HassanCS/TCRa_HLA_peptide_esm2_t6_8M_UR50D_up_to_epoch_2

SentenceTransformer based on facebook/esm2_t6_8M_UR50D

This is a sentence-transformers model finetuned from facebook/esm2_t6_8M_UR50D. It maps sentences & paragraphs to a 320-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: facebook/esm2_t6_8M_UR50D
• Maximum Sequence Length: 1026 tokens
• Output Dimensionality: 320 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': 1026, 'do_lower_case': False}) with Transformer model: EsmModel 
  (1): Pooling({'word_embedding_dimension': 320, '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})
)

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("HassanCS/TCRa_HLA_peptide_esm2_t6_8M_UR50D_up_to_epoch_2")
# Run inference
sentences = [
    'G E S V G L H L P T L S V Q E G D N S I I N C A Y S N S A S D Y F I W Y K Q E S G K G P Q F I I D I R S N M D K R Q G Q R V T V L L N K T V K H L S L Q I A A T Q P G D S A V Y F C C A E I W D Y G Q N F V F F G P G T R L S V L P Y',
    'R K E V E Q D P G P F N V P E G A T V A F N C T Y S N S A S Q S F F W Y R Q D C R K E P K L L M S V Y S S G N E D G R F T A Q L N R A S Q Y I S L L I R D S K L S D S A T Y L C C V V I K A A G N K L T F F G G G T R V L V K P N',
    'G Q N I D Q P T E M T A T E G A I V Q I N C T Y Q T S G F N G L F W Y Q Q H A G E A P T F L S Y N V L D G L E E K G R F S S F L S R S K G Y S Y L L L K E L Q M K D S A S Y L C A V R E G G G A D G L T F G K G T H L I I Q P Y',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 320]

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

Evaluation

Metrics

Semantic Similarity

• Dataset: all-dev
• Evaluated with EmbeddingSimilarityEvaluator

Metric	Value
pearson_cosine	0.8217
spearman_cosine	0.8623

Training Details

Training Dataset

Unnamed Dataset

• Size: 528,048 training samples
• Columns: sentence1, sentence2, and score
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	score
  type	string	string	float
  details	min: 108 tokens mean: 116.06 tokens max: 126 tokens	min: 107 tokens mean: 116.14 tokens max: 125 tokens	min: 0.0 mean: 0.37 max: 0.97

• Samples:

  sentence1	sentence2	score
  A Q S V S Q H N H H V I L S E A A S L E L G C N Y S Y G G T V N L F W Y V Q Y P G Q H L Q L L L K Y F S G D P L V K G I K G F E A E F I K S K F S F N L R K P S V Q W S D T A E Y F C A V N A R R N T P L V F G K G T R L S V I A N	A Q S V S Q H N H H V I L S E A A S L E L G C N Y S Y G G T V N L F W Y V Q Y P G Q H L Q L L L K Y F S G D P L V K G I K G F E A E F I K S K F S F N L R K P S V Q W S D T A E Y F C A V T S G R G S Q G N L I F G K G T K L S V K P N	0.05165289256198347
  K Q E V T Q I P A A L S V P E G E N L V L N C S F T D S A I Y N L Q W F R Q D P G K G L T S L L L I Q S S Q R E Q T S G R L N A S L D K S S G R S T L Y I A A S Q P G D S A T Y L C C A V N S V S G A G S Y Q L T F F G K G T K L S V I P N	G E N V E Q H P S T L S V Q E G D S A V I K C T Y S D S A S N Y F P W Y K Q E L G K G P Q L I I D I R S N V G E K K D Q R I A V T L N K T A K H F S L H I T E T Q P E D S A V Y F C C A A N N Q G G K L I F F G Q G T E L S V K P N	0.04132231404958678
  K Q E V T Q I P A A L S V P E G E N L V L N C S F T D S A I Y N L Q W F R Q D P G K G L T S L L L I Q S S Q R E Q T S G R L N A S L D K S S G R S T L Y I A A S Q P G D S A T Y L C C A V A G G T S Y G K L T F F G Q G T I L T V H P N	K Q E V T Q I P A A L S V P E G E N L V L N C S F T D S A I Y N L Q W F R Q D P G K G L T S L L L I Q S S Q R E Q T S G R L N A S L D K S S G R S T L Y I A A S Q P G D S A T Y L C C A V N S P G S G A G S Y Q L T F F G K G T K L S V I P N	0.018595041322314043

• Loss: CoSENTLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

• Size: 58,673 evaluation samples
• Columns: sentence1, sentence2, and score
• Approximate statistics based on the first 1000 samples:

  	sentence1	sentence2	score
  type	string	string	float
  details	min: 107 tokens mean: 116.01 tokens max: 124 tokens	min: 106 tokens mean: 116.0 tokens max: 126 tokens	min: 0.0 mean: 0.39 max: 0.97

• Samples:

  sentence1	sentence2	score
  A Q T V T Q S Q P E M S V Q E A E T V T L S C T Y D T S E S N Y Y L F W Y K Q P P S R Q M I L V I R Q E A Y K Q Q N A T E N R F S V N F Q K A A K S F S L K I S D S Q L G D T A M Y F C A L W S G G G A D G L T F G K G T H L I I Q P Y	S Q Q G E E D P Q A L S I Q E G E N A T M N C S Y K T S I N N L Q W Y R Q N S G R G L V H L I L I R S N E R E K H S G R L R V T L D T S K K S S S L L I T A S R A A D T A S Y F C A R S R N K Q G G I F F F G Q G T E L S V K P N	0.04132231404958678
  Q K E V E Q N S G P L S V P E G A I A S L N C T Y S D R G S Q S F F W Y R Q Y S G K S P E L I M F I Y S N G D K E D G R F T A Q L N K A S Q Y V S L L I R D S Q P S D S A T Y L C C A V T T Q G G S E K L V F F G K G T K L T V N P Y	G E D V E Q S L F L S V R E G D S S V I N C T Y T D S S S T Y L Y W Y K Q E P G A G L Q L L T Y I F S N M D M K Q D Q R L T V L L N K K D K H L S L R I A D T Q T G D S A I Y F C C A E D K D A R L M F F G D G T Q L V V K P N	0.018595041322314043
  G E N V E Q H P S T L S V Q E G D S A V I K C T Y S D S A S N Y F P W Y K Q E L G K G P Q L I I D I R S N V G E K K D Q R I A V T L N K T A K H F S L H I T E T Q P E D S A V Y F C C A A S I G Q G G K L I F F G Q G T E L S V K P N	G E N V E Q H P S T L S V Q E G D S A V I K C T Y S D S A S N Y F P W Y K Q E L G K G P Q L I I D I R S N V G E K K D Q R I A V T L N K T A K H F S L H I T E T Q P E D S A V Y F C C A A S N P G G G N K L T F F G T G T Q L K V E L N	0.8347107438016529

• Loss: CoSENTLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• learning_rate: 0.001
• weight_decay: 0.0001
• num_train_epochs: 2
• fp16: True
• load_best_model_at_end: True

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: epoch
• prediction_loss_only: True
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• 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: 0.001
• weight_decay: 0.0001
• 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: True
• 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
• 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: proportional

Training Logs

Epoch	Step	Training Loss	Validation Loss	all-dev_spearman_cosine
0.0485	100	10.3293	-	-
0.0969	200	10.3109	-	-
0.1454	300	10.2921	-	-
0.1939	400	10.2746	-	-
0.2424	500	10.2654	-	-
0.2908	600	10.2522	-	-
0.3393	700	10.2391	-	-
0.3878	800	10.2306	-	-
0.4363	900	10.2105	-	-
0.4847	1000	10.1984	-	-
0.5332	1100	10.1897	-	-
0.5817	1200	10.1767	-	-
0.6302	1300	10.1702	-	-
0.6786	1400	10.1586	-	-
0.7271	1500	10.145	-	-
0.7756	1600	10.1227	-	-
0.8240	1700	10.1221	-	-
0.8725	1800	10.1022	-	-
0.9210	1900	10.0882	-	-
0.9695	2000	10.0733	-	-
1.0	2063	-	10.0591	0.7452
1.0179	2100	10.0549	-	-
1.0664	2200	10.0362	-	-
1.1149	2300	10.0149	-	-
1.1634	2400	10.0047	-	-
1.2118	2500	9.9963	-	-
1.2603	2600	9.9767	-	-
1.3088	2700	9.97	-	-
1.3572	2800	9.9527	-	-
1.4057	2900	9.9496	-	-
1.4542	3000	9.9261	-	-
1.5027	3100	9.9258	-	-
1.5511	3200	9.9106	-	-
1.5996	3300	9.9015	-	-
1.6481	3400	9.8807	-	-
1.6966	3500	9.8732	-	-
1.7450	3600	9.8636	-	-
1.7935	3700	9.8633	-	-
1.8420	3800	9.8469	-	-
1.8905	3900	9.8462	-	-
1.9389	4000	9.8301	-	-
1.9874	4100	9.8251	-	-
2.0	4126	-	9.8359	0.8623

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.11.13
• Sentence Transformers: 4.1.0
• Transformers: 4.53.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.9.0
• Datasets: 4.4.1
• 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",
}

CoSENTLoss

@online{kexuefm-8847,
    title={CoSENT: A more efficient sentence vector scheme than Sentence-BERT},
    author={Su Jianlin},
    year={2022},
    month={Jan},
    url={https://kexue.fm/archives/8847},
}