First evaluation suggest only 14B (dense) performance?
#33
by
rtzurtz
- opened
Hello team Tencent,
according to the majority of comments and this evaluation / benchmark: x(dot)com/ArtificialAnlys/status/1942375426174902354, Hunyuan-A13B's performance disappoints vs what one would expect based on your published benchmark and its parameter size (80B)?
According to that Artificial Analysis URL, Hunyuan-A13B is on par with Qwen3-14B, and it is on par with Qwen3-30B-A3B, a much fewer parameter model (30B vs your's 80B). Based on your benchmark and Hunyuan-A13B's model size (80B), I was hoping it would be on par with Qwen3-32B or better.
Can you please look into it, maybe a fix needs to be applied somewhere.