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 14, 05:30 UTC
[Bug][Benchmark] Fix duplicate req in oversampling (#26140)
Signed-off-by: Ekagra Ranjan <3116519+ekagra-ranjan@users.noreply.github.com> Co-authored-by: Roger Wang <hey@rogerw.io>
Ekagra Ranjan committed
Oct 2, 2025 at 22:55 UTC
ad2d78801642b50b65fe5761c941ed97ed1285fd
1 file changed
+11
-5
vllm/benchmarks/datasets.py
+11
-5
@@ -235,16 +235,22 @@ class BenchmarkDataset(ABC):
235
236
if len(requests) < num_requests:
237
random.seed(self.random_seed)
238
- additional = deepcopy(
239
- random.choices(requests, k=num_requests - len(requests))
240
- )
241
- for i in range(len(additional)):
242
- req = additional[i]
238
+ needed = num_requests - len(requests)
239
+ additional = []
240
+ for i in range(needed):
241
+ req = deepcopy(random.choice(requests))
242
req.request_id = request_id_prefix + str(len(requests) + i)
243
+ additional.append(req)
244
requests.extend(additional)
245
logger.info("Oversampled requests to reach %d total samples.",
246
num_requests)
247
248
+ ids = [req.request_id for req in requests]
249
+ if len(ids) != len(set(ids)):
250
+ raise ValueError("Duplicate request_id found in the sampled "
251
+ "requests. Please ensure that each request_id "
252
+ "is unique.")
253
+
254
255
# -----------------------------------------------------------------------------
256
# Utility Functions and Global Caches