Closed
[Requirement|需求建议]: 新增MxToBlockMxQuant torch_extension接口 #4213
yin-pengAdd on  Jul 21Closed at  Jul 24
yin-pengCollaborators
Jul 21 Create

Thanks for sending an requirement! Please fill in the following template to help quickly solve your problem.

Backgroud(背景信息)

大模型推理中MX (Microscaling) 格式量化转换的核心算子,用于将 FP4 低精度数据结合E8M0缩放系数转换为FP8分块量化格式。
本次则是新增torch.ops.cann_ops_nn.mx_to_block_mx_quant接口。

Origin(信息来源)

PanGu

Benefit / Necessity (价值/作用)

Design(设计方案)

likedislike
Yyin-pengCollaborators
Jul 21 Added label: requirement
yuning_chenyuning_chenCollaborators
Jul 21 Assign yin-peng as responsible person
CANN-robotCANN-robotCollaborators
Jul 24 Close issue
CANN-robotCANN-robotCollaborators
Jul 24 Added label: resolved