README.md

metadata

language: ar
license: apache-2.0
library_name: peft
base_model: UBC-NLP/MARBERT
tags:
  - arabic
  - dialect-classification
  - lora

HammaLoRAMarBert

Advanced Arabic Dialect Classification Model with Complete Training Metrics

Full Training History

epoch	train_loss	eval_loss	train_accuracy	eval_accuracy	f1	precision	recall
1	1.23777	1.22883	0.696741	0.706742	0.679064	0.704547	0.696741
2	0.740169	0.733708	0.790397	0.800562	0.791608	0.797789	0.790397
3	0.601572	0.617834	0.818182	0.821348	0.819564	0.824729	0.818182
4	0.562756	0.585901	0.824363	0.816292	0.825464	0.835567	0.824363
5	0.497183	0.534541	0.839411	0.832022	0.839956	0.842141	0.839411
6	0.467484	0.529349	0.848964	0.830899	0.850348	0.855113	0.848964
7	0.447877	0.52692	0.851773	0.832022	0.852826	0.857268	0.851773
8	0.44038	0.525875	0.854021	0.830337	0.855092	0.860913	0.854021
9	0.416875	0.513681	0.863886	0.835955	0.865207	0.870201	0.863886
10	0.397198	0.498091	0.868506	0.839888	0.869502	0.872867	0.868506
11	0.396181	0.509205	0.86757	0.835955	0.869238	0.875968	0.86757
12	0.38368	0.494237	0.873064	0.838764	0.87361	0.875448	0.873064
13	0.377543	0.496908	0.874001	0.83764	0.874749	0.877947	0.874001
14	0.371016	0.491708	0.877435	0.841573	0.878057	0.880101	0.877435
15	0.370049	0.493832	0.877872	0.840449	0.878651	0.881198	0.877872

Label Mapping:

{0: 'Egypt', 1: 'Iraq', 2: 'Lebanon', 3: 'Morocco', 4: 'Saudi_Arabia', 5: 'Sudan', 6: 'Tunisia'}

USAGE Example:

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="Hamma-16/HammaLoRAMarBert",
    device="cuda" if torch.cuda.is_available() else "cpu"
)

sample_text = "شلونك اليوم؟"
result = classifier(sample_text)
print(f"Text: {sample_text}")
print(f"Predicted: {result[0]['label']} (confidence: {result[0]['score']:.1%})")