dice-research/amharic-property-retriever-mbert

SentenceTransformer based on google-bert/bert-base-multilingual-cased

This is a sentence-transformers model finetuned from google-bert/bert-base-multilingual-cased. It maps sentences & paragraphs to a 768-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: google-bert/bert-base-multilingual-cased
• Maximum Sequence Length: 64 tokens
• Output Dimensionality: 768 dimensions
• Similarity Function: Cosine Similarity

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, 'architecture': 'BertModel'})
  (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})
)

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.

import json
from sentence_transformers import SentenceTransformer, util

# load trained model
model = SentenceTransformer("dice-research/amharic-property-retriever-mbert")

# input field you want to test
query = "book's ቋንቋ"


# load all candidate properties from dataset
properties = set()

with open("dice-research/amharic-property-mapping", "r", encoding="utf-8") as f:
    for line in f:
        row = json.loads(line)
        properties.add(row["property_text"])

properties = list(properties)

# compute embeddings
query_emb = model.encode(query, convert_to_tensor=True, normalize_embeddings=True)
prop_emb = model.encode(properties, convert_to_tensor=True, normalize_embeddings=True)

# compute similarity
scores = util.cos_sim(query_emb, prop_emb)[0]

# get top 5 predictions
top_k = 5
top_results = scores.topk(top_k)

print("Input:", query)
print("\nTop 5 predictions:")

for idx, score in zip(top_results.indices, top_results.values):
    print(properties[idx], score.item())

Evaluation

Metrics

Information Retrieval

• Dataset: dev-ir
• Evaluated with InformationRetrievalEvaluator

Metric	Value
cosine_accuracy@1	0.4409
cosine_accuracy@3	0.5538
cosine_accuracy@5	0.6344
cosine_accuracy@10	0.7151
cosine_precision@1	0.4409
cosine_precision@3	0.1846
cosine_precision@5	0.1269
cosine_precision@10	0.0715
cosine_recall@1	0.4409
cosine_recall@3	0.5538
cosine_recall@5	0.6344
cosine_recall@10	0.7151
cosine_ndcg@10	0.5657
cosine_mrr@10	0.5192
cosine_map@100	0.5311

Training Details

Training Dataset

• Size: 1,224 training samples
• Columns: anchor and positive
• Approximate statistics based on the first 1000 samples:

  	anchor	positive
  type	string	string
  details	min: 5 tokens mean: 7.17 tokens max: 11 tokens	min: 3 tokens mean: 3.52 tokens max: 8 tokens

• Samples:

  anchor	positive
  case's header	title
  airline's ንብረቶች	assets
  person's እናት	mother

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false
  }
  

Evaluation Dataset

• Size: 186 evaluation samples
• Columns: anchor and positive
• Approximate statistics based on the first 186 samples:

  	anchor	positive
  type	string	string
  details	min: 5 tokens mean: 7.16 tokens max: 12 tokens	min: 3 tokens mean: 3.51 tokens max: 9 tokens

• Samples:

  anchor	positive
  soccer player's ዓመታት	years
  case's rowclass	class
  type's accessyear	year

• Loss: MultipleNegativesRankingLoss with these parameters:

  {
      "scale": 20.0,
      "similarity_fct": "cos_sim",
      "gather_across_devices": false
  }
  

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: epoch
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 64
• learning_rate: 2e-05
• num_train_epochs: 4
• warmup_ratio: 0.1
• load_best_model_at_end: True
• 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: 32
• per_device_eval_batch_size: 64
• 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: linear
• lr_scheduler_kwargs: None
• 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
• bf16: False
• 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: 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}
• parallelism_config: 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
• project: huggingface
• trackio_space_id: trackio
• 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: no
• 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: True
• prompts: None
• batch_sampler: no_duplicates
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Epoch	Step	Training Loss	Validation Loss	dev-ir_cosine_ndcg@10
0.6410	25	2.67	-	-
1.0	39	-	0.6878	0.4881
1.2821	50	1.6894	-	-
1.9231	75	1.643	-	-
2.0	78	-	0.6251	0.5394
2.5641	100	1.3576	-	-
3.0	117	-	0.6248	0.5641
3.2051	125	1.2821	-	-
3.8462	150	1.2421	-	-
4.0	156	-	0.608	0.5657

• The bold row denotes the saved checkpoint.

Framework Versions

• Python: 3.9.2
• Sentence Transformers: 5.1.2
• Transformers: 4.57.6
• PyTorch: 2.8.0+cu128
• Accelerate: 1.10.1
• Datasets: 4.5.0
• Tokenizers: 0.22.2