The custom adding of already downloaded models is only available on first initialisation, or after clearing cache and data (effectively the same) as mentioned in another issue.
But that aside, no models selected this way work - despite all working on the prior version.
In addition, even the models downloaded through the inbuilt list fail to run.
Case in point, Gemma 4 E4B IT. It reports memory issues (see screenshot).
It never did that on the previous version, and that model runs in every other local LLM app I have.
I've included a screenshot of it running in Box for example.
Not to mention they I can run the 12GB Gemma 4 26B APEX Mini model, Gemma 3 12B Q4_K_M, Qwen 3.5-9B and so on without issue on my Pixel 10 Pro XL (16GB RAM, G5 Tensor, masses of free space).
Something has definitely gone wrong with the loading or memory handling! I basically can't run anything on TN now!
TN error in screenshots and then import and inference of the very same Gemma 4 model in Box!
The custom adding of already downloaded models is only available on first initialisation, or after clearing cache and data (effectively the same) as mentioned in another issue.
But that aside, no models selected this way work - despite all working on the prior version.
In addition, even the models downloaded through the inbuilt list fail to run.
Case in point, Gemma 4 E4B IT. It reports memory issues (see screenshot).
It never did that on the previous version, and that model runs in every other local LLM app I have.
I've included a screenshot of it running in Box for example.
Not to mention they I can run the 12GB Gemma 4 26B APEX Mini model, Gemma 3 12B Q4_K_M, Qwen 3.5-9B and so on without issue on my Pixel 10 Pro XL (16GB RAM, G5 Tensor, masses of free space).
Something has definitely gone wrong with the loading or memory handling! I basically can't run anything on TN now!
TN error in screenshots and then import and inference of the very same Gemma 4 model in Box!