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
[Log] Optimize kv cache memory log from Bytes to GiB (#25204)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Sep 23, 2025 at 12:44 UTC
846197f505fb7b7861a9688e2a09b514b9d3133d
1 file changed
+6
-4
vllm/v1/worker/gpu_worker.py
+6
-4
@@ -383,11 +383,13 @@ class Worker(WorkerBase):
383
f"for non-torch memory, and {GiB(cuda_graph_memory_bytes)} "
384
f"GiB for CUDAGraph memory. Replace gpu_memory_utilization "
385
f"config with `--kv-cache-memory="
386
- f"{kv_cache_memory_bytes_to_requested_limit}` to fit into "
387
- f"requested memory, or `--kv-cache-memory="
388
- f"{kv_cache_memory_bytes_to_gpu_limit}` to fully "
386
+ f"{kv_cache_memory_bytes_to_requested_limit}` "
387
+ f"({GiB(kv_cache_memory_bytes_to_requested_limit)} GiB) to fit "
388
+ f"into requested memory, or `--kv-cache-memory="
389
+ f"{kv_cache_memory_bytes_to_gpu_limit}` "
390
+ f"({GiB(kv_cache_memory_bytes_to_gpu_limit)} GiB) to fully "
391
f"utilize gpu memory. Current kv cache memory in use is "
390
- f"{int(self.available_kv_cache_memory_bytes)} bytes.")
392
+ f"{GiB(self.available_kv_cache_memory_bytes)} GiB.")
393
394
logger.debug(msg)
395