Kundyzka/bert-base-multilingual-informatics-kaz

Description

This model is a fine-tuned version of google-bert/bert-base-multilingual-cased using the Kundyzka/informatics_kaz dataset. Developed by Kundyz Maksutova, PhD Candidate, this model is specifically optimized for question-answering tasks in the Kazakh language, with a focus on computer science and informatics.

Key Features:

• Developer: Kundyz Maksutova, PhD Candidate
• Base Model: google-bert/bert-base-multilingual-cased
• Dataset: Kundyzka/informatics_kaz
• Language: Kazakh (kk)
• Task: Question Answering (pipeline_tag: question-answering)
• Library: adapter-transformers

Performance:

This model demonstrates significant improvements after fine-tuning, as highlighted by the following metrics:

• Before Training:
  • F1 Score: 24.586
  • Exact Match (EM): 11.818
• After Training:
  • F1 Score: 63.317
  • Exact Match (EM): 43.162

These metrics were evaluated on the Kundyzka/informatics_kaz dataset, indicating a substantial enhancement in the model’s ability to handle domain-specific questions.

Intended Use:

This model is intended for question-answering applications in the Kazakh language. Potential use cases include:

• Educational Platforms: Assisting students with queries in informatics and computer science.
• Research Projects: Supporting research in Kazakh natural language processing.
• AI Applications: Enhancing intelligent systems, chatbots, or virtual assistants requiring Kazakh language support.

Limitations:

• Domain-Specific Training: The model is optimized for informatics and computer science topics, and performance may degrade on unrelated queries.
• Language Support: The model supports only the Kazakh language and does not handle multilingual tasks.
• Bias: Potential biases in the dataset may influence model outputs.

Tags:

• computerscience
• informatics
• question-answering
• Kazakh
• adapter-transformers

This model is a step forward in enabling high-quality question-answering systems for low-resource languages like Kazakh. For further details, customization, or fine-tuning, refer to the model repository.