Add deepagents_rl package sources
Browse files
src/deepagents_rl/train/__pycache__/swe_env.cpython-312.pyc
ADDED
|
Binary file (25.2 kB). View file
|
|
|
src/deepagents_rl/train/swe_env.py
CHANGED
|
@@ -513,6 +513,28 @@ class DeepAgentSWEEnv:
|
|
| 513 |
|
| 514 |
self._last_report = report
|
| 515 |
self._last_reward = reward
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 516 |
self._end_trace()
|
| 517 |
return reward
|
| 518 |
|
|
|
|
| 513 |
|
| 514 |
self._last_report = report
|
| 515 |
self._last_reward = reward
|
| 516 |
+
|
| 517 |
+
# W&B: one point per completed rollout (train + ckpt eval)
|
| 518 |
+
try:
|
| 519 |
+
import wandb
|
| 520 |
+
|
| 521 |
+
if wandb.run is not None:
|
| 522 |
+
tools_used = [c.get("name") for c in self._tool_calls]
|
| 523 |
+
wandb.log(
|
| 524 |
+
{
|
| 525 |
+
"rollout/reward_total": float(reward),
|
| 526 |
+
"rollout/pass_frac": float(report.get("pass_frac") or 0.0),
|
| 527 |
+
"rollout/resolved": 1.0 if report.get("resolved") else 0.0,
|
| 528 |
+
"rollout/n_tool_calls": float(len(self._tool_calls)),
|
| 529 |
+
"rollout/instance_id": self._task.get("instance_id") or "",
|
| 530 |
+
"rollout/tools_used": ",".join(tools_used),
|
| 531 |
+
"rollout/resolution": str(report.get("resolution") or ""),
|
| 532 |
+
"rollout/grader_error": str(report.get("error") or ""),
|
| 533 |
+
}
|
| 534 |
+
)
|
| 535 |
+
except Exception:
|
| 536 |
+
pass
|
| 537 |
+
|
| 538 |
self._end_trace()
|
| 539 |
return reward
|
| 540 |
|