BERT Medium Amharic Text Embedding

This is a sentence-transformers model finetuned from yosefw/bert-medium-am-embed on the json dataset. It maps sentences & paragraphs to a 512-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: yosefw/bert-medium-am-embed
Maximum Sequence Length: 512 tokens
Output Dimensionality: 512 dimensions
Similarity Function: Cosine Similarity
Training Dataset:
- json
Language: en
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': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 512, '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})
  (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("rasyosef/bert-amharic-text-embedding-medium")
# Run inference
sentences = [
  "የተደጋገመው የመሬት መንቀጥቀጥና የእሳተ ገሞራ ምልክት በአፋር ክልል",
  "ከተደጋጋሚ መሬት መንቀጥቀጥ በኋላ አፋር ክልል እሳት ከመሬት ውስጥ ሲፈላ ታይቷል፡፡ ከመሬት ውስጥ እሳትና ጭስ የሚተፋው እንፋሎቱ ዛሬ ማለዳውን 11 ሰዓት ግድም ከከባድ ፍንዳታ በኋላየተስተዋለ መሆኑን የአከባቢው ነዋሪዎች እና ባለስልጣናት ለዶቼ ቬለ ተናግረዋል፡፡ አለት የሚያፈናጥር እሳት ነው የተባለው እንፋሎቱ በክልሉ ጋቢረሱ (ዞን 03) ዱለቻ ወረዳ ሰጋንቶ ቀበሌ መከሰቱን የገለጹት የአከባቢው የአይን እማኞች ከዋናው ፍንዳታ በተጨማሪ በዙሪያው ተጨማሪ ፍንዳታዎች መታየት ቀጥሏል ባይ ናቸው፡፡",
  "ለኢትዮጵያ ብሔራዊ ባንክ ዋጋን የማረጋጋት ቀዳሚ ዓላማ ጋር የተጣጣሙ የገንዘብ ፖሊሲ ምክረ ሀሳቦችን እንዲሰጥ የተቋቋመው የኢትዮጵያ ብሔራዊ ባንክ የገንዘብ ፖሊሲ ኮሚቴ እስካለፈው ህዳር ወር የነበረው እአአ የ2024 የዋጋ ግሽበት በተለይምምግብ ነክ ምርቶች ላይ ከአንድ ዓመት በፊት ከነበው ጋር ሲነጻጸር መረጋጋት ማሳየቱን ጠቁሟል፡፡ ዶይቼ ቬለ ያነጋገራቸው የአዲስ አበባ ነዋሪዎች ግን በዚህ የሚስማሙ አይመስልም፡፡ ከአምና አንጻር ያልጨመረ ነገር የለም ባይ ናቸው፡፡ የኢኮኖሚ  ባለሙያም በሰጡን አስተያየት ጭማሪው በሁሉም ረገድ የተስተዋለ በመሆኑ የመንግስት ወጪን በመቀነስ ግብርናው ላይ አተኩሮ መስራት ምናልባትም የዋጋ መረጋጋቱን ሊያመጣ ይችላል ይላሉ፡፡"
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 512]

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

Evaluation

Metrics

Information Retrieval

Datasets: dim_512, dim_384, dim_256, dim_128 and dim_64
Evaluated with InformationRetrievalEvaluator

Metric	dim_512	dim_384	dim_256	dim_128	dim_64
cosine_accuracy@1	0.581	0.5765	0.5714	0.5573	0.5162
cosine_accuracy@3	0.7081	0.7042	0.7061	0.6946	0.6519
cosine_accuracy@5	0.7587	0.7559	0.7514	0.7369	0.701
cosine_accuracy@10	0.8175	0.8149	0.8059	0.7931	0.7632
cosine_precision@1	0.581	0.5765	0.5714	0.5573	0.5162
cosine_precision@3	0.236	0.2347	0.2354	0.2315	0.2173
cosine_precision@5	0.1517	0.1512	0.1503	0.1474	0.1402
cosine_precision@10	0.0817	0.0815	0.0806	0.0793	0.0763
cosine_recall@1	0.581	0.5765	0.5714	0.5573	0.5162
cosine_recall@3	0.7081	0.7042	0.7061	0.6946	0.6519
cosine_recall@5	0.7587	0.7559	0.7514	0.7369	0.701
cosine_recall@10	0.8175	0.8149	0.8059	0.7931	0.7632
cosine_ndcg@10	0.6958	0.6919	0.6863	0.6734	0.6374
cosine_mrr@10	0.6572	0.653	0.6482	0.6353	0.5975
cosine_map@100	0.6628	0.6587	0.6543	0.6414	0.6043

Training Details

Training Dataset

json

Dataset: json
Size: 28,046 training samples
Columns: anchor and positive
Approximate statistics based on the first 1000 samples:
anchor positive
type string string
details
min: 4 tokens
mean: 15.02 tokens
max: 38 tokens

min: 47 tokens
mean: 213.84 tokens
max: 512 tokens

	anchor	positive
type	string	string
details	min: 4 tokens mean: 15.02 tokens max: 38 tokens	min: 47 tokens mean: 213.84 tokens max: 512 tokens

Samples:

anchor	positive
`የዱር እንስሳት ከሰዎች ጋር በሚኖራቸው ቁርኝት ለኮሮናቫይረስ ተጋላጭ እንዳይሆኑ የመከላከል ተግባራትን እያከናወኑ መሆኑን ባለስልጣኑ አስታወቀ፡፡`	ባሕርዳር፡ ግንቦት 18/2012 ዓ.ም (አብመድ) የአማራ ክልል የአካባቢ፣ የደንና የዱር እንስሳት ጥበቃና ልማት ባለስልጣን በሚያስተዳድራቸው ብሔራዊ ፓርኮች እና የማኅበረሰብ ጥብቅ ሥፍራዎች ከኮሮናቫይረስ ተጋላጭነት ለመከላከል እየሠራ መሆኑን አስታውቋል፡፡የባለስልጣኑ የኮሙዩኒኬሽን ዳይሬክተር ጋሻው እሸቱ 10 በሚሆኑ ብሔራዊ ፓርኮችና የማኅበረሰብ ጥብቅ ሥፍራዎች የኮሮና ቫይረስን መከላከል በሚቻልባቸው ቅድመ ተግባራት እና ርምጃዎች ላይ መምከራቸውን ተናግረዋል፡፡ የዱር እንስሳት በመንጋ የሚኖሩ፣ እርስ በርሳቸው ተመጋጋቢ፣ ከሰዎች እና ከቤት እንስሳቶች ጋር ሊቀላቀሉ የሚችሉ በመሆናቸው በኮሮናቫይረስ ከተጋለጡ ‘‘የኮሮናቫይረስ ተጋላጭነት በብርቅየ የዱር እንስሳት ብዝኃ ሕይወት ላይ ስጋት መሆን የለበትም’’ ያሉት አቶ ጋሻው በፓርኮቹ ውስጥ ለሚሠሩ የጥበቃ፣ ስካውት እና ለጽሕፈት ቤት ሠራተኞች በዘርፉ ላይ ያተኮረ የኮሮናቫይረስ መከላከያ ትምህርቶችን እና የቁሳቁስ ድጋፎችን ማድረጋቸውን አስታውቀዋል፡፡
`የትግራይ ክልል የአየር መሥመር ለአገልግሎት ክፍት ሆነ፡፡`	የትግራይ ክልል የአየር መሥመር ለአገልግሎት ክፍት ሆነ፡፡ ባሕር ዳር፡ ታኅሣሥ 05/2013 ዓ.ም (አብመድ) በሰሜን ኢትዮጵያ ትግራይ ክልል የህግ ማስከበር ሂደትን ተከትሎ ተዘግቶ የነበረው የአየር ክልል ከዛሬ ታህሣሥ 5/2013 ዓ.ም ከቀኑ 8 ሰዓት ጀምሮ በሰሜን የኢትዮጵያ የአየር ክልል ውስጥ የሚያቋርጡ የአለም አቀፍ እና የሃገር ውስጥ የበረራ መስመሮች ለአገልግሎት ክፍት ሆነዋል፡፡ አገልግሎት መሥጠት የሚችሉ ኤርፖርቶች በረራ ማስተናገድ የሚችሉ መሆኑንም የኢትዮጵያ ሲቪል አቪዬሽን ባለስልጣን ገልጿል::
`የአውሮፓ ኢንቨስትመንት ባንክ ለመንግሥት 76 ሚሊዮን ዶላር ሊያበድር ነው`	በዳዊት እንደሻውየአውሮፓ ኢንቨስትመንት ባንክ ጽሕፈት ቤቱን በአዲስ አበባ ከከፈተ ከሁለት ዓመት በኋላ ትልቅ ነው የተባለለትን የ76 ሚሊዮን ዶላር ብድር ስምምነት ለመፈራረም፣ ኃላፊዎቹን ወደ ኢትዮጵያ ይልካል፡፡ከወር በፊት በኢትዮጵያ መንግሥትና በባንኩ መካከል የተደረገው ይኼ የብድር ስምምነት፣ የኢትዮጵያ ልማት ባንክ በሊዝ ፋይናንሲንግ ለአነስተኛና ለመካከለኛ ኢንተርፕራይዞች ለሚያደርገው እገዛ ይውላል፡፡የአውሮፓ ኢንቨስትመንት ባንክ ምክትል ፕሬዚዳንት ፒም ቫን በሌኮም፣ እንዲሁም ሌሎች ኃላፊዎች ይመጣሉ ተብሎ ይጠበቃል፡፡በዚህም መሠረት የባንኩ ኃላፊዎች ከገንዘብና ኢኮኖሚ ትብብር ሚኒስቴር ጋር አድርገውት ከነበረው ስምምነት የሚቀጥልና ተመሳሳይ የሆነ ስምምነት፣ ከኢትዮጵያ ልማት ባንክ ጋር እንደሚያደርጉ ይጠበቃል፡፡እ.ኤ.አ. እስከ 2022 ድረስ የሚቀጥለው አነስተኛና መካከለኛ ኢንተርፕራይዞችን የማገዝ ፕሮጀክት 276 ሚሊዮን ዶላር ወጪ የሚያስወጣ ሲሆን፣ ባለፈው ዓመት የዓለም ባንክ ወደ 200 ሚሊዮን ዶላር ብድር ሰጥቷል፡፡በአውሮፓ ኢንቨስትመንት ባንክ የሚሰጠው ብድር፣ የኢትዮጵያ ልማት ባንክን የሊዝ ፋይናንሲንግ ሥራ እንደሚያግዝ ጉዳዩ የሚመለከታቸው የልማት ባንክ ኃላፊዎች ለሪፖርተር ተናግረዋል፡፡ ‹‹በተጨማሪም የውጭ ምንዛሪ እጥረቱን ለማቃለል ያግዛል፤›› ሲሉ ኃላፊው ገልጸዋል፡፡በልማት ባንክ በኩል የሚደረገው እገዛ በሁለት መስኮቶች የሚወጣ ሲሆን፣ አንደኛው በቀጥታ በባንክ እንደ ሊዝ ፋይናንሲንግ ሲሰጥ ሌላው ደግሞ እንደ መሥሪያ ካፒታል ልማት ባንክ ለመረጣቸው 12 ባንኮችና ዘጠኝ ማይክሮ ፋይናንሶች ይሰጣል፡፡የአውሮፓ ኢንቨስትመንት ባንክ በኢትዮጵያ መንቀሳቀስ ከጀመረ ከ1980ዎቹ ጀምሮ ወደ ግማሽ ቢሊዮን ዶላር የሚጠጋ ለኃይል፣ ለኮሙዩኒኬሽንና ለግሉ ዘርፍ ኢ...

Loss: MatryoshkaLoss with these parameters:

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        512,
        384,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

Training Hyperparameters

Non-Default Hyperparameters

eval_strategy: epoch
per_device_train_batch_size: 64
per_device_eval_batch_size: 64
gradient_accumulation_steps: 2
num_train_epochs: 4
lr_scheduler_type: cosine
warmup_ratio: 0.1
fp16: True
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: 64
per_device_eval_batch_size: 64
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: 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: 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}
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

Epoch	Step	Training Loss	dim_512_cosine_ndcg@10	dim_384_cosine_ndcg@10	dim_256_cosine_ndcg@10	dim_128_cosine_ndcg@10	dim_64_cosine_ndcg@10
0.0456	10	14.3172	-	-	-	-	-
0.0911	20	11.9004	-	-	-	-	-
0.1367	30	8.2867	-	-	-	-	-
0.1822	40	4.869	-	-	-	-	-
0.2278	50	3.7541	-	-	-	-	-
0.2733	60	3.1055	-	-	-	-	-
0.3189	70	2.6283	-	-	-	-	-
0.3645	80	2.2792	-	-	-	-	-
0.4100	90	2.0364	-	-	-	-	-
0.4556	100	1.9502	-	-	-	-	-
0.5011	110	1.6862	-	-	-	-	-
0.5467	120	1.6991	-	-	-	-	-
0.5923	130	1.5849	-	-	-	-	-
0.6378	140	1.3585	-	-	-	-	-
0.6834	150	1.464	-	-	-	-	-
0.7289	160	1.6712	-	-	-	-	-
0.7745	170	1.4967	-	-	-	-	-
0.8200	180	1.4184	-	-	-	-	-
0.8656	190	1.2148	-	-	-	-	-
0.9112	200	1.3443	-	-	-	-	-
0.9567	210	1.1794	-	-	-	-	-
1.0	220	1.1257	0.6572	0.6578	0.6471	0.6308	0.5837
1.0456	230	1.2824	-	-	-	-	-
1.0911	240	1.2316	-	-	-	-	-
1.1367	250	1.1745	-	-	-	-	-
1.1822	260	0.9189	-	-	-	-	-
1.2278	270	0.977	-	-	-	-	-
1.2733	280	0.9832	-	-	-	-	-
1.3189	290	0.9445	-	-	-	-	-
1.3645	300	0.8845	-	-	-	-	-
1.4100	310	0.754	-	-	-	-	-
1.4556	320	0.7767	-	-	-	-	-
1.5011	330	0.6453	-	-	-	-	-
1.5467	340	0.6502	-	-	-	-	-
1.5923	350	0.6711	-	-	-	-	-
1.6378	360	0.6081	-	-	-	-	-
1.6834	370	0.5782	-	-	-	-	-
1.7289	380	0.793	-	-	-	-	-
1.7745	390	0.6978	-	-	-	-	-
1.8200	400	0.7294	-	-	-	-	-
1.8656	410	0.6582	-	-	-	-	-
1.9112	420	0.5806	-	-	-	-	-
1.9567	430	0.5558	-	-	-	-	-
2.0	440	0.5417	0.6831	0.6801	0.6744	0.6640	0.6246
2.0456	450	0.6179	-	-	-	-	-
2.0911	460	0.5952	-	-	-	-	-
2.1367	470	0.604	-	-	-	-	-
2.1822	480	0.4688	-	-	-	-	-
2.2278	490	0.4907	-	-	-	-	-
2.2733	500	0.5165	-	-	-	-	-
2.3189	510	0.4703	-	-	-	-	-
2.3645	520	0.4971	-	-	-	-	-
2.4100	530	0.4522	-	-	-	-	-
2.4556	540	0.4145	-	-	-	-	-
2.5011	550	0.344	-	-	-	-	-
2.5467	560	0.392	-	-	-	-	-
2.5923	570	0.3371	-	-	-	-	-
2.6378	580	0.3402	-	-	-	-	-
2.6834	590	0.3535	-	-	-	-	-
2.7289	600	0.4581	-	-	-	-	-
2.7745	610	0.3701	-	-	-	-	-
2.8200	620	0.4221	-	-	-	-	-
2.8656	630	0.3886	-	-	-	-	-
2.9112	640	0.3828	-	-	-	-	-
2.9567	650	0.3737	-	-	-	-	-
3.0	660	0.3318	0.6921	0.6887	0.6852	0.6699	0.6339
3.0456	670	0.4025	-	-	-	-	-
3.0911	680	0.4092	-	-	-	-	-
3.1367	690	0.3605	-	-	-	-	-
3.1822	700	0.3218	-	-	-	-	-
3.2278	710	0.3362	-	-	-	-	-
3.2733	720	0.3451	-	-	-	-	-
3.3189	730	0.3476	-	-	-	-	-
3.3645	740	0.3594	-	-	-	-	-
3.4100	750	0.3324	-	-	-	-	-
3.4556	760	0.3144	-	-	-	-	-
3.5011	770	0.2667	-	-	-	-	-
3.5467	780	0.3241	-	-	-	-	-
3.5923	790	0.253	-	-	-	-	-
3.6378	800	0.2916	-	-	-	-	-
3.6834	810	0.2632	-	-	-	-	-
3.7289	820	0.348	-	-	-	-	-
3.7745	830	0.2788	-	-	-	-	-
3.8200	840	0.3224	-	-	-	-	-
3.8656	850	0.3144	-	-	-	-	-
3.9112	860	0.2926	-	-	-	-	-
3.9567	870	0.3002	-	-	-	-	-
3.9841	876	-	0.6958	0.6919	0.6863	0.6734	0.6374

The bold row denotes the saved checkpoint.

Framework Versions

Python: 3.10.12
Sentence Transformers: 3.3.1
Transformers: 4.47.1
PyTorch: 2.5.1+cu121
Accelerate: 1.2.1
Datasets: 3.2.0
Tokenizers: 0.21.0

Citation

BibTeX

Sentence Transformers

@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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

rasyosef
/

bert-amharic-text-embedding-medium