metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - generated_from_trainer
  - dataset_size:98646
  - loss:MultipleNegativesRankingLoss
base_model: intfloat/multilingual-e5-base
widget:
  - source_sentence: >-
      Instruct: Given a dialogue context, retrieve relevant followup phrase that
      align with the context

      Dialogue Context: bot_0: Hey, how are things today?
    sentences:
      - >-
        Followup phrase: i am painting a mural in bedroom of Sunset on beach
        with black backgound.
      - 'Followup phrase: I have a kitten.'
      - 'Followup phrase: I live in Georgia.'
  - source_sentence: >-
      Instruct: Given a dialogue context, retrieve relevant followup phrase that
      align with the context

      Dialogue Context: bot_0: Greetings from the urban jungle!

      bot_1: Hi there from the beach!

      bot_0: I just dyed my hair red to match my parrot's feathers.

      bot_1: That's funny. I dyed my hair brown to match my horses.

      bot_0: Cool! Her name is tasha and I became vegetarian because of her.

      bot_1: I heard bar b que parrot can be good but I want to be vegetarian
      too.
    sentences:
      - 'Followup phrase: I bought deer meat but haven''t used it yet.'
      - 'Followup phrase: I am attending law school. I have one more year to go.'
      - 'Followup phrase: My favorite thing to do is hike. I like eggplant jerky.'
  - source_sentence: >-
      Instruct: Given a dialogue context, retrieve relevant followup phrase that
      align with the context

      Dialogue Context: bot_0: Hi. I'm turning 29 soon. How old are you?

      bot_1: I'm 97 and I am vegan.

      bot_0: Wow. That's great. What are you doing for spring break.

      bot_1: I'm going to hunt animals and eat them..

      bot_0: I'm going to the beach. I thought you were vegan?

      bot_1: It comes and goes I change from second to second..

      bot_0: I'm driving to the beach in my new car my mom bought.

      bot_1: Oh ok I'm driving a tomato my mom bought me..

      bot_0: I'm going to wear a new dress I bought for easter.

      bot_1: That sounds so interesting... Wow.

      bot_0: What else do you do?
    sentences:
      - 'Followup phrase: My favorite cake is chocolate with extra frosting.'
      - >-
        Followup phrase: I went to school in San Francisco. I love the city. I
        live in San Francisco.
      - >-
        Followup phrase: I am thinking of traveling to Ireland to catch a
        leprechaun.
  - source_sentence: >-
      Instruct: Given a dialogue context, retrieve relevant followup phrase that
      align with the context

      Dialogue Context: bot_0: Hi, I want to talk to you about god.

      bot_1: Do you speak more than one language, I do.

      bot_0: Well, yes, I speak a little spanish, un poco.

      bot_1: That's cool I speak both spanish and english very fluently.

      bot_0: Magnifico! I need to talk to some of my workers in spanish, daily.

      bot_1: Ah, so you must be older than me as I am still in school.

      bot_0: Yes, I am middle-aged and work in a clothing factory with lots of
      menial workers.

      bot_1: What do you do in your spare time, I like to hike and walk on the
      beach.
    sentences:
      - 'Followup phrase: I am not familiar with Insane Clown Posse.'
      - >-
        Followup phrase: I am not a health nut. My family is not vegan. We try
        to avoid processed foods.
      - 'Followup phrase: I have Yorkshire Terrier puppies.'
  - source_sentence: >-
      Instruct: Given a dialogue context, retrieve relevant followup phrase that
      align with the context

      Dialogue Context: bot_0: Hi, I love country music. Do you have a favorite
      type of music?

      bot_1: I like classic country music, like patsy cline and hank snow.

      bot_0: That's awesome. I live in the country so it fits me, where do you
      live?

      bot_1: Up in alpine county ca. It's lovely up here in early spring.

      bot_0: I like spring, but winter is my favorite time of year.

      bot_1: I am so done with winter. Tough getting the kids to school. No snow
      days.

      bot_0: I understand, we live on a corn and bean farm and it can be hard.

      bot_1: I love farm-fresh organic food. It's all I make.

      bot_0: That's nice. I love living on the farm and driving the tractors.

      bot_1: I could do that! I would love farm life, but we live in the
      mountains.

      bot_0: That would be amazing! Have you lived there your entire life?

      bot_1: Yes, born and raised here in the sierras. We ski and snowmobile.

      bot_0: Amazing, were your parents from there also? My father is a
      preacher.
    sentences:
      - >-
        Followup phrase: My great-grandpa came to the Sierras in 1903. My
        great-grandpa came to mine gold.
      - 'Followup phrase: I read stealth and espionage themed books.'
      - >-
        Followup phrase: I majored in finance. I'm looking for a job in finance.
        I have a big sweet tooth.
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy
  - cosine_accuracy_threshold
  - cosine_f1
  - cosine_f1_threshold
  - cosine_precision
  - cosine_recall
  - cosine_ap
  - cosine_mcc
model-index:
  - name: SentenceTransformer based on intfloat/multilingual-e5-base
    results:
      - task:
          type: binary-classification
          name: Binary Classification
        dataset:
          name: Unknown
          type: unknown
        metrics:
          - type: cosine_accuracy
            value: 0.8977793653628624
            name: Cosine Accuracy
          - type: cosine_accuracy_threshold
            value: 0.7719855308532715
            name: Cosine Accuracy Threshold
          - type: cosine_f1
            value: 0.6526667569886646
            name: Cosine F1
          - type: cosine_f1_threshold
            value: 0.7546476721763611
            name: Cosine F1 Threshold
          - type: cosine_precision
            value: 0.6602101416171768
            name: Cosine Precision
          - type: cosine_recall
            value: 0.6452938024647259
            name: Cosine Recall
          - type: cosine_ap
            value: 0.7302000254958843
            name: Cosine Ap
          - type: cosine_mcc
            value: 0.5842050499797425
            name: Cosine Mcc

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: Hi, I love country music. Do you have a favorite type of music?\nbot_1: I like classic country music, like patsy cline and hank snow.\nbot_0: That's awesome. I live in the country so it fits me, where do you live?\nbot_1: Up in alpine county ca. It's lovely up here in early spring.\nbot_0: I like spring, but winter is my favorite time of year.\nbot_1: I am so done with winter. Tough getting the kids to school. No snow days.\nbot_0: I understand, we live on a corn and bean farm and it can be hard.\nbot_1: I love farm-fresh organic food. It's all I make.\nbot_0: That's nice. I love living on the farm and driving the tractors.\nbot_1: I could do that! I would love farm life, but we live in the mountains.\nbot_0: That would be amazing! Have you lived there your entire life?\nbot_1: Yes, born and raised here in the sierras. We ski and snowmobile.\nbot_0: Amazing, were your parents from there also? My father is a preacher.",
    'Followup phrase: My great-grandpa came to the Sierras in 1903. My great-grandpa came to mine gold.',
    'Followup phrase: I read stealth and espionage themed books.',
]
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	0.8978
cosine_accuracy_threshold	0.772
cosine_f1	0.6527
cosine_f1_threshold	0.7546
cosine_precision	0.6602
cosine_recall	0.6453
cosine_ap	0.7302
cosine_mcc	0.5842

Training Details

Training Dataset

Unnamed Dataset

Size: 98,646 training samples
Columns: sentence1 and sentence2
Approximate statistics based on the first 1000 samples:
sentence1 sentence2
type string string
details
min: 35 tokens
mean: 143.84 tokens
max: 319 tokens

min: 9 tokens
mean: 15.99 tokens
max: 36 tokens

	sentence1	sentence2
type	string	string
details	min: 35 tokens mean: 143.84 tokens max: 319 tokens	min: 9 tokens mean: 15.99 tokens max: 36 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 live semi-close towkr. I don't own a car. I enjoy running and walking. I live in a small town.`
`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: I enjoy exercising at the gym.`
`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.`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

Size: 67,188 evaluation samples
Columns: sentence1, sentence2, and label
Approximate statistics based on the first 1000 samples:
sentence1 sentence2 label
type string string int
details
min: 39 tokens
mean: 146.95 tokens
max: 294 tokens

min: 10 tokens
mean: 18.06 tokens
max: 74 tokens

0: ~83.30%
1: ~16.70%

	sentence1	sentence2	label
type	string	string	int
details	min: 39 tokens mean: 146.95 tokens max: 294 tokens	min: 10 tokens mean: 18.06 tokens max: 74 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: Hi, I'm shelby, nice to meet you!`	`Followup phrase: Name is Jessica.`	`1`
`Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: Hi, I'm shelby, nice to meet you!`	`Followup phrase: I think fast food is dangerous.`	`0`
`Instruct: Given a dialogue context, retrieve relevant followup phrase that align with the context Dialogue Context: bot_0: Hi, I'm shelby, nice to meet you!`	`Followup phrase: I also have many allergies. I get hay fever and it makes me sneeze.`	`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	2.4265	-	-
0.2026	200	2.1867	-	-
0.3040	300	2.1658	-	-
0.4053	400	2.0992	-	-
0.5066	500	2.0569	-	-
0.6079	600	2.0444	-	-
0.7092	700	2.0274	-	-
0.8105	800	2.0095	-	-
0.9119	900	2.0072	-	-
1.0	987	-	6.0145	0.7222
1.0132	1000	1.9768	-	-
1.1145	1100	1.932	-	-
1.2158	1200	1.9352	-	-
1.3171	1300	1.8966	-	-
1.4184	1400	1.9461	-	-
1.5198	1500	1.8999	-	-
1.6211	1600	1.9098	-	-
1.7224	1700	1.9049	-	-
1.8237	1800	1.9113	-	-
1.9250	1900	1.9211	-	-
2.0	1974	-	6.2014	0.7302

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}
}