zero9tech/Qwen2.5-Coder-3B-Data-Science-Insight-TR-7.6K
The Qwen2.5-Coder-3B-Data-Science-Insight-TR-7.6K model by Zero9 Tech is a 3.1 billion parameter language model developed for data mining and applied data science decision support. It features continued pre-training with Turkish adaptation and specialized fine-tuning on a data scientist dialogue dataset. This model is optimized for generating decision-oriented responses, including method selection, alternative comparisons, risk signals, and validation steps, making it suitable for analytical and strategic data science applications.
Loading preview...
Model Overview
Qwen2.5-Coder-3B-Data-Science-Insight-TR-7.6K is a 3.1 billion parameter model developed by Zero9 Tech, specifically engineered for data mining and applied data science decision support.
Key Capabilities
- Turkish Language Adaptation: Underwent continued pre-training (CPT) with approximately 10% adaptation using Wikimedia/Wikipedia data (48,148 records) to enhance its understanding and generation capabilities in Turkish.
- Domain-Specific Fine-tuning: Specialized through Supervised Fine-Tuning (SFT) on the
murataksit34/veri-bilimci-diyalog-8k-trdataset, focusing on data scientist dialogues. - Decision-Oriented Responses: Optimized to produce answers that aid in decision-making processes, covering aspects such as:
- Method selection
- Comparison of alternatives
- Identification of risk signals
- Validation steps
Training Details
The model's training involved a two-stage process:
- Continued Pre-Training (CPT): Focused on adapting to the Turkish language.
- Domain Expertise SFT: Utilized a dataset of 7,656 records, split into 6,124 for training and 1,532 for testing, to instill specialized data science knowledge.
Ideal Use Cases
This model is particularly well-suited for applications requiring analytical insights and strategic guidance within data science contexts, where clear, decision-focused outputs are paramount.