yosriku/gemma300m-specialization-v1

SentenceTransformer

This model was finetuned with Unsloth.

based on unsloth/embeddinggemma-300m

This is a sentence-transformers model finetuned from unsloth/embeddinggemma-300m. 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: unsloth/embeddinggemma-300m
• Maximum Sequence Length: 1024 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': 1024, 'do_lower_case': False, 'architecture': 'PeftModelForFeatureExtraction'})
  (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): Dense({'in_features': 768, 'out_features': 3072, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (3): Dense({'in_features': 3072, 'out_features': 768, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity'})
  (4): 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
queries = [
    "insomnia akut",
]
documents = [
    'Psychiatry',
    'Dermatology (Skin)',
    'General Surgery',
]
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings.shape)
# [1, 768] [3, 768]

# Get the similarity scores for the embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[0.6914, 0.5889, 0.4946]], dtype=torch.float16)

Evaluation

Metrics

Information Retrieval

• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.0465
cosine_accuracy@3	0.1395
cosine_accuracy@5	0.186
cosine_accuracy@10	0.2791
cosine_precision@1	0.0465
cosine_precision@3	0.0465
cosine_precision@5	0.0372
cosine_precision@10	0.0279
cosine_recall@1	0.0465
cosine_recall@3	0.1395
cosine_recall@5	0.186
cosine_recall@10	0.2791
cosine_ndcg@10	0.1513
cosine_mrr@10	0.112
cosine_map@100	0.1278

Training Details

Training Dataset

Unnamed Dataset

• Size: 151 training samples
• Columns: query and expected_target
• Approximate statistics based on the first 151 samples:

  	query	expected_target
  type	string	string
  details	min: 3 tokens mean: 9.07 tokens max: 44 tokens	min: 4 tokens mean: 6.17 tokens max: 11 tokens

• Samples:

  query	expected_target
  nyeri tangan sangat lama, spesialis ortopedik di siloam denpasar	Orthopaedics (Bone)
  mau cek jantung soalnya sering deg-degan	Cardiology (Heart)
  jadwal dokter bedah di Siloam Semanggi kapan ya?	General Surgery

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Evaluation Dataset

Unnamed Dataset

• Size: 19 evaluation samples
• Columns: query and expected_target
• Approximate statistics based on the first 19 samples:

  	query	expected_target
  type	string	string
  details	min: 4 tokens mean: 7.0 tokens max: 14 tokens	min: 4 tokens mean: 6.11 tokens max: 8 tokens

• Samples:

  query	expected_target
  anak saya demam tinggi	Pediatrics
  interventional cardiology	Cardiology (Heart)
  tulang	Orthopaedics (Bone)

• Loss: MultipleNegativesRankingLoss with these parameters:

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

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 64
• per_device_eval_batch_size: 64
• gradient_accumulation_steps: 2
• learning_rate: 2e-05
• num_train_epochs: 10
• max_steps: 240
• warmup_ratio: 0.03
• prompts: {'query': 'task: search result | query: {query}', 'expected_target': 'title: none | text: {expected_target}'}
• batch_sampler: no_duplicates

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: 2
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 2e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 10
• max_steps: 240
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.03
• 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}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• 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: {'query': 'task: search result | query: {query}', 'expected_target': 'title: none | text: {expected_target}'}
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	cosine_ndcg@10
-1	-1	-	-	0.1513
2.6667	5	2.4	1.5137	-
5.0	10	1.7563	1.2100	-
7.6667	15	1.1973	1.1016	-
10.0	20	0.6759	1.0469	-
12.6667	25	0.4918	0.9702	-
15.0	30	0.194	1.0283	-
17.6667	35	0.061	1.1143	-
20.0	40	0.0436	1.1846	-
22.6667	45	0.0216	1.2246	-
25.0	50	0.0068	0.9980	-
27.6667	55	0.0204	0.9604	-
30.0	60	0.0098	1.0459	-
32.6667	65	0.0043	1.0967	-
35.0	70	0.004	1.0381	-
37.6667	75	0.0029	1.2236	-
40.0	80	0.0062	1.2285	-
42.6667	85	0.0021	1.3682	-
45.0	90	0.0007	1.2969	-
47.6667	95	0.0016	1.3076	-
50.0	100	0.0004	1.4443	-
52.6667	105	0.0006	1.2227	-
55.0	110	0.0007	1.2383	-
57.6667	115	0.0003	1.6729	-
60.0	120	0.0004	1.6797	-
62.6667	125	0.0005	1.6914	-
65.0	130	0.0004	1.6885	-
67.6667	135	0.0002	1.5742	-
70.0	140	0.0001	1.3975	-
72.6667	145	0.0002	1.2949	-
75.0	150	0.0002	1.2549	-
77.6667	155	0.0003	1.2939	-
80.0	160	0.0002	1.3262	-
82.6667	165	0.0002	1.3604	-
85.0	170	0.0002	1.3916	-
87.6667	175	0.0002	1.4180	-
90.0	180	0.0002	1.4395	-
92.6667	185	0.0002	1.4531	-
95.0	190	0.0002	1.4922	-
97.6667	195	0.0002	1.5186	-
100.0	200	0.0001	1.5420	-
102.6667	205	0.0002	1.5576	-
105.0	210	0.0001	1.5713	-
107.6667	215	0.0002	1.5791	-
110.0	220	0.0001	1.5869	-
112.6667	225	0.0002	1.5908	-
115.0	230	0.0001	1.5928	-
117.6667	235	0.0002	1.5947	-
120.0	240	0.0001	1.5947	-

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.2.2
• Transformers: 4.56.2
• PyTorch: 2.9.0+cu126
• Accelerate: 1.12.0
• Datasets: 4.0.0
• 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{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}
}