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
There are models that have context windows much higher than default value going up to 1M. However, the dispite showing model supports 131k windows or 1M. When trying to load with higher value, it gives an error. Once it fails to load with max supported context, lowering it doesn't help either. the model just continuesly give the same error until it is loaded in default value or similar. Sometimes, it doesn't load at all.
I know loading with a higher window takes more ram, but 64gb Ram should be able to load a 7/3b model with a 131k+ window.
Besides just showing the max window supported by the model, it should also show that the max window can be loaded given the machines hardware. This way, models can be used with their max potential with any given machine.
Currently, while downloading any model, it does show that the ram amount is required to load the model, but that might be with the default value of 2k/4k.
The text was updated successfully, but these errors were encountered:
There are models that have context windows much higher than default value going up to 1M. However, the dispite showing model supports 131k windows or 1M. When trying to load with higher value, it gives an error. Once it fails to load with max supported context, lowering it doesn't help either. the model just continuesly give the same error until it is loaded in default value or similar. Sometimes, it doesn't load at all.
I know loading with a higher window takes more ram, but 64gb Ram should be able to load a 7/3b model with a 131k+ window.
Besides just showing the max window supported by the model, it should also show that the max window can be loaded given the machines hardware. This way, models can be used with their max potential with any given machine.
Currently, while downloading any model, it does show that the ram amount is required to load the model, but that might be with the default value of 2k/4k.
The text was updated successfully, but these errors were encountered: