文件最后提交记录最后更新时间
qbmm mxa8w4 modify check of k <= 64 Co-authored-by: yuanshuai000<yuanshuai70@huawei.com> # message auto-generated for no-merge-commit merge: !3063 merge master into master qbmm mxa8w4 modify check of k <= 64 Created-by: yuanshuai000 Commit-by: yuanshuai000 Merged-by: ascend-robot Description: 1. PR描述 修改npu_quant_matmul接口ge_converter中的校验,在mxA8W4数据流中,因为k的约束从 k>=64放开到 k >=8 , 而k<=64时scale是1维tensor,存在transpose->reshape优化导致NZ格式无法正确传递的问题,因此图模式对k的校验改为 k>64 2. 来源 issue: https://gitcode.com/Ascend/torchair/issues/478 3. 涉及文件 python/torchair/_ge_concrete_graph/ge_converter/custom/quantmatmul.py 修改mxA8W4-weightNZ数据流对k的校验为 k > 64 See merge request: Ascend/torchair!30639 小时前
add npu_format_cast_via_cpu + add quantmatmul lite converter # message auto-generated for no-merge-commit merge: !3092 merge master into master add npu_format_cast_via_cpu + add quantmatmul lite converter Created-by: zihan0007 Commit-by: zihan0007 Merged-by: ascend-robot Description: 合入来源: torchair 新增npu_format_cast_via_cpu接口: 新增npu_format_cast_via_cpu接口 支持npuTensor在cpu将数据排布从Nd转为Nz,当前支持int8/uint8 Nd输入转为Fractal_nz输出 ge_pass增加QuantBatchMatmulV3前的Bitcast和Transpose算子互换位置的能力,方便后续Transpose融合 增加lite converter支持QuantBatchMatmulV3 A8W4(int4)场景的converter逻辑 See merge request: Ascend/torchair!30927 天前
Support independent packaging and deployment of npugraph_ex Co-authored-by: xuzhenhua2025<xuzhenhua4@huawei.com> # message auto-generated for no-merge-commit merge: !2680 merge master into master Support independent packaging and deployment of npugraph_ex Created-by: xuzhenhua2025 Commit-by: xuzhenhua2025 Merged-by: ascend-robot Description: 1. npugraph_ex支持独立打包安装 2. npugraph_ex支持在torch_npu编译安装时自动集成到torch_npu/dynamo中 3. npugraph_ex解除对torchair C++层实现的依赖: a. static_kernel中torchair.core.GetSocName() 替换为 torch_npu.npu.get_device_name() b. static_kernel中torchair.core.AclopStartDumpArgs() 替换为torch_npu._C._aclop_start_dump() See merge request: Ascend/torchair!26802 个月前