Implement a rule-based-reward GRPO loop on a toy arithmetic task rewarding a correct boxed

Super-hard

Reasoning & Test-Time Compute4 / 4

Chapter 144 / 4

Reasoning & Test-Time Compute

Implement a rule-based-reward GRPO loop on a toy arithmetic task rewarding a correct boxed

Super-hard

Implement a rule-based-reward GRPO loop on a toy arithmetic task rewarding a correct boxed answer plus a format reward for <think>/<answer> tags. [DeepSeek]

Target interface

make_actions(target)

Implement the function/class skeleton in the editor. Any correct approach is accepted.

Hints0 / 2

Reference solution

Your own code stays in the editor.

solution.pyPython

Local draft

import numpy as np

def make_actions(target):
    raise NotImplementedError

def reward(parsed, target, well_formed):
    raise NotImplementedError

def softmax(z):
    raise NotImplementedError

⌘/Ctrl↵to submit

AI review

No review yet

Submit your solution for a verdict, a score out of 100 and concrete feedback. Any correct approach passes.

Chapter 144 / 4

Reasoning & Test-Time Compute

Implement a rule-based-reward GRPO loop on a toy arithmetic task rewarding a correct boxed

Super-hard

Implement a rule-based-reward GRPO loop on a toy arithmetic task rewarding a correct boxed answer plus a format reward for <think>/<answer> tags. [DeepSeek]

Target interface

make_actions(target)

Implement the function/class skeleton in the editor. Any correct approach is accepted.

Hints0 / 2

Reference solution

Your own code stays in the editor.