Er-Medium-12.5M
FromZero SLM (Qwen3.5 lineage, 12,497,520 parameters) trained on ~34.8B tokens from educational and web corpora. Base continuation model - not chat-tuned.
Model pageAskJASP is the public site for FromZero - tiny, open language models you can browse, understand, and run yourself. Explore the catalog, read how they work, and try them in the browser.
We do not host free online inference. Download weights and run locally or on your own stack.
Catalog
Base completion models. Small, open, and meant for research and experimentation - not production assistants.
FromZero SLM (Qwen3.5 lineage, 12,497,520 parameters) trained on ~34.8B tokens from educational and web corpora. Base continuation model - not chat-tuned.
Model pageTiny FromZero model trained on synthetic data. Extreme size - not for factual Q&A or chat.
Model pageExtremely small FromZero SLM. Useful for toy demos and architecture experiments.
Model pageSwipe for more
Get started
Open the catalog and choose a FromZero checkpoint on Hugging Face.
Use hf download (Hugging Face CLI) or the HF UI to pull the model folder locally.
Load with Transformers (or your stack). Prefer completion-style prompts for base models.
Browser Lab
WebGPU or WASM, fully on-device. Prompts never leave your machine; weights download from Hugging Face. An experimental completion playground - not a hosted chat service.