here is an image showing the execution

retrying the same run, runs everything perfectly, seems like model serialisation is broken or takes too long and hence mode training fails

I narrowed it down to the Hyperparameters definition. I think it’s an older way of doing things that doesn’t work on newer Flyte installations. The docs say > If you’re using Flytekit version < v1.11.1, you will need to add

from dataclasses_json import dataclass_json

to your imports and decorate your dataclass with

@dataclass_json

. which it seems like flyte-school is using. Just removing

@dataclass_json

didn’t make it work for me, thought. But this test workflow runs fine on flyte v1.15.3:

from dataclasses import dataclass
from datetime import datetime
from typing import Optional

import flytekit as fl


@dataclass
class Datum:
    x: int
    y: Optional[int]


@fl.task()
def multiply(x: int, y: Optional[int]) -> Datum:
    print(f"[{datetime.now()}] multiplying {x} by {y}")
    return Datum(x=x*y, y=None)


@fl.task()
def add(d: Datum) -> Datum:
    print(f"[{datetime.now()}] adding {d.x} by {d.y}")
    return Datum(x=0, y=d.x+9)


@fl.workflow
def wkflw() -> Optional[int]:
    d = multiply(x=2,y=10)
    r = add(d)
    return r.y