Model Overview
Dannys0n/Qwen3-1.7B-seed_gen_voronoi is a specialized 2 billion parameter language model, fine-tuned from the Qwen/Qwen3-1.7B architecture. It leverages QLoRA (4-bit) with supervised fine-tuning over 3 epochs, using a learning rate of 0.0002 and a LoRA rank of 16 with alpha 32. The model was trained on the Dannys0n/test-dataset.
Key Capabilities
- Voronoi Seed Generation: The primary function is to generate seed vectors as structured JSON for creating voronoi diagrams.
- Hotspot Load Balancing: It can place shard centers in a normalized 2D world, considering hotspot positions, weights, and radii to balance load.
- Structured JSON Output: Designed to return only valid JSON, making it suitable for programmatic integration.
Intended Use Cases
This model is a test model developed for the CS-394/594 class at DigiPen. It is particularly useful for:
- Generating Voronoi Visualizations: Users can prompt the model with example inputs to receive seed vectors, which can then be used to generate voronoi diagrams. A Python notebook is available for local voronoi visualization.
Limitations
- Single-Turn Conversations: This model is designed for single-turn interactions and does not support long, multi-turn conversations.