Issues with the Merge Kit
#3
by
rikunarita
- opened
I tried the merge kit you recommended (https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb? usp=sharing#scrollTo=1Wq4SB9A_9ic) with Qwen2.5-Coder-7B and Deepseek-coder-V2-Lite. The model repository and README were created, but the crucial model weights themselves were not saved. How can I resolve this issue?
Did you set the runtime to TPU? Maybe you didnt have the setup script correct
I was using a GPU. Does it fail with a GPU?
So turns out it doesn't matter if you are using a TPU CPU or GPU; since the model weights and architectures are different, it will always fail.
Now that does mean you can still do a frankenmerge but it will probably just not work