Jeffy ships text classification tool that runs and retrains on CPU
What happened
Jeffy released a local pretrained text classification tool, reported on the Hacker News front page on October 4. It offers 13 pretrained classifiers that run and retrain on CPU without a GPU, with a Python SDK, an HTTP API and training on custom data. It pairs a shared bge-large-en-v1.5 encoder with a logistic regression head. The report cites about 2 GB of memory and 50–80 ms p50 end-to-end latency per text on Linux aarch64.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageShow HN: Local pretrained classifiers, GPU not needed
Jeffy 提供 13 个预训练文本分类器，可在 CPU 上本地运行和重训，并支持 Python SDK、HTTP API 与自定义数据训练。工具采用共享编码器 bge-large-en-v1.5 与逻辑回归分类头，运行内存约 ~2 GB，Linux aarch64 上单条文本端到端 p50 延迟为 50–80 ms。
Heat over time
Not enough continuous observations to draw a trend yet.