You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hey, first of all thanks for the awesome work in this repository 🙌
I've just tried the new Qwen2.5 models, but looking through the config.ts I realized that some of the models seem to link to different versions on Huggingface. The one in the screenshot, for example, has the model id of Qwen2.5, but it seems to link to Qwen2.0.
Now I'm wondering: is only the meta information incorrect, or am I actually using the older 2.0 model, when I choose this model id? 🤔
The text was updated successfully, but these errors were encountered:
Ahh thanks for the catch! I took a pass and Qwen2.5-0.5B-Instruct-q0f16-MLC should be the only one mistaken here.
Note that model_id and model should match (i.e. the weights). But not necessarily the model_lib. In this case, Qwen2.5-0.5B can reuse the WASM for Qwen2-0.5B, since there is no architectural change, only weights change.
Hey, first of all thanks for the awesome work in this repository 🙌
I've just tried the new Qwen2.5 models, but looking through the config.ts I realized that some of the models seem to link to different versions on Huggingface. The one in the screenshot, for example, has the model id of Qwen2.5, but it seems to link to Qwen2.0.
Now I'm wondering: is only the meta information incorrect, or am I actually using the older 2.0 model, when I choose this model id? 🤔
The text was updated successfully, but these errors were encountered: