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.

python

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("pj-mathematician/JobGTE-multilingual-base-v2")
# Run inference
sentences = [
    'Volksvertreter',
    'Parlamentarier',
    'Oberbürgermeister',
]
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	0.6667	0.1243	0.2956	0.6796	0.7088	0.6485	0.7667
cosine_accuracy@20	0.9905	1.0	0.9754	0.9806	0.9553	0.9324	0.9843
cosine_accuracy@50	0.9905	1.0	0.9852	0.9903	0.9802	0.9683	0.9932
cosine_accuracy@100	0.9905	1.0	0.9901	0.9903	0.9901	0.9849	0.9958
cosine_accuracy@150	0.9905	1.0	0.9901	0.9903	0.9938	0.9886	0.9974
cosine_accuracy@200	0.9905	1.0	0.9901	0.9903	0.9958	0.9938	0.9979
cosine_precision@1	0.6667	0.1243	0.2956	0.6796	0.7088	0.6485	0.7667
cosine_precision@20	0.5148	0.5759	0.5103	0.4883	0.1216	0.1209	0.1387
cosine_precision@50	0.32	0.3923	0.3694	0.2963	0.0512	0.0514	0.0581
cosine_precision@100	0.1905	0.2566	0.2397	0.1788	0.0261	0.0265	0.0296
cosine_precision@150	0.1362	0.1928	0.1808	0.1278	0.0175	0.0179	0.0199
cosine_precision@200	0.1054	0.1528	0.1462	0.0999	0.0132	0.0135	0.0149
cosine_recall@1	0.0685	0.0036	0.0111	0.0693	0.2738	0.2436	0.2569
cosine_recall@20	0.5491	0.3853	0.3208	0.5251	0.899	0.8787	0.9157
cosine_recall@50	0.7554	0.566	0.5042	0.7083	0.9459	0.932	0.9583
cosine_recall@100	0.8503	0.6899	0.6173	0.8169	0.9651	0.9596	0.9765
cosine_recall@150	0.8995	0.754	0.6848	0.8613	0.9732	0.9718	0.9834
cosine_recall@200	0.9208	0.7858	0.7253	0.8898	0.9791	0.98	0.9865
cosine_ndcg@1	0.6667	0.1243	0.2956	0.6796	0.7088	0.6485	0.7667
cosine_ndcg@20	0.6952	0.6169	0.5378	0.6681	0.7815	0.7448	0.8002
cosine_ndcg@50	0.723	0.5914	0.5288	0.6857	0.7944	0.7595	0.8125
cosine_ndcg@100	0.7733	0.6235	0.5552	0.7379	0.7986	0.7657	0.8167
cosine_ndcg@150	0.7947	0.6557	0.5888	0.7577	0.8001	0.7682	0.8181
cosine_ndcg@200	0.8039	0.6717	0.6092	0.7697	0.8012	0.7696	0.8187
cosine_mrr@1	0.6667	0.1243	0.2956	0.6796	0.7088	0.6485	0.7667
cosine_mrr@20	0.8183	0.5581	0.5165	0.8159	0.7804	0.7324	0.8422
cosine_mrr@50	0.8183	0.5581	0.5168	0.8163	0.7813	0.7335	0.8425
cosine_mrr@100	0.8183	0.5581	0.5168	0.8163	0.7814	0.7338	0.8425
cosine_mrr@150	0.8183	0.5581	0.5168	0.8163	0.7814	0.7338	0.8425
cosine_mrr@200	0.8183	0.5581	0.5168	0.8163	0.7814	0.7338	0.8426
cosine_map@1	0.6667	0.1243	0.2956	0.6796	0.7088	0.6485	0.7667
cosine_map@20	0.5566	0.4841	0.3984	0.5222	0.7071	0.6646	0.7007
cosine_map@50	0.5534	0.4304	0.3603	0.5083	0.7107	0.6684	0.7046
cosine_map@100	0.5852	0.4374	0.3632	0.5372	0.7113	0.6693	0.7054
cosine_map@150	0.5943	0.4527	0.3782	0.5454	0.7114	0.6695	0.7055
cosine_map@200	0.5976	0.4593	0.3863	0.5495	0.7115	0.6696	0.7056
cosine_map@500	0.6016	0.472	0.3992	0.5542	0.7116	0.6697	0.7057

Training Details

Training Datasets

full_en

Dataset: full_en
Size: 28,880 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 3 tokens
mean: 5.68 tokens
max: 11 tokens
min: 3 tokens
mean: 5.76 tokens
max: 12 tokens
Samples:
anchor positive
air commodore flight lieutenant
command and control officer flight officer
air commodore command and control officer
Loss: GISTEmbedLoss with these parameters:
json
1{'guide': SentenceTransformer( 2 (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 3 (1): Pooling({'word_embedding_dimension': 384, '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}) 4 (2): Normalize() 5), 'temperature': 0.01, 'margin_strategy': 'absolute', 'margin': 0.0}

full_de

Dataset: full_de
Size: 23,023 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 3 tokens
mean: 7.99 tokens
max: 30 tokens
min: 3 tokens
mean: 8.19 tokens
max: 30 tokens
Samples:
anchor positive
Staffelkommandantin Kommodore
Luftwaffenoffizierin Luftwaffenoffizier/Luftwaffenoffizierin
Staffelkommandantin Luftwaffenoffizierin
Loss: GISTEmbedLoss with these parameters:
json
1{'guide': SentenceTransformer( 2 (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 3 (1): Pooling({'word_embedding_dimension': 384, '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}) 4 (2): Normalize() 5), 'temperature': 0.01, 'margin_strategy': 'absolute', 'margin': 0.0}

full_es

Dataset: full_es
Size: 20,724 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 3 tokens
mean: 9.13 tokens
max: 32 tokens
min: 3 tokens
mean: 8.84 tokens
max: 32 tokens
Samples:
anchor positive
jefe de escuadrón instructor
comandante de aeronave instructor de simulador
instructor oficial del Ejército del Aire
Loss: GISTEmbedLoss with these parameters:
json
1{'guide': SentenceTransformer( 2 (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 3 (1): Pooling({'word_embedding_dimension': 384, '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}) 4 (2): Normalize() 5), 'temperature': 0.01, 'margin_strategy': 'absolute', 'margin': 0.0}

full_zh

Dataset: full_zh
Size: 30,401 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 5 tokens
mean: 7.15 tokens
max: 14 tokens
min: 5 tokens
mean: 7.46 tokens
max: 21 tokens
Samples:
anchor positive
技术总监 技术和运营总监
技术总监 技术主管
技术总监 技术艺术总监
Loss: GISTEmbedLoss with these parameters:
json
1{'guide': SentenceTransformer( 2 (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 3 (1): Pooling({'word_embedding_dimension': 384, '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}) 4 (2): Normalize() 5), 'temperature': 0.01, 'margin_strategy': 'absolute', 'margin': 0.0}

mix

Dataset: mix
Size: 21,760 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 2 tokens
mean: 6.71 tokens
max: 19 tokens
min: 2 tokens
mean: 7.69 tokens
max: 19 tokens
Samples:
anchor positive
technical manager Technischer Direktor für Bühne, Film und Fernsehen
head of technical directora técnica
head of technical department 技术艺术总监
Loss: GISTEmbedLoss with these parameters:
json
1{'guide': SentenceTransformer( 2 (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: BertModel 3 (1): Pooling({'word_embedding_dimension': 384, '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}) 4 (2): Normalize() 5), 'temperature': 0.01, 'margin_strategy': 'absolute', 'margin': 0.0}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 128
per_device_eval_batch_size: 128
gradient_accumulation_steps: 2
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: 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: 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	-	0.7447	0.6125	0.5378	0.7240	0.7029	0.6345	0.7437
0.0082	1	4.3088	-	-	-	-	-	-	-
0.8230	100	1.9026	-	-	-	-	-	-	-
1.6502	200	0.9336	0.8024	0.6703	0.6109	0.7695	0.7914	0.7594	0.8136
2.4774	300	0.161	-	-	-	-	-	-	-
3.3045	400	0.1398	0.8039	0.6717	0.6092	0.7697	0.8012	0.7696	0.8187

Framework Versions

Python: 3.11.11
Sentence Transformers: 4.1.0
Transformers: 4.51.2
PyTorch: 2.6.0+cu124
Accelerate: 1.6.0
Datasets: 3.5.0
Tokenizers: 0.21.1

Citation

BibTeX

Sentence Transformers

bibtex

@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",
}

GISTEmbedLoss

bibtex

@misc{solatorio2024gistembed,
    title={GISTEmbed: Guided In-sample Selection of Training Negatives for Text Embedding Fine-tuning},
    author={Aivin V. Solatorio},
    year={2024},
    eprint={2402.16829},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

	anchor	positive
type	string	string
details	min: 3 tokens mean: 5.68 tokens max: 11 tokens	min: 3 tokens mean: 5.76 tokens max: 12 tokens

anchor	positive
`air commodore`	`flight lieutenant`
`command and control officer`	`flight officer`
`air commodore`	`command and control officer`

anchor	positive
`Staffelkommandantin`	`Kommodore`
`Luftwaffenoffizierin`	`Luftwaffenoffizier/Luftwaffenoffizierin`
`Staffelkommandantin`	`Luftwaffenoffizierin`

anchor	positive
`jefe de escuadrón`	`instructor`
`comandante de aeronave`	`instructor de simulador`
`instructor`	`oficial del Ejército del Aire`

anchor	positive
`技术总监`	`技术和运营总监`
`技术总监`	`技术主管`
`技术总监`	`技术艺术总监`

anchor	positive
`technical manager`	`Technischer Direktor für Bühne, Film und Fernsehen`
`head of technical`	`directora técnica`
`head of technical department`	`技术艺术总监`