Description
This model was developed by Kundyz Maksutova, PhD Candidate, as part of research on improving question-answering systems in the Kazakh language. It is a fine-tuned version of kz-transformers/kaz-roberta-conversational using the Kundyzka/informatics_kaz dataset. The model is optimized for answering questions in Kazakh, with a primary focus on computer science and related fields.
Key Features:
- Developer: Kundyz Maksutova, PhD Candidate
- Base Model:
kz-transformers/kaz-roberta-conversational
- Dataset:
Kundyzka/informatics_kaz
- Language: Kazakh (
kk)
- Task: Question Answering (
pipeline_tag: question-answering)
- Library:
adapter-transformers
Performance:
The model achieves the following performance metrics, highlighting its improvement after fine-tuning:
- Before Training:
- F1 Score: 17.797
- Exact Match (EM): 7.662
- After Training:
- F1 Score: 67.788
- Exact Match (EM): 51.428
These metrics were evaluated on the Kundyzka/informatics_kaz dataset, demonstrating a significant improvement in performance and reliability for domain-specific questions.
Intended Use:
This model is designed to handle natural language questions in the Kazakh language. It is particularly well-suited for:
- Educational Platforms: Assisting students with questions in computer science.
- Research Projects: Facilitating studies and experiments in Kazakh natural language processing.
- AI Applications: Powering chatbots and intelligent systems requiring accurate and domain-specific answers.
Limitations:
- Domain Dependency: The model is fine-tuned for computer science topics, and performance may degrade on unrelated queries.
- Bias: The training dataset may introduce biases that could affect the model’s responses.
- Language: The model supports only the Kazakh language and is not designed for multilingual use.
Tags:
computersciencequestion-answeringKazakhadapter-transformers
This model contributes to advancing natural language processing for low-resource languages like Kazakh, with a focus on computer science applications. For further details, fine-tuning guidelines, or customization, refer to the model repository.
