Simon Jégou
Simon Jégou
Hello, I would like to run the different CLIP models on high definition non-square images (e.g. 720p or 1080p). Is there a simple way to do so without deforming the...
Hello, I want to programmatically push a dataset to Darwin using the `RemoteDataset.push` method. For my use case I need to both keep the directory structure (kwarg `preserve_folders=True`) and to...
Hello, Flash attention ([paper](https://arxiv.org/abs/2205.14135)) is a fast, memory-efficient and exact implementation of attention. It could speed up training and inference of ViT-B to ViT-g models by a factor 1.5 to...
Hello, I am working on a practical use-case of Document understanding and wondering if I could leverage models such as Layout(X)LM. The goal is to extract key informations from the...
Hello, Vision transformers in timm currently use a custom implementation of attention instead of `nn.MultiheadAttention`. Pytorch 2.0 will come with [flash attention](https://arxiv.org/abs/2205.14135) which is an exact implementation of attention, but...
ConvNext
Hello, Have you considered using the ConvNext architecture for training DINOv2? ConvNext has shown to have improved performance and lower latency on tasks such as CLIP. For example, in the...
Dear @rolux, Many thanks for porting the work of @Puzer for StyleGan2. I noticed the optimization sometimes fails due to bad initialization of the dlatent variable `W`. I tried to...
Hello, just reopening this [issue](https://github.com/Dao-AILab/flash-attention/issues/330) as I would love to use FA2 on T4 GPUs ^^
### Feature request Include the [wise-ft](https://arxiv.org/abs/2109.01903) method in `LoraModel` models. ### Motivation Wise-ft interpolates between the weights $W_{base}$ of a base model and the weights $W_{ft}$ of a finetuned model...
Hello, Could you add the chat format for the phi3 models ? It is described [here](https://huggingface.co/microsoft/Phi-3-medium-128k-instruct) : ```bash \nQuestion\n\nAnswer\n ``` Maybe something like this in `llama_chat_format.py` ? ```python @register_chat_format("phi3") def...