BGE m3 Uzbek Legal Matryoshka

This is a sentence-transformers model finetuned from BAAI/bge-m3 on the json dataset. 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: BAAI/bge-m3
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 1024 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json
Language: uz
License: apache-2.0

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': 8192, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (1): Pooling({'word_embedding_dimension': 1024, '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("fitlemon/bge-m3-uz-legal-matryoshka")
# Run inference
sentences = [
    "Intizomiy jazolar qanday qo'llaniladi?",
    'Intizomiy javobgarlik xodim tomonidan intizomiy qilmish (ushbu Kodeks 301-moddasining \nikkinchi qismi) sodir etilganligi uchun yuzaga keladigan va ushbu xodimga nisbatan intizomiy jazo \nchorasi qo‘llanilishida ifodalanadigan yuridik javobgarlikdir. \nIntizomiy javobgarlikning turlari umumiy va maxsus intizomiy javobgarlikdan iboratdir. \nUmumiy intizomiy javobgarlik ushbu Kodeks va ichki mehnat tartibi qoidalari bilan tartibga \nsolinadigan javobgarlik bo‘lib, u xodimga nisbatan ushbu Kodeksning 312-moddasida nazarda \ntutilgan intizomiy jazo choralaridan birini qo‘llashdan iborat va barcha xodimlarga nisbatan  tatbiq \netiladi, bundan o‘zi uchun maxsus intizomiy javobgarlik belgilangan shaxslar mustasno. \nMaxsus intizomiy javobgarlik xodimlarning faqat alohida toifalari uchun qonunda, \nshuningdek intizom to‘g‘risidagi ustavlar va nizomlarda nazarda tutilgan hamda xodimga nisbatan tegishli qonunda, intizom haqidagi ustavda va nizomda nazarda tutilgan intizomi y choralarni \nqo‘llashdan iborat javobgarlikdir.',
    'Ushbu Kodeksda nazarda tuti lgan asoslardan tashqari, O‘zbekiston Respublikasi hududida \nmehnat faoliyatini amalga oshirish huquqiga doir tasdiqnomaning amal qilish muddati tugatilganligi \nyoki bekor qilinganligi chet el fuqarosi bilan tuzilgan mehnat shartnomasini bekor qilish uchun a sos \nbo‘ladi.  \nO‘zbekiston Respublikasi hududida mehnat faoliyatini amalga oshirish huquqiga doir \ntasdiqnomaning amal qilish muddati tugashi munosabati bilan uning amal qilish muddati tugagan \nkunda mehnat shartnomasi bekor qilinishi lozim.  \nO‘zbekiston Resp ublikasi hududida mehnat faoliyatini amalga oshirish huquqiga doir \ntasdiqnoma bekor qilinganda mehnat shartnomasi ish beruvchi O‘zbekiston Respublikasi Bandlik va \nmehnat munosabatlari vazirligi huzuridagi Tashqi mehnat migratsiyasi agentligining tegishli \nxabarnomasini olgan kunda bekor qilinishi lozim.',
]
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]

Evaluation

Metrics

Information Retrieval

Datasets: dim_1024, dim_768, dim_512, dim_256, dim_128 and dim_64
Evaluated with InformationRetrievalEvaluator

Metric	dim_1024	dim_768	dim_512	dim_256	dim_128	dim_64
cosine_accuracy@1	0.4915	0.4839	0.4858	0.4687	0.4668	0.482
cosine_accuracy@3	0.8178	0.8083	0.8027	0.7894	0.7837	0.7552
cosine_accuracy@5	0.8956	0.8918	0.8843	0.8805	0.8653	0.8558
cosine_accuracy@10	0.9469	0.9431	0.9431	0.9374	0.9279	0.9241
cosine_precision@1	0.4915	0.4839	0.4858	0.4687	0.4668	0.482
cosine_precision@3	0.2726	0.2694	0.2676	0.2631	0.2612	0.2517
cosine_precision@5	0.1791	0.1784	0.1769	0.1761	0.1731	0.1712
cosine_precision@10	0.0947	0.0943	0.0943	0.0937	0.0928	0.0924
cosine_recall@1	0.4915	0.4839	0.4858	0.4687	0.4668	0.482
cosine_recall@3	0.8178	0.8083	0.8027	0.7894	0.7837	0.7552
cosine_recall@5	0.8956	0.8918	0.8843	0.8805	0.8653	0.8558
cosine_recall@10	0.9469	0.9431	0.9431	0.9374	0.9279	0.9241
cosine_ndcg@10	0.7326	0.7268	0.7245	0.712	0.7055	0.7055
cosine_mrr@10	0.6621	0.6556	0.6529	0.6382	0.6328	0.6348
cosine_map@100	0.6646	0.6584	0.6555	0.6411	0.6362	0.6383

Training Details

Training Dataset

json

Dataset: json
Size: 4,737 training samples
Columns: question and chunk
Approximate statistics based on the first 1000 samples:
question chunk
type string string
details
min: 10 tokens
mean: 22.31 tokens
max: 45 tokens

min: 30 tokens
mean: 269.53 tokens
max: 520 tokens

	question	chunk
type	string	string
details	min: 10 tokens mean: 22.31 tokens max: 45 tokens	min: 30 tokens mean: 269.53 tokens max: 520 tokens

Samples:

question	chunk
`Noqulay tabiiy-iqlim sharoitlaridagi ish uchun mehnatga haq to‘lash koeffitsiyentlari haqida {chapter} va {section}da nima deyiladi?`	Noqulay tabiiy-iqlim sharoitlaridagi ish uchun mehnatga haq to‘lash koeffitsiyenti ayrim hududlardagi mehnat sharoitlarining xususiyatlari inobatga olingan holda xodimlarga to‘lanadigan kompensatsiya xususiyatiga ega bo‘lgan ustama turidir. Koeffitsiyentlarning eng kam miqdorlari va ularni qo‘llash tartibi O‘zbekiston Respublikasi Vazirlar Mahkamasi tomonidan belgilanadi. 28-bob. Xodim mehnatining xususiyati bilan bog‘liq bo‘lgan mehnatni huquqiy jihatdan tartibga solishning o‘ziga xos xususiyatlari 1-§. Tashkilot rahbarining, uning o‘rinbosarlarining, tashkilot bosh buxgalterining va tashkilot alohida bo‘linmasi rahbarining mehnatini huquqiy jihatdan tartibga solishning o‘ziga xos xususiyatlari
`Homiladorlik yoki farzandlar borligi bilan bog‘liq sabablarga ko‘ra ishga qabul qilmaslik qonunga xilofmi?`	Ishga qabul qilishni qonunga xilof ravishda rad etishga yo‘l qo‘yilmaydi. Quyidagilar ishga qabul qilishni qonunga xilof ravishda rad etishdir: mehnat va mashg‘ulotlar sohasida kamsitishni taqiqlash to‘g‘risidagi talablarni buzish; ish beruvchi tomonidan ishga taklif etilgan shaxslarni ishga qabul qilmaslik; ish beruvchi qonunga muvofiq mehnat shartnomasini tuzishi shart bo‘lgan shaxslarni (ish o‘rinlarining belgilangan eng k am soni hisobiga ishga yuborilgan shaxslarni, ish beruvchi alohida asoslar bo‘yicha mehnat shartnomasini bekor qilgan shaxslarni, ular qayta ishga qabul qilingan taqdirda va boshqalarni) ishga qabul qilmaslik; homiladorlik yoki farzandlar borligi bilan bog‘liq sabablarga ko‘ra ishga qabul qilmaslik; Oldingi tahrirga qarang. sudlanganligi, shu jumladan tugallangan va olib tashlangan sudlanganligi sababli shaxslarni ishga qabul qi lmaslik, bundan qonunchilikda nazarda tutilgan hollar mustasno, yoxud shaxslarni ularning qarindoshlari sudlanganligi, shu j...
`Xizmat safariga yuborish uchun nogironligi bo‘lgan xodimlarning roziligi qanday ahamiyatga ega {chapter} va {section}da?`	`Nogironligi bo‘lgan xodimlarni xizmat safariga yuborishga, tungi ishlarga, ish vaqtidan tashqari ishlarga hamda dam olish va ishlanmaydigan bayram kunlaridagi ishlarga jalb qilishga faqat ularning roziligi bilan, agar ushbu xodimlar uchun bunday ishlar tibbiy -ijtimoiy ekspert komissiyasi tavsiyalarida taqiqlanmagan bo‘lsa, yo‘l qo‘yiladi.`

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        1024,
        768,
        512,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
learning_rate: 2e-05
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
fp16: True
tf32: False
load_best_model_at_end: True
optim: adamw_torch_fused
batch_sampler: no_duplicates

All Hyperparameters

Click to expand

overwrite_output_dir: False
do_predict: False
eval_strategy: epoch
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: 2e-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: 4
max_steps: -1
lr_scheduler_type: cosine
lr_scheduler_kwargs: {}
warmup_ratio: 0.1
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: False
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: 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_fused
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: no_duplicates
multi_dataset_batch_sampler: proportional

Training Logs