test: adapt real-quant assertions and refit diagnostics to vllm 0.25

- the real-quant test scripts assert on the engine log's quantization= tag,
  which on vLLM 0.25 prints the registered NeMo config name
  (e.g. nemo_modelopt_w4a16_nvfp4) instead of modelopt; accept both (the
  w4a16-real GB200 run trained and produced healthy metrics but failed only
  this grep)
- log the full traceback when an IPC weight batch load fails: the refit
  manifest only records the exception message, which for bare assertions
  (e.g. 'AssertionError: ') leaves nothing to diagnose in CI logs

Signed-off-by: Terry Kong <terryk@nvidia.com>

Author: terrykong

nemo_rl/models/generation/vllm/vllm_backend.py

@@ -656,6 +656,10 @@ def update_weights_via_ipc_zmq(self) -> bool:
                         self._load_weights(weights)
                     except Exception as error:
                         batch_error = error
+                        # The manifest only keeps the exception message; log
+                        # the full traceback so loader failures stay
+                        # diagnosable from worker logs.
+                        logger.exception("IPC weight batch load failed")
                     finally:
                         # Synchronize before releasing or ACKing an IPC allocation,
                         # including when a loader failed after scheduling CUDA work.

tests/test_suites/llm/grpo-nanov3-30ba3b-4n4g-megatron-qa-nvfp4-w4a16-real.sh

@@ -38,7 +38,7 @@ uv run --no-sync tests/json_dump_tb_logs.py "$LOG_DIR" --output_path "$JSON_METR
 
 grep -q "VllmQuantInternalWorkerExtension" "$RUN_LOG"
 grep -q "Detected ModelOpt NVFP4 checkpoint" "$RUN_LOG"
-grep -q "quantization=modelopt" "$RUN_LOG"
+grep -qE "quantization=(nemo_)?modelopt" "$RUN_LOG"  # vLLM 0.25 logs the registered NeMo quant name (e.g. nemo_modelopt_w4a16_nvfp4)
 assert_not_grep "FakeQuantWorker" "$RUN_LOG" \
     "Real-quant run unexpectedly used FakeQuantWorker"
 assert_not_grep "VLLM_QUANT_CFG" "$RUN_LOG" \

tests/test_suites/llm/grpo-nemotron3-super-120BA12B-16n4g-megatron-qa-nvfp4-w4a16-real-300step.sh

@@ -40,7 +40,7 @@ uv run --no-sync tests/json_dump_tb_logs.py "$LOG_DIR" --output_path "$JSON_METR
 # Real-quant rollout must go through the ModelOpt NVFP4 vLLM kernel path.
 grep -q "VllmQuantInternalWorkerExtension" "$RUN_LOG"
 grep -q "Detected ModelOpt NVFP4 checkpoint" "$RUN_LOG"
-grep -q "quantization=modelopt" "$RUN_LOG"
+grep -qE "quantization=(nemo_)?modelopt" "$RUN_LOG"  # vLLM 0.25 logs the registered NeMo quant name (e.g. nemo_modelopt_w4a16_nvfp4)
 grep -q "nvfp4_experts_weightonly.yaml" "$RUN_LOG"
 assert_not_grep "FakeQuantWorker" "$RUN_LOG" \
     "Real-quant run unexpectedly used FakeQuantWorker"

tests/test_suites/llm/grpo-nemotron3-super-120BA12B-16n4g-megatron-qa-nvfp4-w4a4-real-300step.sh

@@ -40,7 +40,7 @@ uv run --no-sync tests/json_dump_tb_logs.py "$LOG_DIR" --output_path "$JSON_METR
 # Real-quant rollout must go through the ModelOpt NVFP4 vLLM kernel path.
 grep -q "VllmQuantInternalWorkerExtension" "$RUN_LOG"
 grep -q "Detected ModelOpt NVFP4 checkpoint" "$RUN_LOG"
-grep -q "quantization=modelopt" "$RUN_LOG"
+grep -qE "quantization=(nemo_)?modelopt" "$RUN_LOG"  # vLLM 0.25 logs the registered NeMo quant name (e.g. nemo_modelopt_w4a16_nvfp4)
 grep -q "examples/modelopt/quant_configs/nvfp4_experts.yaml" "$RUN_LOG"
 assert_not_grep "FakeQuantWorker" "$RUN_LOG" \
     "Real-quant run unexpectedly used FakeQuantWorker"

tests/test_suites/llm/grpo-qwen3-30ba3b-4n4g-megatron-qa-nvfp4-w4a4-real.sh

@@ -40,7 +40,7 @@ uv run --no-sync tests/json_dump_tb_logs.py "$LOG_DIR" --output_path "$JSON_METR
 
 grep -q "VllmQuantInternalWorkerExtension" "$RUN_LOG"
 grep -q "Detected ModelOpt NVFP4 checkpoint" "$RUN_LOG"
-grep -q "quantization=modelopt" "$RUN_LOG"
+grep -qE "quantization=(nemo_)?modelopt" "$RUN_LOG"  # vLLM 0.25 logs the registered NeMo quant name (e.g. nemo_modelopt_w4a16_nvfp4)
 assert_not_grep "FakeQuantWorker" "$RUN_LOG" \
     "Real-quant run unexpectedly used FakeQuantWorker"
 assert_not_grep "VLLM_QUANT_CFG" "$RUN_LOG" \