arun11karthik/cellsense-fim-1.5b
arun11karthik/cellsense-fim-1.5b is a 1.5 billion parameter fill-in-the-middle (FIM) code completion model, fine-tuned from Qwen/Qwen2.5-Coder-1.5B by arun11karthik. It is specifically designed for Jupyter notebooks, supporting a 32K-token context window. This model excels at providing repository-aware, local-import-aware, and task-aware code completions, making it highly effective for in-notebook development workflows.
Loading preview...
CellSense-FIM 1.5B: Notebook-Native Code Completion
CellSense-FIM 1.5B is a 1.5 billion parameter model from the CellSense-FIM family, developed by arun11karthik. Fine-tuned from Qwen/Qwen2.5-Coder-1.5B, it specializes in fill-in-the-middle (FIM) code completion within Jupyter notebooks, leveraging a substantial 32K-token context window.
Key Capabilities & Differentiators
- Notebook-Optimized FIM: Unlike general code models, CellSense-FIM is trained on a unique dataset that captures the specific context of Jupyter notebooks.
- Repository-Aware: Integrates surrounding files from the same repository, ensuring completions align with project conventions and helpers.
- Local-Import-Aware: Understands local imports, providing accurate signatures for calls to user-defined code.
- Task-Aware: Conditions completions on recently accessed files, reflecting the developer's current focus.
- Enhanced Performance: Achieves significant gains in completion quality metrics like edit similarity (0.07 → 0.72) and BLEU (4.5 → 56.0) compared to its base model, as well as improved token accuracy and CodeBLEU.
Recommended Usage
This model is best utilized with the CellSense JupyterLab Plugin, which automatically assembles the necessary repository, import, and task context into the model's native FIM prompt format. It supports serving via vLLM for GPU-accelerated inference or Ollama for fully local, CPU-based execution, with GGUF quantizations available for efficient deployment.