James Conley

I am running Doccano through docker on ubuntu

I've been hitting this same issue whenever I try to use gradient checkpointing with Lora. Not sure what the underlying problem is but seems to be tied to that.

> Hi ! Can you call `model.enable_input_require_grads()` before `get_peft_model` ? You have to use the `main` branch of `transformers` though (`pip install --upgrade git+https://github.com/huggingface/transformers`) This worked for me, thanks!

Hi! I have the cookies pulled from firefox, but am having trouble adding them. It looks like it's trying to login through the main page still, and is giving the...

Super cool idea! I have a lot of things to do still and only a little time here and there to work on this, but maybe sometime down the road....

Most of the time in processing appears to be the cv2/numpy code for extracting the detected word images from the main image. I swapped this code out for a simple...

> Please fill the issue templates for the specific bug in Accelerate or close this. There is no point opening an issue in different repos if it's all the same....

Here's the full original issue in PEFT ``` The examples/conditional_generation/peft_lora_seq2seq_accelerate_fsdp.py appears to be incompatible with FSDP. It appears others have also noticed the issue. https://github.com/h2oai/h2o-llmstudio/issues/98 I've created a stripped down...

My own example/details # System Info Relevant Package Versions ``` torch==2.0.1 torchaudio==2.0.2+cu118 torchvision==0.15.2+cu118 peft==0.3.0 accelerate==0.20.3 ``` Using base container nvidia/cuda:11.8.0-devel-ubuntu22.04 in docker on a linux box with 2x A6000 GPUs...

> Hello, FSDP with PEFT isn't leading to any memory savings when compared to plain pytorch. see this [pytorch/pytorch#91165 (comment)](https://github.com/pytorch/pytorch/issues/91165#issuecomment-160080533), It also shows how to use FSDP with PEFT nonetheless....