Field-adaptive-bi-encoder / README.md

mudasir13cs

Add README.md

b93e3f8 verified about 1 month ago

preview code

raw

history blame

27.4 kB

metadata

tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - dense
  - generated_from_trainer
  - dataset_size:32000
  - loss:MultipleNegativesRankingLoss
base_model: sentence-transformers/all-MiniLM-L6-v2
widget:
  - source_sentence: how to present delivery offers creatively
    sentences:
      - >-
        Introducing the "Delivery Offer with Fresh Raw Plums" presentation
        template, featuring a vibrant poster adorned with lush plums, some
        artfully flying through the air. Perfect for marketing ads, business
        finance, and fashion style industries, this template adds a fresh twist
        to presentations. Ideal for holidays, celebrations, and city-themed
        events, it captures the essence of express courier and delivery
        services. Engage your audience with this visually captivating and
        professionally designed template.
      - >-
        Elevate your presentations with the "Inclusive Urban Design Contribution
        Gratitude" template. This visually engaging design features a
        certificate motif, perfect for recognizing achievements in construction
        and urban development. Ideal for industries like Marketing Ads,
        Entertainment Leisure, and Business Finance, this template is also
        suited for Holidays Celebration, Fashion Style, and Travels Vacations
        contexts. Seamlessly blend Courier, Delivery, and Celebration themes to
        captivate your audience and underscore your message with professional
        flair.
      - >-
        Unleash your creativity with the "Volunteer Work Quote with Animal
        Skull" presentation template. Featuring a striking black and white image
        of a ram, this graphic design is perfect for industries like Marketing
        Ads, Entertainment Leisure, and Services. Ideal for Holidays
        Celebration, Food & Drinks, and Fashion Style presentations, this
        template captivates with its artistic flair. Engage your audience with
        its bold visual elements, making it a standout choice for those seeking
        impactful, professional presentations. Keywords: graphic design, animal
        skull, creative presentation.
  - source_sentence: leisure activities presentation style
    sentences:
      - >-
        Elevate your presentations with the "Fashion Quote Businessman Wearing
        Suit in Grey" template. Featuring sleek visuals of a confident man in a
        suit and tie, this template embodies sophistication and style. Ideal for
        industries like Marketing Ads, Entertainment, Leisure, and Business
        Finance, it seamlessly fits categories such as Holidays Celebration and
        Food & Drinks. Perfect for presentations focused on celebrating,
        socializing, and enjoying life's finer moments, this template ensures
        your message is both impactful and memorable.
      - >-
        Dive into a lush paradise with our "Amazing Tropical Vegetation"
        presentation template. Featuring vibrant green plant imagery against a
        chic white and pink striped backdrop, this template exudes a lively yet
        professional aesthetic. Perfect for marketing ads, entertainment
        projects, or business finance presentations, it's ideal for holiday
        celebrations, leisure activities, or pet-related content. Captivate your
        audience with this versatile, visually stunning template, bringing a
        touch of tropical flair to your next project.
      - >-
        Introducing the "Nature Expo Announcement: Blooming Daisy Flower"
        presentation template, a vibrant blend of lush green fields and pristine
        white flowers, perfect for capturing the essence of nature. Ideal for
        marketing ads, entertainment, and service industries, this template is
        perfect for holidays, celebrations, and food and drinks events. With
        elements like music notes and food frames, it's designed for those
        celebrating nature's beauty. Engage your audience with this captivating,
        nature-inspired template. Perfect for creating an unforgettable Nature
        Expo 2019 presentation.
  - source_sentence: Gaming championship announcement design
    sentences:
      - >-
        Unleash the fun with our Video Games Championship Announcement template!
        Featuring a sleek black background and the intriguing phrase, "Is this
        video game your passion without challenges?" this template is perfect
        for marketing ads and entertainment events. Ideal for holidays,
        celebrations, and leisure activities, it seamlessly integrates a food
        frame for added flair. Engage your audience with its amusing style,
        making it perfect for celebrating gaming enthusiasts and their love for
        competition.
      - >-
        Elevate your presentations with the "Teacher Helping Kids" template,
        featuring heartwarming visuals of a dedicated woman assisting a young
        boy in a vibrant classroom setting. Ideal for marketing ads,
        entertainment, and service industries, this template seamlessly fits
        holiday celebrations and food and drink events. Perfect for parties,
        socializing, and cultural gatherings, it brings an engaging,
        professional touch to your message. Inspire your audience with these
        captivating, relatable scenes that highlight education and community.
      - >-
        Unleash your creativity with the "Do It Yourself Inspirational Banner"
        presentation template. Featuring a sleek black and white logo with the
        phrase 'dot yourself,' this design is framed by a vibrant
        yellow-bordered white background, exuding a modern yet professional
        aesthetic. Perfect for marketing ads, entertainment, and business
        finance sectors, this template is ideal for celebrating holidays,
        showcasing food and drinks, or highlighting cities and places. Keywords:
        courier, delivery, express.
  - source_sentence: holiday celebration presentation design
    sentences:
      - >-
        Elevate your presentations with the "Digital Photography Tips with
        Camera" template, featuring a sleek dark blue background accented by
        vibrant lights and a striking black and white clock image. Perfect for
        professionals in marketing ads, entertainment, and leisure services,
        this template is ideal for holidays, food and drinks, and fashion style
        presentations. Capture your audience's attention and boost your
        Instagram presence with this visually stunning, professional template
        designed for engaging storytelling.
      - >-
        Elevate your marketing campaigns with the "Antique Furniture Ad with
        Luxury Armchair" presentation template. Featuring a sophisticated visual
        style, this template seamlessly integrates a furniture store logo with
        an elegant armchair centerpiece. Perfect for marketing ads in the
        entertainment and leisure industry, business finance presentations, and
        holiday celebrations, this template captures attention effortlessly.
        Ideal for use in courier, delivery, and travel sectors, it ensures your
        message is delivered with timeless elegance and professional flair.
      - >-
        The "Happy Children at Kids Camp" presentation template features vibrant
        visuals of joyful children sitting on lush grass, set against a lively
        green banner backdrop. Perfect for marketing ads, entertainment,
        leisure, and business finance sectors, this template is ideal for
        holiday celebrations, food and drinks promotions, and fashion style
        events. With its engaging food frame and informational infographics,
        it's designed to captivate audiences while effectively conveying your
        message. Keywords: kids camp, celebrations, engaging visuals, marketing.
  - source_sentence: engaging slides for food and drink theme
    sentences:
      - >-
        Elevate your presentations with the "Android Robot Hand" template,
        featuring a sleek, futuristic design. This template includes a
        captivating white robot set against a dynamic blue background,
        complemented by a blue and white bar chart emphasizing the number 1.
        Perfect for industries like Marketing Ads, Business Finance, and Fashion
        Style, it's ideal for discussions on trends and innovations. Engage your
        audience with this modern, trend-focused template, suitable for Courier,
        Delivery, and Express services.
      - >-
        Discover the "Cycling Club Tips" presentation template, featuring a
        minimalistic design with a striking black and white diagonal striped
        background and an image of a woman cycling against a crisp white
        backdrop. Perfect for marketing ads, entertainment, leisure, and
        business finance sectors, this template is ideal for creating engaging
        holiday celebrations or food and drinks presentations. Seamlessly
        integrate keywords like courier, delivery, and express to captivate your
        audience and elevate your message with style and impact.
      - >-
        Elevate your presentations with the "Smartphone Review Man Scrolling
        Phone" template, featuring a sleek design with dynamic visuals of a
        person holding a smartphone against a modern white background adorned
        with triangles. Perfect for marketing ads, business finance, and fashion
        style industries, this template is ideal for holidays, food and drinks,
        and leisure entertainment themes. Enhance your courier and delivery
        presentations or celebration pitches with this versatile, professional
        template that captivates and engages your audience effortlessly.
pipeline_tag: sentence-similarity
library_name: sentence-transformers
metrics:
  - cosine_accuracy@1
  - cosine_accuracy@3
  - cosine_accuracy@5
  - cosine_accuracy@10
  - cosine_precision@1
  - cosine_precision@3
  - cosine_precision@5
  - cosine_precision@10
  - cosine_recall@1
  - cosine_recall@3
  - cosine_recall@5
  - cosine_recall@10
  - cosine_ndcg@10
  - cosine_mrr@10
  - cosine_map@1
  - cosine_map@5
  - cosine_map@10
model-index:
  - name: SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2
    results:
      - task:
          type: information-retrieval
          name: Information Retrieval
        dataset:
          name: validation
          type: validation
        metrics:
          - type: cosine_accuracy@1
            value: 0.45
            name: Cosine Accuracy@1
          - type: cosine_accuracy@3
            value: 0.62
            name: Cosine Accuracy@3
          - type: cosine_accuracy@5
            value: 0.685625
            name: Cosine Accuracy@5
          - type: cosine_accuracy@10
            value: 0.765625
            name: Cosine Accuracy@10
          - type: cosine_precision@1
            value: 0.45
            name: Cosine Precision@1
          - type: cosine_precision@3
            value: 0.20666666666666664
            name: Cosine Precision@3
          - type: cosine_precision@5
            value: 0.137125
            name: Cosine Precision@5
          - type: cosine_precision@10
            value: 0.0765625
            name: Cosine Precision@10
          - type: cosine_recall@1
            value: 0.45
            name: Cosine Recall@1
          - type: cosine_recall@3
            value: 0.62
            name: Cosine Recall@3
          - type: cosine_recall@5
            value: 0.685625
            name: Cosine Recall@5
          - type: cosine_recall@10
            value: 0.765625
            name: Cosine Recall@10
          - type: cosine_ndcg@10
            value: 0.6029385769142473
            name: Cosine Ndcg@10
          - type: cosine_mrr@10
            value: 0.5515214533730152
            name: Cosine Mrr@10
          - type: cosine_map@1
            value: 0.45
            name: Cosine Map@1
          - type: cosine_map@5
            value: 0.5409479166666666
            name: Cosine Map@5
          - type: cosine_map@10
            value: 0.5515214533730158
            name: Cosine Map@10

SentenceTransformer based on sentence-transformers/all-MiniLM-L6-v2

This is a sentence-transformers model finetuned from sentence-transformers/all-MiniLM-L6-v2. 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: sentence-transformers/all-MiniLM-L6-v2
Maximum Sequence Length: 256 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': 256, '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 = [
    'engaging slides for food and drink theme',
    'Elevate your presentations with the "Smartphone Review Man Scrolling Phone" template, featuring a sleek design with dynamic visuals of a person holding a smartphone against a modern white background adorned with triangles. Perfect for marketing ads, business finance, and fashion style industries, this template is ideal for holidays, food and drinks, and leisure entertainment themes. Enhance your courier and delivery presentations or celebration pitches with this versatile, professional template that captivates and engages your audience effortlessly.',
    'Discover the "Cycling Club Tips" presentation template, featuring a minimalistic design with a striking black and white diagonal striped background and an image of a woman cycling against a crisp white backdrop. Perfect for marketing ads, entertainment, leisure, and business finance sectors, this template is ideal for creating engaging holiday celebrations or food and drinks presentations. Seamlessly integrate keywords like courier, delivery, and express to captivate your audience and elevate your message with style and impact.',
]
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, 0.1884, 0.1604],
#         [0.1884, 1.0000, 0.2392],
#         [0.1604, 0.2392, 1.0000]])

Evaluation

Metrics

Information Retrieval

Dataset: validation
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.45
cosine_accuracy@3	0.62
cosine_accuracy@5	0.6856
cosine_accuracy@10	0.7656
cosine_precision@1	0.45
cosine_precision@3	0.2067
cosine_precision@5	0.1371
cosine_precision@10	0.0766
cosine_recall@1	0.45
cosine_recall@3	0.62
cosine_recall@5	0.6856
cosine_recall@10	0.7656
cosine_ndcg@10	0.6029
cosine_mrr@10	0.5515
cosine_map@1	0.45
cosine_map@5	0.5409
cosine_map@10	0.5515

Training Details

Training Dataset

Unnamed Dataset

Size: 32,000 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: 5 tokens
mean: 7.11 tokens
max: 12 tokens

min: 82 tokens
mean: 100.09 tokens
max: 133 tokens

	sentence_0	sentence_1
type	string	string
details	min: 5 tokens mean: 7.11 tokens max: 12 tokens	min: 82 tokens mean: 100.09 tokens max: 133 tokens

Samples:

sentence_0	sentence_1
`entertainment-themed presentation slides`	Ignite creativity with our "Reading Inspiration Books on Shelves" presentation template. Featuring a charming visual style with piles of books on a table, this template is perfect for industries like Marketing, Entertainment, and Services. Ideal for Holidays, Celebrations, and Leisure topics, it seamlessly integrates themes of Food, Drinks, and Games. Captivate your audience with a professional yet engaging backdrop that celebrates creativity and leisure. Perfect for marketers looking to inspire and entertain.
`fashion-forward slides for holidays and celebrations`	Elevate your presentations with the "Insurance Company Successful Business Team" template, featuring a sleek design showcasing an insurance logo and a dynamic duo seated on a couch. A pink shield on a pristine white background adds a touch of elegance. Perfect for marketing ads, entertainment, and fashion, it's ideal for holidays, celebrations, or pet-related themes. Keywords like "courier," "beauty," and "celebration" seamlessly blend, making it a captivating choice for professionals seeking style and impact.
`How to promote a decor event template`	`Unveil your event with our "Interior Decoration Event Announcement Sofa in Grey" template, featuring a chic living room setting with a stylish grey couch and vibrant green plant. Perfect for marketing ads in the entertainment and leisure industries, this template is ideal for holiday celebrations or home decor events. Capture attention with its modern aesthetic and versatile design, seamlessly integrating keywords like courier, delivery, and express to boost your promotional efforts.`

Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
num_train_epochs: 20
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: steps
prediction_loss_only: True
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
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: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 1
num_train_epochs: 20
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: 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: False
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
hub_revision: None
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
liger_kernel_config: None
eval_use_gather_object: False
average_tokens_across_devices: False
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	validation_cosine_ndcg@10
1.0	500	2.3734	0.4626
2.0	1000	1.9095	0.4966
3.0	1500	1.7464	0.5176
4.0	2000	1.6538	0.5309
5.0	2500	1.5949	0.5425
6.0	3000	1.5507	0.5519
7.0	3500	1.5173	0.5605
8.0	4000	1.4871	0.5669
9.0	4500	1.4587	0.5729
10.0	5000	1.4309	0.5763
11.0	5500	1.4214	0.5805
12.0	6000	1.4028	0.5852
13.0	6500	1.3867	0.5894
14.0	7000	1.3745	0.5945
15.0	7500	1.3625	0.5950
16.0	8000	1.3516	0.5982
17.0	8500	1.3453	0.6001
18.0	9000	1.3448	0.6019
19.0	9500	1.3327	0.6023
20.0	10000	1.3323	0.6029

Framework Versions

Python: 3.10.18
Sentence Transformers: 5.0.0
Transformers: 4.54.0.dev0
PyTorch: 2.6.0+cu124
Accelerate: 1.8.1
Datasets: 3.6.0
Tokenizers: 0.21.2

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}
}