Search results

NaN value in loss when running AMP training with B580

ARCARC GPUARC GPU

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

Status: Open.

#814 In intel/intel-extension-for-pytorch;

· xiangyang-95 opened on Apr 24, 2025

When using xpu backend, stable diffusion give bad result (comparing with 'cpu' backend and also opevino)

ARCARC GPUARC GPU

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

LLM

Status: Open.

#713 In intel/intel-extension-for-pytorch;

· LeptonWu opened on Sep 22, 2024

Corruption with Wuerstchen and Stable Cascade models

ARCARC GPUARC GPU

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

dGPU-Max

Status: Open.

#529 In intel/intel-extension-for-pytorch;

· Disty0 opened on Feb 13, 2024

FP16 output on PVC has randomness

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

dGPU-Max

LLM

Status: Open.

#526 In intel/intel-extension-for-pytorch;

· rnwang04 opened on Feb 6, 2024

Bicubic Interpolation output on XPU differs from CPU.

ARCARC GPUARC GPU

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

Status: Open.

#454 In intel/intel-extension-for-pytorch;

· lalalapotter opened on Oct 30, 2023

torch.multinomial behave unexpectedly on float16 GPU input tensor

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

XPU/GPUXPU/GPU specific issuesXPU/GPU specific issues

Status: Open.

#402 In intel/intel-extension-for-pytorch;

· rnwang04 opened on Aug 11, 2023

Invalid output and errors using model = ipex.optimize(model): split master weight unsupported, Conv BatchNorm folding failed, Linear BatchNorm folding failed

ARCARC GPUARC GPU

CorrectnessOutput incorrect or unacceptable accuracy lossOutput incorrect or unacceptable accuracy loss

Status: Open.

#302 In intel/intel-extension-for-pytorch;

· nathanodle opened on Feb 19, 2023