Visp Three.js LoRA Adapter
Visp is a Three.js-focused LoRA adapter trained on top of Qwen/Qwen2.5-Coder-1.5B-Instruct.
This repository contains the selected Visp checkpoint from the local evaluation run:
- Adapter checkpoint:
ckpt_010 - Base model:
Qwen/Qwen2.5-Coder-1.5B-Instruct - Fine-tuning type: LoRA
- LoRA rank: 8
- LoRA scale: 20
- Training framework: MLX LM
Intended Use
Visp is intended for generating and repairing self-contained Three.js / TypeScript browser game examples.
Evaluation Summary
The selected checkpoint was promoted because it had the best local expanded benchmark result while preserving the original benchmark:
- Expanded visual pass rate:
0.96 - Expanded compile pass rate:
1.0 - Original pass rate:
1.0
Local Usage
python3 -m mlx_lm.chat \
--model Qwen/Qwen2.5-Coder-1.5B-Instruct \
--adapter-path ./ckpt_010
If downloaded directly from the Hub, point --adapter-path at the folder containing adapters.safetensors and adapter_config.json.
Notes
This is an adapter-only upload. It does not include the base Qwen model weights.
Hardware compatibility
Log In to add your hardware
Quantized
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for jasontcip/visp-threejs-lora
Base model
Qwen/Qwen2.5-1.5B Finetuned
Qwen/Qwen2.5-Coder-1.5B Finetuned
Qwen/Qwen2.5-Coder-1.5B-Instruct