Web-Search-Beta-Release-Bugs

[3Simplex]

Here we can show off the bugs we find.

Note

The context should now work up to your machines limit or the models limit whichever comes first (131072 = 128k)

Warning

In the first beta release, llama.cpp did not support the rope scaling required for full context (limit is 8192). Soon this will be updated for full 128K functionality.
Setting context higher than 8192 will cause bad output.

Please list the following in your post.

• Operating system
• GPU
• Device setting (Cuda, Vulkan, CPU)
• Context setting.

[niansa]

It's now associating that weird broken HTML-looking stuff with the name of the streamer.

[3Simplex]

This should be solved now. Please give the new beta another test. New Releases

[niansa]

[3Simplex]

Is this maybe related to... (solved in new release?)

Warning

At the time of this release, llama.cpp did not support the rope scaling required for full context (limit is 8192). Soon this will be updated for full 128K functionality.

[3Simplex]

This should be solved now. Please give the new beta another test. New Releases