sanjaymalladi/DataSense-GPU-Full
VISIONConcurrent Unit Cost:1Model Size:5.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 5, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold
DataSense-GPU-Full by sanjaymalladi is a 5.1 billion parameter language model fine-tuned as a data-science agent. It specializes in generating cuDF and cuML code from natural language, enabling significant execution speedups for data processing and machine learning tasks on GPUs. This model is optimized for accelerating data science workflows by leveraging GPU capabilities.
Loading preview...
Overview
DataSense-GPU-Full is a 5.1 billion parameter language model developed by sanjaymalladi, specifically fine-tuned to act as a data-science agent. Its primary function is to translate natural language queries into executable cuDF and cuML code, facilitating GPU-accelerated data processing and machine learning.
Key Capabilities
- GPU-Optimized Code Generation: Generates code utilizing NVIDIA's cuDF and cuML libraries, designed for high-performance computing on GPUs.
- Performance Acceleration: Demonstrates substantial speedups, such as a 215x execution speedup for a Random Forest risk model and a 58.9x speedup for rolling window operations on large datasets.
- Integrated Architecture: The LoRA adapter is fused directly into the base weights, eliminating the need for external PEFT dependencies during deployment.
Good For
- Accelerating Data Science Workflows: Ideal for data scientists and developers looking to leverage GPU power for data manipulation and machine learning tasks.
- Generating cuDF/cuML Code: Users can describe their data science needs in natural language and receive optimized GPU code.
- High-Performance Data Processing: Suitable for applications requiring rapid processing of large datasets, as evidenced by benchmarks on Kaggle Tesla T4 with cuDF 26.02.