A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 07:15 UTC
Fix ValueError: Missing value for tag key(s): model_name,engine. (#19113)
Signed-off-by: Seiji Eicher <seiji@anyscale.com>
Seiji Eicher committed
Jun 4, 2025 at 02:10 UTC
2669a0d7b518371bb1d950425bd64a320010733f
2 files changed
+14
-1
tests/v1/metrics/test_ray_metrics.py
+4
-1
@@ -47,12 +47,15 @@ def test_engine_log_metrics_ray(
47
engine_args, stat_loggers=[RayPrometheusStatLogger])
48
49
for i, prompt in enumerate(example_prompts):
50
- engine.generate(
50
+ results = engine.generate(
51
request_id=f"request-id-{i}",
52
prompt=prompt,
53
sampling_params=SamplingParams(max_tokens=max_tokens),
54
)
55
56
+ async for _ in results:
57
+ pass
58
+
59
# Create the actor and call the async method
60
actor = EngineTestActor.remote() # type: ignore[attr-defined]
61
ray.get(actor.run.remote())
vllm/v1/metrics/ray_wrappers.py
+10
@@ -31,6 +31,16 @@ class RayPrometheusMetric:
31
32
self.metric.set_default_tags(labelskwargs)
33
34
+ if labels:
35
+ if len(labels) != len(self.metric._tag_keys):
36
+ raise ValueError(
37
+ "Number of labels must match the number of tag keys. "
38
+ f"Expected {len(self.metric._tag_keys)}, got {len(labels)}"
39
+ )
40
+
41
+ self.metric.set_default_tags(
42
+ dict(zip(self.metric._tag_keys, labels)))
43
+
44
return self
45
46