Hi There!
Given the model IR
I want to partition this part
lv4: R.Tensor((batch_size, 2, 3), dtype="float32") = R.reshape(lv3, R.shape([batch_size, 2, 3]))
The pattern language used is as follow
from tvm import tir
batch_size = tir.SizeVar("batch_size", "int64")
def make_pattern():
v_reshaped = is_op("relax.reshape")(wildcard(),is_shape([batch_size,2,3]))
return v_reshaped
pattern_table = [("db.reshape", make_pattern())]
mod=relax.transform.FuseOpsByPattern(pattern_table, annotate_codegen=True)(mod)
mod=relax.transform.MergeCompositeFunctions()(mod)
mod=relax.transform.RunCodegen()(mod)
But this thing failed to partition the model even this thing doesn’t work with wildcard. Thank You