Kaustabh Ganguly comments

Results 9 comments of


                                            Kaustabh Ganguly

'Node' object has no attribute 'output_masks'

I am also having a problem in this error . You found any solution ?

TypeError: unsupported operand type(s) for /: 'Tensor' and 'NoneType' when full finetuning gemma3

Again facing same damn issue

TypeError: unsupported operand type(s) for /: 'Tensor' and 'NoneType' when full finetuning gemma3

How to solve this

[Question] How to do an inefence on Gemma3-1b-it

`from transformers import TextIteratorStreamer from threading import Thread from unsloth import FastLanguageModel # 1) Prepare model for efficient generation FastLanguageModel.for_inference(model) # 2) Use a more directive prompt and include BOS...

[Question] How to do an inefence on Gemma3-1b-it

`# Block 15: Optional Inference Test print("\n--- Block 13: Running Basic Inference Test ---") from transformers import TextIteratorStreamer from threading import Thread from unsloth import FastLanguageModel if training_successful and 'final_model'...

[Bug]Expected all tensors to be on the same device, but found at least two devices, cuda:2 and cuda:0!

This is not a bug, unsloth runs on only 1 gpu.. It wont run in multi gpu systems. force it to use only 1 gpu

[Question] Fine-tune Gemma3 OOM

You have to use a smaller max seq length. I trained successfully on H100 80gb vram, with 8000 context window (instruction + response) and it was consuming 69gb vram

[Question] Fine-tune Gemma3 OOM

the size vram consumes increases quadratic with the sequence length for the attention layers

[Question] Fine-tune Gemma3 OOM

@shimmyshimmer hey can you just answer 1 question please - What will be the best open source model below 3b size that is most suitable for medical or clinical domain...