Max Ren

icon company Facebook icon location San Francisco, CA icon location Dev @ PyTorch

Results 59 comments of Max Ren

QB4W SIMD Packing for neondot

> This is broken for us with many errors of the form: > > ``` > src/qb4-packw/gen/qb4-packw-x16c4-gemm-goi-scalar.c:23:38: error: cast from 'const unsigned char *' to 'unsigned int *' drops const...

Execution of method forward failed with status 0x14

@SS-JIA it looks like they're using the Vulkan Delegate here, any ideas?

Run trunk.yml when we update PT pin

Import and land?

Quantization of yolov5 for XNNPACK backend

Hi @EarthMu, for your first two questions, it is a little difficult for us to tell what exactly is going wrong here. Would it be possible to either share your...

Quantization of yolov5 for XNNPACK backend

Hi Sorry, this issues seems like its coming from torch.export.export. Do you mind cross posting this in PyTorch/PyTorch? cc @angelayi @tugsbayasgalan

Quantization of yolov5 for XNNPACK backend

this looks like the issue is coming from export in PyTorch core, I'm closing the issue here for now. Please feel free to reopen if needed.

ARM build failed with recent XNNPACK update: third_party/XNNPACK/src/reference/unary-elementwise.cc:125:14: error: invalid ‘static_cast’ from type ‘xnn_bfloat16’ to type ‘_Float16’

I believe the upstream is pulled in to resolve this in PyTorch: https://github.com/pytorch/pytorch/pull/141943. I believe the change is only on main, and not in the 2.6.0 release

Pull in 16x4 kernel for QP8

We see some rather significant speed up on prefill performance for Llama Models: ### Before: ``` I 00:00:05.587790 executorch:stats.h:84] Prompt Tokens: 64 Generated Tokens: 63 I 00:00:05.587793 executorch:stats.h:90] Model Load...

Move Config Srcs into build_srcs.bzl

@dsharletg

Move Config Srcs into build_srcs.bzl

Seems like the issue was that src/pack-lh.cc actually uses the unary_elementwise_config and reduce_config. So i kept those around and but those as deps for pack_lh. I removed all others and...