[auto_scheduler] how to dump the schedule of Auto-Scheduler

Questions
#1

the deepseek suggest me use mod.script() to dump the schedule of Auto-Scheduler, but on machine it shows AttributeError: <class ‘tvm.te.schedule.Schedule’> has no attribute mod, does my tvm version too old ? （It is tvm 0.19.dev0 on my machine）

image
image951×161 8.09 KB

print (tvm.version) 0.19.dev0

import tvm
from tvm import auto_scheduler

# 1. 定义计算任务
@auto_scheduler.register_workload
def matmul():
    A = tvm.te.placeholder((1024, 1024), name='A')
    B = tvm.te.placeholder((1024, 1024), name='B')
    k = tvm.te.reduce_axis((0, 1024), name='k')
    C = tvm.te.compute((1024, 1024), 
                       lambda i, j: tvm.te.sum(A[i, k] * B[k, j], axis=k), 
                       name='C')
    return [A, B, C]

target = tvm.target.Target("llvm")
task = auto_scheduler.SearchTask(func=matmul, args=(), target=target)

# 2. 调优配置
log_file = "matmul.json"
tune_option = auto_scheduler.TuningOptions(
    num_measure_trials=10,
    measure_callbacks=[auto_scheduler.RecordToFile(log_file)],
    verbose=2,  # 输出详细调优信息
)

# 3. 运行自动调优
task.tune(tune_option)

# 4. 应用最佳调度
sch, args = task.apply_best(log_file)

# 5. 打印生成的Schedule
print("======= Auto-Scheduler优化后的调度 =======")
print(sch.mod.script())