Rudblest/projedanismanai
Rudblest/projedanismanai is a 12 billion parameter instruction-tuned language model, fine-tuned from mistralai/Mistral-Nemo-Instruct-2407 using QLoRA 4-bit on a specialized 3043-example Turkish dataset. It is specifically optimized to act as an AI consultant for TEKNOFEST and TÜBİTAK competition participants, excelling in technical report writing, project idea clarification, and application preparation in Turkish. The model has a maximum sequence length of 3072 tokens and achieved a training loss of 0.3591.
Loading preview...
ProjeDanışmanAi: Turkish AI Consultant for TEKNOFEST and TÜBİTAK
Rudblest/projedanismanai is a 12 billion parameter language model, fine-tuned from the mistralai/Mistral-Nemo-Instruct-2407 base model. Its core purpose is to serve as a dedicated AI consultant for participants in TEKNOFEST and TÜBİTAK competitions, providing support exclusively in Turkish.
Key Capabilities
This model has been fine-tuned using QLoRA 4-bit with unsloth on a unique dataset of 3043 Turkish instruction-output pairs, focusing on competition-specific tasks. It excels at:
- Technical Report Writing: Assisting with TEKNOFEST KTR/PTR technical reports.
- Application Preparation: Supporting TÜBİTAK application processes.
- Project Development: Clarifying project ideas, performing risk analysis, and evaluating feasibility.
- Content Generation: Producing project titles and summaries.
- Presentation Support: Aiding in jury and presentation preparations.
Training Details and Limitations
The model was trained for 5 epochs with a maximum sequence length of 3072 tokens, achieving a training loss of 0.3591. Its specialized training dataset includes categories like rapor_yazimi (report writing), sifirdan_proje (project from scratch), and strateji (strategy).
Important Limitations:
- Language: Responds only in Turkish.
- Scope: Rejects out-of-domain questions (e.g., food, stock market, health).
- Performance: May show reduced performance on general engineering topics outside of TEKNOFEST/TÜBİTAK contexts.