nemo_rl/models/generation/vllm/vllm_worker.py (1)

542-554: ⚠️ Potential issue | 🔴 Critical

Bug: iterating over empty metrics dict instead of vllm_prom_metrics.

Line 549 iterates for metric in metrics, but metrics is the empty dict initialized on line 540. This means vllm_prom_metrics is computed but never read, and the method always returns {}. This should iterate over vllm_prom_metrics.

🐛 Proposed fix

             if hasattr(self.llm, "get_metrics"):
                 vllm_prom_metrics = self.llm.get_metrics()
             else:
                 # The AsyncLLM API does not implement get_metrics so we need to call the prometheus API ourselves
                 from vllm.v1.metrics.reader import get_metrics_snapshot
 
                 vllm_prom_metrics = get_metrics_snapshot()
-            for metric in metrics:
+            for metric in vllm_prom_metrics:
                 if hasattr(metric, "values"):
                     metrics[metric.name] = metric.values
                 elif hasattr(metric, "value"):
                     metrics[metric.name] = metric.value