04 端到端模型整合 课程问题

[Goldsaints]

大佬们，您们好！很感谢您们的分享。我有一个问题，我运行您们分享“04 端到端模型整合 ”课程的代码，出现tvm内部错误，“The truth value of an array with more than one element is ambiguous. Use a.any() or a.all() ”，这是什么原因呢？我试过加上a.any() or a.all(),然后报str 没有 any() 和 all()。
我本地版本是 window10，python3.7，pytorch 1.10+cpu

代码出错的位置：

  @tvm.script.ir_module
  class MyModule:
      @T.prim_func
      def relu0(X: T.Buffer[(1, 128), "float32"],
                Y: T.Buffer[(1, 128), "float32"]):
          T.func_attr({"global_symbol": "relu0", "tir.noalias": True})
          for i, j in T.grid(1, 128):
              with T.block("Y"):
                  vi, vj = T.axis.remap("SS", [i, j])
                  Y[vi, vj] = T.max(X[vi, vj], T.float32(0))
    @T.prim_func
    def linear0(X: T.Buffer[(1, 784), "float32"],
                W: T.Buffer[(128, 784), "float32"],
                B: T.Buffer[(128,), "float32"],
                Z: T.Buffer[(1, 128), "float32"]):
        T.func_attr({"global_symbol": "linear0", "tir.noalias": True})
        Y = T.alloc_buffer((1, 128), "float32")
        for i, j, k in T.grid(1, 128, 784):
            with T.block("Y"):
                vi, vj, vk = T.axis.remap("SSR", [i, j, k])
                with T.init():
                    Y[vi, vj] = T.float32(0)
                Y[vi, vj] = Y[vi, vj] + X[vi, vk] * W[vj, vk]

        for i, j in T.grid(1, 128):
            with T.block("Z"):
                vi, vj = T.axis.remap("SS", [i, j])
                Z[vi, vj] =  Y[vi, vj] + B[vj]

    @T.prim_func
    def linear1(X: T.Buffer[(1,128),"float32"],
                W: T.Buffer[(10,128),"float32"],
                B: T.Buffer[(10,),"float32"],
                Z: T.Buffer[(1,10),"float32"]):
        T.func_attr({"global_symbol":"linear1","tir.noalias":True})
        Y = T.alloc_buffer((1,10),"float32")
        for i, j, k in T.grid(1,10,128):
            with T.block("Y"):
                vi, vj, vk = T.axis.remap("SSR", [i, j, k])
                with T.init():
                    Y[vi, vj] = T.float32(0)
                Y[vi, vj] = Y[vi, vj] + X[vi, vk] * W[vj, vk]
        
        for i, j in T.grid(1,10):
            with T.block("Z"):
                vi, vj = T.axis.remap("SS", [i,j])
                Z[vi, vj] = Y[vi, vj] + B[vj]
    
    @R.function
    def main(x: Tensor((1, 784), "float32"),
             w0: Tensor((128, 784), "float32"),
             b0: Tensor((128,), "float32"),
             w1: Tensor((10, 128), "float32"),
             b1: Tensor((10,), "float32")):
        with R.dataflow():
            lv0 = R.call_tir(linear0, (x, w0, b0), (1, 128), dtype="float32")
            lv1 = R.call_tir(relu0, (lv0,), (1, 128), dtype="float32")
            out = R.call_tir(linear1, (lv1, w1, b1), (1, 10), dtype="float32")
            R.output(out)
        return out

[Hzfengsy]

这是一个旧版本wheel的问题，请考虑更新wheel

[Goldsaints]

好的，我试试，谢谢您！

[Hzfengsy]

目前，windows wheel可能仍存在不稳定性，建议考虑更换Linux系统或者使用colab环境

[Goldsaints]

ok，感谢您的回复！