	300	1.2605	-	-
0.4053	400	1.1947	-	-
0.5066	500	1.1714	-	-
0.6079	600	1.1106	-	-
0.7092	700	1.0978	-	-
0.8105	800	1.0527	-	-
0.9119	900	1.0524	-	-
1.0	987	-	8.1109	0.8790
1.0132	1000	1.0068	-	-
1.1145	1100	0.949	-	-
1.2158	1200	0.9519	-	-
1.3171	1300	0.9364	-	-
1.4184	1400	0.9253	-	-
1.5198	1500	0.9724	-	-
1.6211	1600	0.9227	-	-
1.7224	1700	0.9169	-	-
1.8237	1800	0.9146	-	-
1.9250	1900	0.9029	-	-
2.0	1974	-	8.4529	0.8727
2.0263	2000	0.9073	-	-
2.1277	2100	0.8685	-	-
2.2290	2200	0.8413	-	-
2.3303	2300	0.8763	-	-
2.4316	2400	0.8524	-	-
2.5329	2500	0.8729	-	-
2.6342	2600	0.856	-	-
2.7356	2700	0.8652	-	-
2.8369	2800	0.8768	-	-
2.9382	2900	0.8477	-	-
3.0	2961	-	8.7662	0.8752

Framework Versions

• Python: 3.10.18
• Sentence Transformers: 4.1.0
• Transformers: 4.52.4
• PyTorch: 2.7.1+cu128
• Accelerate: 1.7.0
• Datasets: 3.6.0
• 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",
}

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}
}