已关闭
[Bug-Report|缺陷反馈]: 1. torch_npu调用dynamic_dual_level_mx_quant不使用smooth_scale时报错 2. aclnn文档需要更新 #1876
gt0cd001创建于 4月1日关闭于 4月2日
4月1日 将 gt0cd001 设为负责人
4月2日 修改标题为 “[Bug-Report|缺陷反馈]: torch_npu调用dynamic_dual_level_mx_quant不使用smooth_scale时报错&aclnn文档需要更新”，原标题为“[Bug-Report|缺陷反馈]: torch_npu调用dynamic_dual_level_mx_quant不使用smooth_scale时报错”
4月2日 修改标题为 “[Bug-Report|缺陷反馈]: 1. torch_npu调用dynamic_dual_level_mx_quant不使用smooth_scale时报错 2. aclnn文档需要更新”，原标题为“[Bug-Report|缺陷反馈]: torch_npu调用dynamic_dual_level_mx_quant不使用smooth_scale时报错&aclnn文档需要更新”
4月2日 关闭了 issue
4月2日 添加了label：resolved
Describe the current behavior / 问题描述 (Mandatory / 必填)
import torch import torch_npu x = torch.randn(16384, 2048, dtype=torch.bfloat16).npu() mul_scale = torch.randn(2048, dtype=torch.bfloat16).npu() torch_npu.npu_dynamic_dual_level_mx_quant(x, smooth_scale=mul_scale) # OK torch_npu.npu_dynamic_dual_level_mx_quant(x, smooth_scale=None) # Inner Error torch_npu.npu_dynamic_dual_level_mx_quant(x) # Inner Error报错信息：
The data type of smooth_scale [UNDEFINED] should match the data type of x [DT_BFLOAT16], please check.Environment / 环境信息 (Mandatory / 必填)
Ascend950PR
Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)
import torch import torch_npu x = torch.randn(16384, 2048, dtype=torch.bfloat16).npu() mul_scale = torch.randn(2048, dtype=torch.bfloat16).npu() torch_npu.npu_dynamic_dual_level_mx_quant(x, smooth_scale=mul_scale) # OK torch_npu.npu_dynamic_dual_level_mx_quant(x, smooth_scale=None) # Inner Error torch_npu.npu_dynamic_dual_level_mx_quant(x) # Inner ErrorDescribe the expected behavior / 预期结果 (Mandatory / 必填)
import torch import torch_npu x = torch.randn(16384, 2048, dtype=torch.bfloat16).npu() mul_scale = torch.randn(2048, dtype=torch.bfloat16).npu() torch_npu.npu_dynamic_dual_level_mx_quant(x, smooth_scale=mul_scale) # OK torch_npu.npu_dynamic_dual_level_mx_quant(x, smooth_scale=None) # OK torch_npu.npu_dynamic_dual_level_mx_quant(x) # OKRelated log / screenshot / 日志 / 截图 (Mandatory / 必填)
The data type of smooth_scale [UNDEFINED] should match the data type of x [DT_BFLOAT16], please check.Special notes for this issue/备注 (Optional / 选填)
Describe the current behavior / 问题描述 (Mandatory / 必填)
aclnnDynamicDualLevelMxQuant.md需要更新