Birch-san

icon indicating a website link https://birchlabs.co.uk

icon company some financial technology company icon location United Kingdom icon location A Very Full Stack dev. I like low-level (C++, WASM, LLVM, Graal, DTrace, BPF) and devops (Docker, networking, monitoring, logging). I like performance + scale.

Results 34 issues of Birch-san

Cmake v3.30.2 cudart link error

6 comment

As there wasn't a torch 2.4.0 wheel, I tried building NATTEN myself. It didn't go as smoothly as usual. Most problems were due to cmake giving misleading/incomplete error messages. These...

Support batch size 0

_Downstream PyTorch issue:_ https://github.com/pytorch/pytorch/issues/133780 **Describe the bug** cuDNN frontend rejects batch_size=0 input with `CUDNN_STATUS_BAD_PARAM` **Expected behavior** cuDNN should return to me a tensor [0, num_head, sequence_length, dims_per_head] something like that,...

segfault when latent height or width is not divisible by 4

## Problem If I trace my model like so (weird shape then nice shape), everything is fine: ```python model(randn(1, 4, 150, 157)) model(randn(1, 4, 64, 64)) ``` whereas if I...

Swapping between two different AttentionProcessors

1 comment

thanks for the great work on stable-fast. it compiles quickly and boosts speed a lot. is it possible to support two different compilation graphs? for example swapping the SDXL UNet's...