Spaces:
Running
Running
| from dashboard import view | |
| def test_live_view_distinguishes_rollouts_from_updates_and_escapes_code(): | |
| ep = {"task_id": "task-a", "step": 0, "completions": ["<script>bad()</script>"], | |
| "turns": [{"stage": "timed", "correct": True, "speedup": 0.8, "reward": 0.3}]} | |
| snap = {"manifest": {"steps": 16, "groups": {"training_tasks": ["task-a"], "official_heldout": ["task-b"]}}, | |
| "train": [ep], "before": [ep]} | |
| state = view(snap, cost=2.5) | |
| assert state["completed"] == 0 and not state["steps"][0]["updated"] | |
| assert state["kernel_evals"] == 2 and state["cost"] == 2.5 | |
| assert state["groups"]["training_tasks"]["before"]["correct_pct"] == 100 | |
| assert state["groups"]["official_heldout"]["before"]["episodes"] == 0 | |
| assert "<script>bad()" not in state["sections"]["train"] | |
| assert "<script>" in state["sections"]["train"] | |
| state = view(snap, " 6%| | 1/16 [00:53<13:28]\nPHASE after:") | |
| assert state["steps"][0]["updated"] and state["phase"] == "Final evaluation" | |
| state = view(snap, '{"event":"step","task_id":"task-b","turn":1,"stage":"compile"}\n') | |
| assert state["kernel_evals"] == 1 and "task-b" in state["activity"] | |
| snap["trainer"] = {"global_step": 1, "log_history": [{"step": 1, "epoch": 0.25, "loss": -0.02, | |
| "reward": 0.3, "learning_rate": 5e-5, "grad_norm": 0.01}]} | |
| state = view(snap) | |
| assert state["metrics"][1]["loss"] == -0.02 and state["steps"][0]["epoch"] == 0.25 | |
| assert state["steps"][0]["learning_rate"] == 5e-5 | |