Simulate an Alternating Self-Play Transcript
Beginner
The problem, from first principles
Self-play generates training data by having two copies of a model (or two roles -- a proposer and a critic, a debater on each side) take turns interacting, with the resulting transcript itself becoming a new training example. Simulating this turn-taking correctly is the harness logic underneath any self-play data-generation pipeline.
The task
Write simulate_self_play(agent_a_moves, agent_b_moves, max_turns). Agents "A" and "B" alternate turns starting with A, each drawing from its own pre-scripted move list in order. Build the (agent_name, move) transcript for up to max_turns turns, stopping early the moment the agent whose turn it is has no move left.
Keywords
production-systemssynthetic-dataself-improvement
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.