Micius-Flash, a 27B open engineering model, is released
Original title: Micius-Flash: Lightweight Agentic Intelligence for Engineering
It averages 64.2% on hardware code benchmarks and 62.4% on a semiconductor agent benchmark, above a same-size baseline.
The authors introduce Micius-Flash, a 27-billion-parameter open-source model for engineering with native multimodal input, deep reasoning, and a 256K-token context window. Expert models for semiconductor knowledge, hardware coding, multimodal understanding, and agents are distilled into one model with Multi-Teacher On-Policy Distillation. On more than 20 datasets with about 15K records, it averages 64.2% on hardware code benchmarks, 15.0 points above a same-size baseline, and 62.4% on a semiconductor agent benchmark, 17.2 points above that baseline. Micius Harness connects it to EDA tools, KiCad, and execution environments.
- Training uses a three-stage knowledge-tree data flywheel.
- MOPD supervises sequences generated by Micius itself.
- Case studies cover PCB design, chip workflow automation, and software development.
- The model is released open source for science and engineering agents.
- Products
- Micius HarnessKiCad
- AI models
- Micius-Flash
From the original article
1 min read at huggingface.co
We introduce Micius-Flash, a 27-billion-parameter model for engineering with native multimodal input, deep reasoning and a 256K-token context window. We design the Micius data pipeline as a knowledge-tree-driven data flywheel comprising three stages: knowledge tree construction and expansion, data synthesis, and quality validation with hard-example augmentation.
By Hongcheng Gao, Weiheng Li, Yipeng Wei · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.