Show HN: Sweep, Open-weights 1.5B model for next-edit autocomplete
3 months ago
- #Code Autocomplete
- #AI
- #Machine Learning
- Sweep Next-Edit 1.5B is a model for next-edit autocomplete, quantized to Q8_0 GGUF format.
- It predicts your next code edit before you make it, running locally on your laptop in under 500ms with speculative decoding.
- Outperforms models over 4x its size on next-edit benchmarks.
- Usage involves downloading run_model.py and the model file, then installing dependencies via pip.
- Model details: GGUF format (Q8_0 quantization), 1.5B parameters, 8192 token context length, based on Qwen2.5-Coder.
- Includes a specific prompt format with file context, recent diffs, and current state for predictions.
- Links provided for a blog post with technical details and benchmarks, and a JetBrains Plugin.
- Licensed under Apache 2.0, with 21 downloads last month.
- Hardware compatibility includes 8-bit inference, but no Inference Provider support currently.