The client of Windows cannot download the built-in models of VLM

[jz201113008]

Failed to load files for ggml-org/SmolVLM2-256M-Video-Instruct-GGUF，The built-in retrieval model in the client has no results, only employees recommended by Vision Language Model can see the model list, but cannot select it

[solderzzc]

@jz201113008 , this model is using a different format, please check LFM2-VL 450M which is performing great at that size.

[jz201113008]

The built-in retrieval model in the client has no results, only employees recommended by Vision Language Model can see the model list, but cannot select it，However, it cannot be retrieved or downloaded in the software.

[solderzzc]

@jz201113008 I got it might be hugging face website is blocked, I'll try to get a mirror website to serve the hugging face related functionality.
I guess you could access this https://hf-mirror.com instead of this https://huggingface.co ?

[jz201113008]

In the VLM settings, selecting "auth" enables normal access to huggingface within the application, but it still cannot be downloaded in the model download interface. Additionally, I attempted to use "External (OpenAI-Compatible)" and configured a local service here. However, a prompt indicated that there were no local models on the video segment analysis interface, and when selecting the video segment dialogue, it also indicated a request failure.

[solderzzc]

@jz201113008 Thanks for the detailed bug report, we are working on it for a fix of the external OpenAI-Compatible bug.
Since we are going to support more cloud models so it takes a while to get the full test pass.

[solderzzc]

@jz201113008 , Thanks for providing valuable feedback, we've fixed the bugs but since we enabled may cloud APIs and 10k-20k code base change, so we are testing still. If you want, please check out the latest version ( not yet rolling out ) below.
I've tested it with a RTX 4070 desktop version. Please let me know if this has any issues.

https://releases.sharpai.org/releases/SharpAI-Aegis-Setup-0.2.9-win-x64.exe
https://releases.sharpai.org/releases/SharpAI-Aegis-0.2.9-win-x64.exe

[jz201113008]

This is a screenshot of the configuration of the LLM, but it still cannot perform the search.

This is the place to configure VLM. Enter the employee recommendation model. On the right, you can see the model introduction. However, on the left, the loading still fails, and an error message indicating a failed request is displayed at the top.；
Furthermore, I also tried using the API Providers model. I selected the GLM model and filled in the corresponding APIKEY. Finally, all the "selected" showed as "connection successful" without any errors. However, when switching to the video segment analysis dialogue and the Aegis dialogue interface, the dialogue results all showed "Error: fetch failed".