pj-mathematician/JobGTE-multilingual-base-pruned

Job - Job matching Alibaba-NLP/gte-multilingual-base pruned

Top performing model on TalentCLEF 2025 Task A. Use it for multilingual job title matching

Model Details

Model Description

• Model Type: Sentence Transformer
• Maximum Sequence Length: 512 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': 512, 'do_lower_case': False}) with Transformer model: NewModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, '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("pj-mathematician/JobGTE-multilingual-base-pruned")
# Run inference
sentences = [
    'Entwicklerin für mobile Anwendungen',
    'Mergers-and-Acquisitions-Analyst/Mergers-and-Acquisitions-Analystin',
    'fashion design expert',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

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

Evaluation

Metrics

Information Retrieval

• Datasets: full_en, full_es, full_de, full_zh, mix_es, mix_de and mix_zh
• Evaluated with InformationRetrievalEvaluator

Metric	full_en	full_es	full_de	full_zh	mix_es	mix_de	mix_zh
cosine_accuracy@1	0.6476	0.1189	0.2956	0.6602	0.6297	0.5564	0.6086
cosine_accuracy@20	0.9714	1.0	0.9704	0.9709	0.9106	0.8866	0.9593
cosine_accuracy@50	0.9905	1.0	0.9754	0.9903	0.9496	0.9381	0.9791
cosine_accuracy@100	0.9905	1.0	0.9901	0.9903	0.973	0.9594	0.9875
cosine_accuracy@150	0.9905	1.0	0.9901	0.9903	0.9834	0.9709	0.9911
cosine_accuracy@200	0.9905	1.0	0.9901	0.9903	0.9901	0.9776	0.9937
cosine_precision@1	0.6476	0.1189	0.2956	0.6602	0.6297	0.5564	0.6086
cosine_precision@20	0.4795	0.5268	0.4291	0.4481	0.1117	0.1095	0.1266
cosine_precision@50	0.2884	0.3438	0.298	0.2713	0.0485	0.0481	0.0552
cosine_precision@100	0.173	0.219	0.1943	0.1665	0.0254	0.0253	0.0287
cosine_precision@150	0.1244	0.1658	0.1482	0.1211	0.0172	0.0173	0.0194
cosine_precision@200	0.0986	0.1333	0.1198	0.0953	0.0131	0.0131	0.0147
cosine_recall@1	0.0661	0.0036	0.0111	0.0661	0.2434	0.2093	0.2044
cosine_recall@20	0.5122	0.3541	0.2668	0.4841	0.8288	0.7989	0.8346
cosine_recall@50	0.6835	0.5098	0.4092	0.6568	0.8987	0.8741	0.9096
cosine_recall@100	0.79	0.6076	0.5098	0.7685	0.9399	0.9173	0.9476
cosine_recall@150	0.84	0.6705	0.5729	0.8278	0.9577	0.9424	0.9609
cosine_recall@200	0.8759	0.7125	0.612	0.8617	0.9695	0.9536	0.9698
cosine_ndcg@1	0.6476	0.1189	0.2956	0.6602	0.6297	0.5564	0.6086
cosine_ndcg@20	0.6468	0.5708	0.4696	0.6231	0.701	0.6541	0.6853
cosine_ndcg@50	0.658	0.5355	0.4449	0.6383	0.7201	0.6748	0.7067
cosine_ndcg@100	0.7095	0.559	0.467	0.6917	0.7291	0.6845	0.7154
cosine_ndcg@150	0.731	0.59	0.4982	0.7167	0.7326	0.6894	0.7181
cosine_ndcg@200	0.7461	0.6095	0.5165	0.7303	0.7347	0.6915	0.7198
cosine_mrr@1	0.6476	0.1189	0.2956	0.6602	0.6297	0.5564	0.6086
cosine_mrr@20	0.7902	0.5532	0.5047	0.8016	0.7037	0.6477	0.7237
cosine_mrr@50	0.791	0.5532	0.5048	0.8021	0.705	0.6494	0.7243
cosine_mrr@100	0.791	0.5532	0.505	0.8021	0.7053	0.6497	0.7245
cosine_mrr@150	0.791	0.5532	0.505	0.8021	0.7054	0.6498	0.7245
cosine_mrr@200	0.791	0.5532	0.505	0.8021	0.7055	0.6498	0.7245
cosine_map@1	0.6476	0.1189	0.2956	0.6602	0.6297	0.5564	0.6086
cosine_map@20	0.5026	0.4379	0.3366	0.475	0.6194	0.5648	0.5652
cosine_map@50	0.484	0.3739	0.2853	0.4579	0.6244	0.57	0.5716
cosine_map@100	0.5118	0.3763	0.2818	0.4848	0.6257	0.5714	0.5731
cosine_map@150	0.5202	0.3892	0.2931	0.4937	0.626	0.5719	0.5734
cosine_map@200	0.5249	0.3958	0.2988	0.4978	0.6262	0.572	0.5735
cosine_map@500	0.5304	0.4063	0.3109	0.504	0.6263	0.5723	0.5736

Training Details

Training Dataset

Unnamed Dataset

• Size: 86,648 training samples
• Columns: sentence and label
• Approximate statistics based on the first 1000 samples:

  	sentence	label
  type	string	list
  details	min: 2 tokens mean: 8.25 tokens max: 54 tokens	size: 768 elements

• Samples:

  sentence	label
  	[-0.07171934843063354, 0.03595816716551781, -0.029780959710478783, 0.006593302357941866, 0.040611181408166885, ...]
  airport environment officer	[-0.022075481712818146, 0.02999737113714218, -0.02189866080880165, 0.016531817615032196, 0.012234307825565338, ...]
  Flake操作员	[-0.04815564677119255, 0.023524893447756767, -0.01583661139011383, 0.042527906596660614, 0.03815540298819542, ...]

• Loss: MSELoss

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• gradient_accumulation_steps: 2
• learning_rate: 0.0001
• num_train_epochs: 5
• warmup_ratio: 0.05
• log_on_each_node: False
• fp16: True
• dataloader_num_workers: 4
• ddp_find_unused_parameters: True
• 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: 128
• per_device_eval_batch_size: 128
• 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: 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: 5
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.05
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: False
• 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: True
• dataloader_num_workers: 4
• 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}
• tp_size: 0
• 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: True
• 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
• 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: no_duplicates
• multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	full_en_cosine_ndcg@200	full_es_cosine_ndcg@200	full_de_cosine_ndcg@200	full_zh_cosine_ndcg@200	mix_es_cosine_ndcg@200	mix_de_cosine_ndcg@200	mix_zh_cosine_ndcg@200
-1	-1	-	0.5348	0.4311	0.3678	0.5333	0.2580	0.1924	0.2871
0.0030	1	0.0017	-	-	-	-	-	-	-
0.2959	100	0.001	-	-	-	-	-	-	-
0.5917	200	0.0005	0.6702	0.5287	0.4566	0.6809	0.5864	0.5302	0.4739
0.8876	300	0.0004	-	-	-	-	-	-	-
1.1834	400	0.0004	0.7057	0.5643	0.4790	0.7033	0.6604	0.6055	0.6003
1.4793	500	0.0004	-	-	-	-	-	-	-
1.7751	600	0.0003	0.7184	0.5783	0.4910	0.7127	0.6927	0.6416	0.6485
2.0710	700	0.0003	-	-	-	-	-	-	-
2.3669	800	0.0003	0.7307	0.5938	0.5023	0.7233	0.7125	0.6639	0.6847
2.6627	900	0.0003	-	-	-	-	-	-	-
2.9586	1000	0.0003	0.7371	0.6002	0.5085	0.7228	0.7222	0.6761	0.6998
3.2544	1100	0.0003	-	-	-	-	-	-	-
3.5503	1200	0.0003	0.7402	0.6059	0.5109	0.7279	0.7285	0.6841	0.7120
3.8462	1300	0.0003	-	-	-	-	-	-	-
4.1420	1400	0.0003	0.7449	0.6083	0.5154	0.7294	0.7333	0.6894	0.7176
4.4379	1500	0.0003	-	-	-	-	-	-	-
4.7337	1600	0.0003	0.7461	0.6095	0.5165	0.7303	0.7347	0.6915	0.7198

Framework Versions

• Python: 3.11.11
• Sentence Transformers: 4.1.0
• Transformers: 4.51.3
• PyTorch: 2.6.0+cu124
• Accelerate: 1.6.0
• Datasets: 3.5.0
• Tokenizers: 0.21.1

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",
}

MSELoss

@inproceedings{reimers-2020-multilingual-sentence-bert,
    title = "Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2020",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/2004.09813",
}