This is a Cross Encoder model finetuned from cross-encoder/mmarco-mMiniLMv2-L12-H384-v1 on the vodex-turkish-reranker-triplets dataset using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

⚠️ Domain-Specific Warning

This model was fine-tuned on Turkish data specifically sourced from the telecommunications domain.
While it performs well on telecom-related tasks such as mobile services, billing, campaigns, and subscription details, it may not generalize well to other domains.
Please assess its performance carefully before applying it outside of telecommunications use cases.

Model Description

Model Type: Cross Encoder
Base model: cross-encoder/mmarco-mMiniLMv2-L12-H384-v1
Maximum Sequence Length: 512 tokens
Number of Output Labels: 1 label
Training Dataset:
- vodex-turkish-reranker-triplets
Language: tr
License: apache-2.0

Model Sources

Documentation: Sentence Transformers Documentation
Documentation: Cross Encoder Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Cross Encoders on Hugging Face

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 CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("seroe/mmarco-mMiniLMv2-L12-H384-v1-turkish-reranker-triplet")
# Get scores for pairs of texts
pairs = [
    ['Faturasız tarifelerde yurtdışı mesaj ücretleri ne kadardır?', 'Yurtdışına gönderilen mesajlar için ücret 75 kuruş olarak belirlenmiştir.'],
    ['Kampanya süresince internet hızı nasıl değişebilir?', 'Kampanya süresince, limit ve altyapının desteklediği azami internet hızına kadar internet hızı yükseltilebilir.'],
    ["Vodafone'un tarifelerinde KDV ve ÖİV dahil midir?", "Vodafone'un tarifelerinde belirtilen ücretlere KDV ve ÖİV dahildir."],
    ['Taahhüt süresi dolmadan internet hizmeti iptal edilirse ne olur?', 'Eğer taahhüt süresi bitmeden internet hizmeti iptal edilirse, aboneye sunulan D-Smart hizmeti de iptal edilecektir.'],
    ['Aylık 15 GB ek paketini nereden satın alabilirim?', 'Bu ek paketi almak için hangi kanalları kullanabilirim?'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'Faturasız tarifelerde yurtdışı mesaj ücretleri ne kadardır?',
    [
        'Yurtdışına gönderilen mesajlar için ücret 75 kuruş olarak belirlenmiştir.',
        'Kampanya süresince, limit ve altyapının desteklediği azami internet hızına kadar internet hızı yükseltilebilir.',
        "Vodafone'un tarifelerinde belirtilen ücretlere KDV ve ÖİV dahildir.",
        'Eğer taahhüt süresi bitmeden internet hizmeti iptal edilirse, aboneye sunulan D-Smart hizmeti de iptal edilecektir.',
        'Bu ek paketi almak için hangi kanalları kullanabilirim?',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

Datasets: val-hard and test-hard
Evaluated with CrossEncoderRerankingEvaluator with these parameters:
json
1{ 2 "at_k": 10, 3 "always_rerank_positives": true 4}

Metric	val-hard	test-hard
map	0.6093 (-0.0246)	0.6085 (-0.0178)
mrr@10	0.6085 (-0.0254)	0.6077 (-0.0186)
ndcg@10	0.6994 (+0.0641)	0.6987 (+0.0705)

Training Details

Training Dataset

vodex-turkish-reranker-triplets

Dataset: vodex-turkish-reranker-triplets at ca7d206
Size: 89,964 training samples
Columns: query, positive, and negative

Approximate statistics based on the first 1000 samples:

	query	positive	negative
type	string	string	string
details	min: 20 characters mean: 57.83 characters max: 112 characters	min: 35 characters mean: 92.19 characters max: 221 characters	min: 31 characters mean: 78.41 characters max: 143 characters

Samples:

query	positive	negative
`Faturasız tarifelerde yurtdışı mesaj ücretleri ne kadardır?`	`Yurtdışına gönderilen mesajlar için ücret 75 kuruş olarak belirlenmiştir.`	`Faturasız tarifelerde yurtdışı mesaj ücretleri 10 kuruş olarak uygulanmaktadır.`
`Kampanya süresince internet hızı nasıl değişebilir?`	`Kampanya süresince, limit ve altyapının desteklediği azami internet hızına kadar internet hızı yükseltilebilir.`	`Kampanya süresince internet hızı sabit kalır ve değişiklik yapılamaz.`
`Vodafone'un tarifelerinde KDV ve ÖİV dahil midir?`	`Vodafone'un tarifelerinde belirtilen ücretlere KDV ve ÖİV dahildir.`	`Vodafone tarifelerinde KDV ve ÖİV, abonelerin talep etmesi durumunda eklenmektedir.`

Loss: CachedMultipleNegativesRankingLoss with these parameters:
json
1{ 2 "scale": 10.0, 3 "num_negatives": 4, 4 "activation_fn": "torch.nn.modules.activation.Sigmoid", 5 "mini_batch_size": 32 6}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: steps
per_device_train_batch_size: 1024
per_device_eval_batch_size: 1024
learning_rate: 5e-07
weight_decay: 0.1
max_grad_norm: 0.8
warmup_ratio: 0.25
bf16: True
dataloader_num_workers: 8
load_best_model_at_end: True
group_by_length: 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: 1024
per_device_eval_batch_size: 1024
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-07
weight_decay: 0.1
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
max_grad_norm: 0.8
num_train_epochs: 3
max_steps: -1
lr_scheduler_type: linear
lr_scheduler_kwargs: {}
warmup_ratio: 0.25
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: True
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: 8
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: True
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: False
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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss	val-hard_ndcg@10	test-hard_ndcg@10
1.125	100	1.3041	0.7093 (+0.0740)	0.7065 (+0.0783)
2.25	200	0.9232	0.6994 (+0.0641)	0.6987 (+0.0705)

Framework Versions

Python: 3.10.12
Sentence Transformers: 4.2.0.dev0
Transformers: 4.46.3
PyTorch: 2.5.1+cu124
Accelerate: 1.6.0
Datasets: 3.6.0
Tokenizers: 0.20.3

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",
}