课程4代码问题请教

[the-mountain-rcj]

课程4中MyModule构造代码如下
代码在pycharm中执行出错，colab中成功运行。

from __future__ import annotations

import numpy as np
import tvm
from tvm import relax
from tvm.ir.module import IRModule
from tvm.script import relax as R
from tvm.script import tir as T

import torch
import torchvision

test_data = torchvision.datasets.FashionMNIST(
    root="data",
    train=False,
    download=True,
    transform=torchvision.transforms.ToTensor()
)
test_loader = torch.utils.data.DataLoader(test_data, batch_size=1, shuffle=True)
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']

img, label = next(iter(test_loader))
img = img.reshape(1, 28, 28).numpy()

import matplotlib.pyplot as plt

# plt.figure()
# plt.imshow(img[0])
# plt.colorbar()
# plt.grid(False)
# plt.show()
print("Class:", class_names[label[0]])


# TVM

@tvm.script.ir_module
class MyModule:
    @T.prim_func
    def relu0(X: T.Buffer[(1, 128), "float32"],
              Y: T.Buffer[(1, 128), "float32"]):
        # function attr dict
        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

print(MyModule.script())

[Hzfengsy]

请尝试更新本地mlc的pip package。 mlc的package是nightly build的，建议及时更新