TechWolf/JobBERT-v3

SentenceTransformer based on sentence-transformers/paraphrase-multilingual-mpnet-base-v2

This is a sentence-transformers model specifically trained for job title matching and similarity. It's finetuned from sentence-transformers/paraphrase-multilingual-mpnet-base-v2 on a large dataset of job titles and their associated skills/requirements across multiple languages. The model maps English, Spanish, German and Chinese job titles and descriptions to a 1024-dimensional dense vector space and can be used for semantic job title matching, job similarity search, and related HR/recruitment tasks.

The model was presented in the paper Multilingual JobBERT for Cross-Lingual Job Title Matching.

Model Details

Model Description

• —Model Type: Sentence Transformer
• —Base model: sentence-transformers/paraphrase-multilingual-mpnet-base-v2 <!-- at revision 84fccfe766bcfd679e39efefe4ebf45af190ad2d -->
• —Maximum Sequence Length: 64 tokens
• —Output Dimensionality: 1024 dimensions
• —Similarity Function: Cosine Similarity
• —Training Dataset: 4 x 5.2M high-quality job title - skills pairs in English, Spanish, German and Chinese

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': 64, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (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): Asym(
    (anchor-0): Dense({'in_features': 768, 'out_features': 1024, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
    (positive-0): Dense({'in_features': 768, 'out_features': 1024, 'bias': True, 'activation_function': 'torch.nn.modules.activation.Tanh'})
  )
)

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load and use the model with the following code:

import torch
import numpy as np
from tqdm.auto import tqdm
from sentence_transformers import SentenceTransformer
from sentence_transformers.util import batch_to_device, cos_sim

# Load the model
model = SentenceTransformer("TechWolf/JobBERT-v3")

def encode_batch(jobbert_model, texts):
    features = jobbert_model.tokenize(texts)
    features = batch_to_device(features, jobbert_model.device)
    features["text_keys"] = ["anchor"]
    with torch.no_grad():
        out_features = jobbert_model.forward(features)
    return out_features["sentence_embedding"].cpu().numpy()

def encode(jobbert_model, texts, batch_size: int = 8):
    # Sort texts by length and keep track of original indices
    sorted_indices = np.argsort([len(text) for text in texts])
    sorted_texts = [texts[i] for i in sorted_indices]
    
    embeddings = []
    
    # Encode in batches
    for i in tqdm(range(0, len(sorted_texts), batch_size)):
        batch = sorted_texts[i:i+batch_size]
        embeddings.append(encode_batch(jobbert_model, batch))
    
    # Concatenate embeddings and reorder to original indices
    sorted_embeddings = np.concatenate(embeddings)
    original_order = np.argsort(sorted_indices)
    return sorted_embeddings[original_order]

# Example usage
job_titles = [
    'Software Engineer',
    '高级软件开发人员',  # senior software developer
    'Produktmanager',  # product manager
    'Científica de datos'  # data scientist
]

# Get embeddings
embeddings = encode(model, job_titles)

# Calculate cosine similarity matrix
similarities = cos_sim(embeddings, embeddings)
print(similarities)

The output will be a similarity matrix where each value represents the cosine similarity between two job titles:

tensor([[1.0000, 0.8087, 0.4673, 0.5669],
        [0.8087, 1.0000, 0.4428, 0.4968],
        [0.4673, 0.4428, 1.0000, 0.4292],
        [0.5669, 0.4968, 0.4292, 1.0000]])

<!--

Direct Usage (Transformers)

<details><summary>Click to see the direct usage in Transformers</summary>

</details> -->

<!--

Downstream Usage (Sentence Transformers)

You can finetune this model on your own dataset.

<details><summary>Click to expand</summary>

</details> -->

<!--

Out-of-Scope Use

List how the model may foreseeably be misused and address what users ought not to do with the model. -->

<!--

Bias, Risks and Limitations

What are the known or foreseeable issues stemming from this model? You could also flag here known failure cases or weaknesses of the model. -->

<!--

Recommendations

What are recommendations with respect to the foreseeable issues? For example, filtering explicit content. -->

Training Details

Training Dataset

• —Size: 21,123,868 training samples
• —Columns: <code>anchor</code> and <code>positive</code>
• —Approximate statistics based on the first 1000 samples: | | anchor | positive | |:--------|:----------------------------------------------------------------------------------|:-----------------------------------------------------------------------------------| | type | string | string | | details | <ul><li>min: 4 tokens</li><li>mean: 10.56 tokens</li><li>max: 38 tokens</li></ul> | <ul><li>min: 19 tokens</li><li>mean: 61.08 tokens</li><li>max: 64 tokens</li></ul> |
• —Samples: | anchor | positive | |:-----------------------------------------------|:------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | <code>通信与培训专员</code> | <code>deliver online training, liaise with educational support staff, interact with an audience, construct individual learning plans, lead a team, develop corporate training programmes, learning technologies, communication, identify with the company's goals, address an audience, learning management systems, use presentation software, motivate others, provide learning support, engage with stakeholders, identify skills gaps, meet expectations of target audience, develop training programmes</code> | | <code>Associate Infrastructure Engineer</code> | <code>create solutions to problems, design user interface, cloud technologies, use databases, automate cloud tasks, keep up-to-date to computer trends, work in teams, use object-oriented programming, keep updated on innovations in various business fields, design principles, Angular, adapt to changing situations, JavaScript, Agile development, manage stable, Swift (computer programming), keep up-to-date to design industry trends, monitor technology trends, web programming, provide mentorship, advise on efficiency improvements, adapt to change, JavaScript Framework, database management systems, stimulate creative processes</code> | | <code>客户顾问/出纳</code> | <code>customer service, handle financial transactions, adapt to changing situations, have computer literacy, manage cash desk, attend to detail, provide customer guidance on product selection, perform multiple tasks at the same time, carry out financial transactions, provide membership service, manage accounts, adapt to change, identify customer's needs, solve problems</code> |
• —Loss: <code>CachedMultipleNegativesRankingLoss</code> with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "mini_batch_size": 512
  }

Training Hyperparameters

Non-Default Hyperparameters

• —overwrite_output_dir: True
• —per_device_train_batch_size: 2048
• —per_device_eval_batch_size: 2048
• —num_train_epochs: 1
• —fp16: True

All Hyperparameters

<details><summary>Click to expand</summary>

• —overwrite_output_dir: True
• —do_predict: False
• —eval_strategy: no
• —prediction_loss_only: True
• —per_device_train_batch_size: 2048
• —per_device_eval_batch_size: 2048
• —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: 5e-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: 1
• —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: False
• —ignore_data_skip: False
• —fsdp: []
• —fsdp_min_num_params: 0
• —fsdp_config: {'minnumparams': 0, 'xla': False, 'xlafsdpv2': False, 'xlafsdpgrad_ckpt': False}
• —fsdp_transformer_layer_cls_to_wrap: None
• —accelerator_config: {'splitbatches': False, 'dispatchbatches': None, 'evenbatches': True, 'useseedablesampler': True, 'nonblocking': False, 'gradientaccumulationkwargs': 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

</details>

Training Logs

Environmental Impact

Carbon emissions were measured using CodeCarbon.

• —Energy Consumed: 1.944 kWh
• —Carbon Emitted: 0.717 kg of CO2
• —Hours Used: 5.34 hours

Training Hardware

• —On Cloud: Yes
• —GPU Model: 1 x NVIDIA A100-SXM4-40GB
• —CPU Model: Intel(R) Xeon(R) CPU @ 2.20GHz
• —RAM Size: 83.48 GB

Framework Versions

• —Python: 3.10.16
• —Sentence Transformers: 4.1.0
• —Transformers: 4.48.3
• —PyTorch: 2.6.0+cu126
• —Accelerate: 1.3.0
• —Datasets: 3.5.1
• —Tokenizers: 0.21.0

Citation

BibTeX

JobBERT-v3 Paper

@misc{decorte2025multilingualjobbertcrosslingualjob,
      title={Multilingual JobBERT for Cross-Lingual Job Title Matching}, 
      author={Jens-Joris Decorte and Matthias De Lange and Jeroen Van Hautte},
      year={2025},
      eprint={2507.21609},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2507.21609}, 
}

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",
}

CachedMultipleNegativesRankingLoss

@misc{gao2021scaling,
    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
    year={2021},
    eprint={2101.06983},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

<!--

Glossary

Clearly define terms in order to be accessible across audiences. -->

<!--

Model Card Authors

Lists the people who create the model card, providing recognition and accountability for the detailed work that goes into its construction. -->

<!--

Model Card Contact

Provides a way for people who have updates to the Model Card, suggestions, or questions, to contact the Model Card authors. -->