[Bug] RuntimeError: Trying to backward through the graph a second time

[nguyendinhduybigtreetc]

Hello, i got this error when training whiper v3. does anyone know how to debug this:

---

RuntimeError Traceback (most recent call last)
/tmp/ipython-input-773422404.py in <cell line: 0>()
----> 1 trainer_stats = trainer.train()

12 frames
/usr/local/lib/python3.12/dist-packages/torch/_functorch/_aot_autograd/runtime_wrappers.py in backward(ctx, *flat_args)
2196 def backward(ctx, *flat_args):
2197 all_args = _backward_prologue_functional(
-> 2198 ctx.saved_tensors,
2199 ctx.symints,
2200 CompiledFunction.metadata,

RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed). Saved intermediate values of the graph are freed when you call .backward() or autograd.grad(). Specify retain_graph=True if you need to backward through the graph a second time or if you need to access saved tensors after calling backward.

[Datta0]

Hey @nguyendinhduybigtreetc a little more info on your setup would help.. like
Model name
how did you load the model
which training are you referring to (SFT/GRPO etc)
Which GPU are you running this on
which unsloth and unsloth zoo version are you on

[nguyendinhduybigtreetc]

@Datta0
I am training Whiper v3 large. All setup and loading of model with dataset are default from colab file on website
https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Whisper.ipynb#scrollTo=QmUBVEnvCDJv