shawhin/clip-title-thumbnail-embeddings

SentenceTransformer based on sentence-transformers/clip-ViT-L-14

This is a sentence-transformers model finetuned from sentence-transformers/clip-ViT-L-14 on the yt-title-thumbnail-pairs dataset. It maps sentences & paragraphs to a None-dimensional dense vector space and can be used for semantic textual similarity, semantic search, paraphrase mining, text classification, clustering, and more.

A YouTube video and blog post walking through the model training are linked below.

Links:

• Video link
• Blog link
• Fine-tuned Model

Model Details

Model Description

• Model Type: Sentence Transformer
• Base model: sentence-transformers/clip-ViT-L-14
• Maximum Sequence Length: None tokens
• Output Dimensionality: None dimensions
• Similarity Function: Cosine Similarity
• Training Dataset:
  • yt-title-thumbnail-pairs

Model Sources

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

Full Model Architecture

SentenceTransformer(
  (0): 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("shawhin/clip-title-thumbnail-embeddings")
# Run inference
sentences = [
    'My $100,000+ Data Science Resume (what got me hired)',
    'The Mapper Algorithm | Overview & Python Example Code',
    'How to Build Data Pipelines for ML Projects (w/ Python Code)',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

Evaluation

Metrics

Triplet

• Datasets: yt-title-thumbnail-train and yt-title-thumbnail-valid
• Evaluated with TripletEvaluator

Metric	yt-title-thumbnail-train	yt-title-thumbnail-valid
cosine_accuracy	1.0	1.0

Training Details

Training Dataset

yt-title-thumbnail-pairs

• Dataset: yt-title-thumbnail-pairs at f97327c
• Size: 53 training samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 53 samples:

  	anchor	positive	negative
  type	PIL.JpegImagePlugin.JpegImageFile	string	string
  details		min: 9 tokens mean: 15.04 tokens max: 27 tokens	min: 10 tokens mean: 15.3 tokens max: 27 tokens

• Samples:

  anchor	positive	negative
  	Multimodal RAG: A Beginner-friendly Guide (with Python Code)	What Nature Can Teach Us About Business...
  	Detecting Power Laws in Real-world Data	w/ Python Code
  	I Quit My Job… Here’s How Much I Made 1 Year Later	Persistent Homology

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

yt-title-thumbnail-pairs

• Dataset: yt-title-thumbnail-pairs at f97327c
• Size: 11 evaluation samples
• Columns: anchor, positive, and negative
• Approximate statistics based on the first 11 samples:

  	anchor	positive	negative
  type	PIL.JpegImagePlugin.JpegImageFile	string	string
  details		min: 8 tokens mean: 14.27 tokens max: 21 tokens	min: 8 tokens mean: 14.36 tokens max: 19 tokens

• Samples:

  anchor	positive	negative
  	I Was Wrong About AI Consulting (what I learned)	How to Make a Data Science Portfolio With GitHub Pages (2024)
  	My $100,000+ Data Science Resume (what got me hired)	The Mapper Algorithm
  	4 Skills You Need to Be a Full-Stack Data Scientist	Fine-Tuning Text Embeddings For Domain-specific Search (w/ Python)

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• learning_rate: 0.0001
• num_train_epochs: 2

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• 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: 0.0001
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 2
• 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
• 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
• dispatch_batches: None
• split_batches: 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: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	yt-title-thumbnail-train_cosine_accuracy	yt-title-thumbnail-valid_cosine_accuracy
0	0	-	-	0.9623	1.0
0.25	1	2.0056	-	-	-
0.5	2	1.9543	-	-	-
0.75	3	1.6954	-	-	-
1.0	4	0.7505	1.4916	-	-
1.25	5	1.5534	-	-	-
1.5	6	1.2892	-	-	-
1.75	7	1.3283	-	-	-
2.0	8	0.3315	1.4990	1.0	1.0

Framework Versions

• Python: 3.12.8
• Sentence Transformers: 3.3.1
• Transformers: 4.48.0
• PyTorch: 2.3.0
• Accelerate: 1.3.0
• Datasets: 3.2.0
• Tokenizers: 0.21.0

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}
}