Zach Mueller comments

Results 411 comments of


                                            Zach Mueller

trafficstars

Save model callback breaks multi-gpu training

@jph00 feel free to assign to me here!

Save model callback breaks multi-gpu training

@caleb-depotanalytics can you post the full trace for me? I received something like: ```python Traceback (most recent call last): File "/home/zach_mueller_huggingface_co/fastai_bug.py", line 19, in learn.fit_flat_cos(2, 1e-3, cbs=[SaveModelCallback()]) File "/home/zach_mueller_huggingface_co/miniconda3/envs/accelerate/lib/python3.9/site-packages/fastai/callback/schedule.py", line...

Save model callback breaks multi-gpu training

@caleb-depotanalytics can you verify using `pip install git+https://github.com/muellerzr/fastai@fix-save` fixes this for you? Thanks!

Add a .gitignore and make HF deps fully optional

No worries @danielhanchen, no rush :) the integration on our side can still be *made*, just not merged until this and I probably won’t get to that until later today/tommorow,...

Target is not converted to 'Long' error

How are you making your databunch?

Issue running lr_find after importing fastdebug

@kevinbird15 is this still true on the latest fastai version? (2.4?)

Apply deprecated `evaluation_strategy`

@BenjaminBossan yes, time has come :)

Issue with the unet example on a different model, reduction dim mismatch

cc @sayakpaul +related accelerate issue: https://github.com/huggingface/accelerate/issues/2497

fix prompt tunning + deepspeed zero3 + checkpoint_saving hang issue

Again though: this is deepspeed specific based on model sharding/splitting, so we should only modify the code for deepspeed specifically, unless there is a reason not to. (Aka, deepspeed and...

fix prompt tunning + deepspeed zero3 + checkpoint_saving hang issue

Great, let's adjust the logic for that case specifically then, and leave the already existing base logic in-tact then