-
From https://github.com/google/jax/issues/20856,
`jax.nn.softmax` returns different results under `jax.jit` because the `jax.jit` version calls oneDNN custom call.
Code:
```
import jax
import j…
-
Hi, CTranslate2 uses oneDNN. oneDNN latest versions has [support for AMD GPU](https://github.com/oneapi-src/oneDNN/tree/master/src/gpu/amd). It [require Intel oneAPI DPC++](https://developer.codeplay.…
-
### 🚀 The feature, motivation and pitch
Feature request: Onednn backend support for quantized operators, INT8 at least.
Motivation: Given the acceptable accuracies from INT8 quantized inference an…
-
PR #366 with [latest commit@ebf149c](https://github.com/oneapi-src/oneDNN/commit/ebf149c48ecdeea5279003c25e20d6d8c7a657a3) and PR #330 with [commit@59e6099](https://github.com/oneapi-src/oneDNN/commit…
-
### 🚀 The feature, motivation and pitch
# Motivation
Intel GPUs could significantly improve the workload performance. As described in [[RFC] Intel GPU Upstreaming](https://github.com/pytorch/pytor…
-
### 🐛 Describe the bug
## Situation
I'm reporting a "bug", however there is no explicit error message and it's a change in the observed behavior during inference. This begun with `torch==2.4.0` but …
-
Hello 1dnn team,
Just asking if it is really expected for ba matmul to be so slow: eg:
```
M=63448
K=640
N=2
tag time
ab 18
ba 1790
```
With benchdnn :
```
wtambellini@lawtambe3 onednn-…
-
From https://github.com/pytorch/pytorch/pull/134282#issuecomment-2307157197, in the aarch64 dashboard results, if we benchmark with fp16, it is 2x~10x slower than bf16, often causing timeout in cases.…
-
Hi dear team,
Is there any other way to accelerate this conv (ic=16, oc=16, height=208, width=32, stride=1, kernel=3) on a single core?
```bash
ONEDNN_VERBOSE=1 numactl -C 1 -m 0 ./benchdnn --m…
-
Is there a method to trace the execution of oneDNN API calls while executing inference with BigDL-LLM?