Error in MergedLinear about lora

[edwardjhu]

Thanks for catching this! I fixed it here: https://github.com/microsoft/LoRA/tree/bugfix_MergedLinear

Could you give it a try and see if this resolves the issue for you?

[maohaotiancs]

I have encountered a similar problem.
When I training on gpu, error happened.

[Baijiong-Lin]

Thanks for catching this! I fixed it here: https://github.com/microsoft/LoRA/tree/bugfix_MergedLinear

Could you give it a try and see if this resolves the issue for you?

@edwardjhu I got the same error as @maohaotiancs.

model.eval() can be run normally in both fan_in_fan_out=True and fan_in_fan_out=False cases but the forward process has an error when model.train().

[Baijiong-Lin]

@edwardjhu @maohaotiancs I have fixed this bug (#105).

[edwardjhu]

Looks good to me! Thanks @Baijiong-Lin

If this is now resolved, feel free to close the issue and I'll merge the patch with main