SentenceTransformer based on sentence-transformers/clip-ViT-B-32

This is a sentence-transformers model finetuned from sentence-transformers/clip-ViT-B-32 on the unsplash-lite dataset. It maps sentences & paragraphs to a 512-dimensional dense vector space and can be used for retrieval.

Model Details

Model Description

Model Type: Sentence Transformer
Base model: sentence-transformers/clip-ViT-B-32
Maximum Sequence Length: 77 tokens
Output Dimensionality: 512 dimensions
Similarity Function: Cosine Similarity
Supported Modalities: Text, Image
Training Dataset:
- unsplash-lite

Model Sources

Documentation: Sentence Transformers Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Sentence Transformers on Hugging Face

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'transformer_task': 'feature-extraction', 'modality_config': {'text': {'method': 'get_text_features', 'method_output_name': 'pooler_output'}, 'image': {'method': 'get_image_features', 'method_output_name': 'pooler_output'}}, 'module_output_name': 'sentence_embedding', 'architecture': 'CLIPModel'})
)

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
queries = [
    'assets/image_0.jpg',
]
documents = [
    'assets/image_1.jpg',
    'assets/image_2.jpg',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 512] [2, 512]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.0571, -0.0265]])

Evaluation

Metrics

Information Retrieval

Dataset: unsplash-dev
Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.63
cosine_accuracy@3	0.83
cosine_accuracy@5	0.89
cosine_accuracy@10	0.91
cosine_precision@1	0.63
cosine_precision@3	0.2767
cosine_precision@5	0.178
cosine_precision@10	0.091
cosine_recall@1	0.63
cosine_recall@3	0.83
cosine_recall@5	0.89
cosine_recall@10	0.91
cosine_ndcg@10	0.7791
cosine_mrr@10	0.7358
cosine_map@100	0.7392

Training Details

Training Dataset

unsplash-lite

Dataset: unsplash-lite at 3afcfc7
Size: 24,896 training samples
Columns: image and caption
Approximate statistics based on the first 100 samples:
image

caption
type

image

string

modality

image

text
details
min: 640x340 px
mean: 640x679 px
max: 640x1138 px
min: 2 tokens
mean: 25.48 tokens
max: 73 tokens

Samples:

image	caption
	`lighthouse, beach, moody, outdoors, water, wave, beacon, ocean, sea, stormy, crashing, porthcawl, sea waves, tower, nature, swell, building, architecture, surf`
	`experimental, person, human, bubble`
	`nature, outdoors, night, aurora`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false,
    "directions": [
        "query_to_doc"
    ],
    "partition_mode": "joint",
    "hardness_mode": null,
    "hardness_strength": 0.0
}

Evaluation Dataset

unsplash-lite

Dataset: unsplash-lite at 3afcfc7
Size: 100 evaluation samples
Columns: image and caption
Approximate statistics based on the first 100 samples:
image

caption
type

image

string

modality

image

text
details
min: 640x301 px
mean: 640x687 px
max: 640x1138 px
min: 2 tokens
mean: 28.81 tokens
max: 70 tokens

Samples:

image	caption
	`windbreaker, style, person, female, clothing, fur coat, apparel, raincoat, human`
	`nature, backgrounds, outdoors, wallpapers, fog`
	`waves, texture, wind wave, cloudy, water housing, blue, barrel, experimental, california, swimming, southern california, bubble, canon, wave, surf, sky, reflection, surfing`

Loss: MultipleNegativesRankingLoss with these parameters:

{
    "scale": 20.0,
    "similarity_fct": "cos_sim",
    "gather_across_devices": false,
    "directions": [
        "query_to_doc"
    ],
    "partition_mode": "joint",
    "hardness_mode": null,
    "hardness_strength": 0.0
}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 32
num_train_epochs: 1
learning_rate: 2e-05
warmup_steps: 0.1
bf16: True
per_device_eval_batch_size: 32
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

per_device_train_batch_size: 32
num_train_epochs: 1
max_steps: -1
learning_rate: 2e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0.1
optim: adamw_torch_fused
optim_args: None
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 1.0
label_smoothing_factor: 0.0
bf16: True
fp16: False
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: huggingface
trackio_space_id: None
trackio_bucket_id: None
trackio_static_space_id: None
per_device_eval_batch_size: 32
prediction_loss_only: True
eval_on_start: False
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: 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
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: True
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_static_graph: None
ddp_backend: None
ddp_timeout: 1800
fsdp: []
fsdp_config: {‘min_num_params’: 0, ‘xla’: False, ‘xla_fsdp_v2’: False, ‘xla_fsdp_grad_ckpt’: False}
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: None
local_rank: -1
prompts: None
batch_sampler: no_duplicates
multi_dataset_batch_sampler: proportional
router_mapping: {}
learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	unsplash-dev_cosine_ndcg@10
-1	-1	-	-	0.7142
0.0501	39	0.9333	-	-
0.1003	78	0.5194	0.3630	0.7751
0.1504	117	0.6026	-	-
0.2005	156	0.5822	0.3062	0.8042
0.2506	195	0.5841	-	-
0.3008	234	0.5302	0.2984	0.7983
0.3509	273	0.5120	-	-
0.4010	312	0.5243	0.3266	0.7791

Training Time

Training: 12.0 minutes

Framework Versions

Python: 3.11.6
Sentence Transformers: 5.6.0.dev0
Transformers: 5.8.0.dev0
PyTorch: 2.10.0+cu128
Accelerate: 1.13.0.dev0
Datasets: 4.8.4
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{oord2019representationlearningcontrastivepredictive,
      title={Representation Learning with Contrastive Predictive Coding},
      author={Aaron van den Oord and Yazhe Li and Oriol Vinyals},
      year={2019},
      eprint={1807.03748},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/1807.03748},
}