This is a sentence-transformers model finetuned from heydariAI/persian-embeddings. It maps sentences & paragraphs to a 1024-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: heydariAI/persian-embeddings
Maximum Sequence Length: 512 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity

Model Sources

Training Notebook: Persian Sentiment PyTorch NLP on Kaggle
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: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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})
)

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("sentence_transformers_model_id")
# Run inference
sentences = [
    'کلا در مرتبسازی ماژول تعمیرگاه خودرو بیحاشیه قابلدرک بود و مورد خاصی جلب توجه نکرد',
    'کلا در مرتبسازی ماژول تعمیرگاه خودرو بیحاشیه قابلدرک بود و مورد خاصی جلب توجه نکرد',
    'به نظر من این ماه فروشگاه پوشاک زنانه استاندارد نسخه آزمایشی نامنظم بود و خوب پیش نرفت و مورد خاصی جلب توجه نکرد',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

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

Training Notebook

The model was fine-tuned using a PyTorch-based NLP training workflow on Kaggle:

Persian Sentiment PyTorch NLP — Kaggle Notebook

The notebook contains the complete training pipeline, including:

Loading the Persian sentiment dataset from Kaggle
Reading the dataset directly from Kaggle's input directory
Preparing sentence pairs and their similarity labels
Fine-tuning heydariAI/persian-embeddings
Training with SoftmaxLossClassification
Using mixed-precision training with FP16
Saving and exporting the resulting Sentence Transformer model
Evaluating the model through sentence embeddings and cosine similarity

The training dataset remains hosted on Kaggle and is consumed by the notebook through Kaggle's mounted dataset directory. The original Parquet format can be used directly without requiring conversion to CSV, provided that the appropriate Parquet reader is available in the notebook environment.

The notebook is intended to make the training process reproducible and to provide a practical reference for users who want to inspect, reproduce, or extend the fine-tuning procedure.

Training Details

Training Dataset

Persian Sentiment Similarity Dataset

The model was trained on a Persian-language sentence-pair dataset hosted on Kaggle and used by the official training notebook.

Source: Persian Sentiment PyTorch NLP dataset and training workflow on Kaggle

Size: 315,000 training samples
Columns: sentence_0, sentence_1, and label
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1 label
type string string int
details
min: 5 tokens
mean: 28.43 tokens
max: 176 tokens
min: 5 tokens
mean: 28.43 tokens
max: 176 tokens
0: ~34.70%
1: ~31.10%
2: ~34.20%

Samples:

sentence_0	sentence_1	label
`از دید من پس از خروج و ورود مجدد سالن زیبایی زنانه اصلی کمی کمکیفیت بود و کند بود جای اصلاح دارد پیشنهاد میشود بازنگری شود و توضیح بیشتری ندارم`	`از دید من پس از خروج و ورود مجدد سالن زیبایی زنانه اصلی کمی کمکیفیت بود و کند بود جای اصلاح دارد پیشنهاد میشود بازنگری شود و توضیح بیشتری ندارم`	`0`
`از بزرگترین ایرادات فیدیبو اینکه رده سنی رو نمینویسه و سبک کتابم مشخص نمیکنه که عاشقانس علمی تخیلیه و`	`از بزرگترین ایرادات فیدیبو اینکه رده سنی رو نمینویسه و سبک کتابم مشخص نمیکنه که عاشقانس علمی تخیلیه و`	`2`
`من این کتاب و نخوندم ولی متنیش که تومن هست رو هشت تومن میفروشن این کتاب حتما باید خلاصه باشه که با وجود صوتی بودن شش تومن`	`من این کتاب و نخوندم ولی متنیش که تومن هست رو هشت تومن میفروشن این کتاب حتما باید خلاصه باشه که با وجود صوتی بودن شش تومن`	`2`

Loss: [object Object].SoftmaxLossClassification

Training Hyperparameters

Non-Default Hyperparameters

num_train_epochs: 2
fp16: True
multi_dataset_batch_sampler: round_robin

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: no
prediction_loss_only: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
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
num_train_epochs: 2
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: {'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: 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: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: round_robin

Training Logs