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, 23:45 UTC
[Misc] Improve memory profiling debug message (#21429)
Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com>
Ye (Charlotte) Qi committed
Jul 26, 2025 at 07:07 UTC
a40a8506df78b965f034aeae260e664c63a7f5d5
1 file changed
+14
-4
vllm/v1/worker/gpu_worker.py
+14
-4
@@ -246,11 +246,21 @@ class Worker(WorkerBase):
246
available_kv_cache_memory = self.requested_memory \
247
- profile_result.non_kv_cache_memory
248
249
+ unrequested_memory = self.init_snapshot.free_memory \
250
+ - self.requested_memory
251
logger.debug(
250
- "Initial free memory: %.2f GiB, free memory: %.2f GiB, "
251
- "requested GPU memory: %.2f GiB",
252
- GiB(self.init_snapshot.free_memory), GiB(free_gpu_memory),
253
- GiB(self.requested_memory))
252
+ "Initial free memory: %.2f GiB; "
253
+ "Requested memory: %.2f (util), %.2f GiB",
254
+ GiB(self.init_snapshot.free_memory),
255
+ self.cache_config.gpu_memory_utilization,
256
+ GiB(self.requested_memory),
257
+ )
258
+ logger.debug(
259
+ "Free memory after profiling: %.2f GiB (total), "
260
+ "%.2f GiB (within requested)",
261
+ GiB(free_gpu_memory),
262
+ GiB(free_gpu_memory - unrequested_memory),
263
+ )
264
logger.debug(profile_result)
265
logger.info("Available KV cache memory: %.2f GiB",
266
GiB(available_kv_cache_memory))