Hunyuan-A13B Technical Report
Tencent’s Hunyuan-A13B is an open-source MoE model with 80B total parameters and 13B active at inference.
The report says the model was trained on a filtered 20T-token corpus with added emphasis on STEM data. It uses supervised fine-tuning and large-scale reinforcement learning, plus a dual-mode chain-of-thought setup for fast or slower reasoning depending on task complexity. The authors claim competitive results across math, science, coding, language understanding, and agent tasks, with throughput aimed at latency-sensitive deployment. HF Daily Papers' note
The report says the model was trained on a filtered 20T-token corpus with added emphasis on STEM data. It uses supervised fine-tuning and large-scale reinforcement learning, plus a dual-mode chain-of-thought setup for fast or slower reasoning depending on task complexity. The authors claim competitive results across math, science, coding, language understanding, and agent tasks, with throughput aimed at latency-sensitive deployment. HF Daily Papers' note
score 7