erow
### Current Behavior Enter a deadlock. The program cannot go on. ### Expected Behavior _No response_ ### Steps To Reproduce It is hard to reproduce because it works the most...
This pull reduces the memory usage. Now it allocates a max shape to decode the image. #365
I found that `self.loader.graph.allocate_memory` takes a lot of memory. It seems work for reusing the memory. ```python >>> ffcv/loader/epoch_iterator.py if not hasattr(self.loader,'memory_allocations'): print("Memory Allocation 19219.1 MiB for bs=128 and batches_ahead=10")...
Hi, I found a problem of FFCV, which may not the fault of the library, but this library suffers from it. I use the same model to evaluate the accuracy...
Contrastive learning is quite popular now. So, it is important to support multiple views. I implement a simple function to create multiple views, but I still hope FFCV can support...
 This is Figure 7 MIG distribution on dSprites. The MIG score should be non-negative, and the scores are low. I analyze the pre-trained models and get the following results....
I deleted a project, but I found the local files were still there.
Key changes: 1. use crop decode strategy: see in [Towards Pretraining Masked Autoencoders in One Day](https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&ved=2ahUKEwixyurunvCIAxVqXUEAHRrtO1oQFnoECBgQAQ&url=https%3A%2F%2Farxiv.org%2Fabs%2F2404.00509&usg=AOvVaw1TvQxWpX5otBUJb3dlNJlx&opi=89978449) ! this needs turbo-jpeg 3.0 2. save memory allocation, see https://github.com/erow/ffcv/blob/main/ffcv/fields/rgb_image.py#L107 3. transformations Comparison...