fix: change GB to GiB in logging close #14979 (#15807)

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

vllm/v1/core/kv_cache_utils.py

@@ -488,9 +488,9 @@ def check_enough_kv_cache_memory(vllm_config: VllmConfig,
     if needed_memory > available_memory:
         raise ValueError(
             f"To serve at least one request with the models's max seq len "
-            f"({max_model_len}), ({needed_memory/1024/1024/1024:.2f} GB KV "
+            f"({max_model_len}), ({needed_memory/1024/1024/1024:.2f} GiB KV "
             f"cache is needed, which is larger than the available KV cache "
-            f"memory ({available_memory/1024/1024/1024:.2f} GB). Try "
+            f"memory ({available_memory/1024/1024/1024:.2f} GiB). Try "
             f"increasing `gpu_memory_utilization` or decreasing "
             f"`max_model_len` when initializing the engine.")
 

vllm/v1/worker/gpu_model_runner.py

@@ -24,8 +24,8 @@ from vllm.multimodal.utils import group_mm_inputs_by_modality
 from vllm.sampling_params import SamplingType
 from vllm.sequence import IntermediateTensors
 from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, DeviceMemoryProfiler,
-                        LayerBlockType, LazyLoader, cdiv, check_use_alibi,
+                        GiB_bytes, LayerBlockType, LazyLoader, cdiv,
-                        is_pin_memory_available)
+                        check_use_alibi, is_pin_memory_available)
 from vllm.v1.attention.backends.flash_attn import FlashAttentionMetadata
 from vllm.v1.core.encoder_cache_manager import compute_encoder_budget
 from vllm.v1.kv_cache_interface import (FullAttentionSpec, KVCacheConfig,
@@ -1206,8 +1206,8 @@ class GPUModelRunner(LoRAModelRunnerMixin):
                                                   self.device)
             time_after_load = time.perf_counter()
         self.model_memory_usage = m.consumed_memory
-        logger.info("Model loading took %.4f GB and %.6f seconds",
+        logger.info("Model loading took %.4f GiB and %.6f seconds",
-                    self.model_memory_usage / float(2**30),
+                    self.model_memory_usage / GiB_bytes,
                     time_after_load - time_before_load)
 
     def _get_prompt_logprobs_dict(

vllm/worker/model_runner.py

@@ -1143,8 +1143,8 @@ class GPUModelRunnerBase(ModelRunnerBase[TModelInputForGPU]):
             time_after_load = time.perf_counter()
 
         self.model_memory_usage = m.consumed_memory
-        logger.info("Model loading took %.4f GB and %.6f seconds",
+        logger.info("Model loading took %.4f GiB and %.6f seconds",
-                    self.model_memory_usage / float(2**30),
+                    self.model_memory_usage / GiB_bytes,
                     time_after_load - time_before_load)
         if self.prompt_adapter_config:
             self.prompt_adapter_manager = LRUCacheWorkerPromptAdapterManager(

vllm/worker/xpu_model_runner.py

@@ -25,7 +25,7 @@ from vllm.multimodal import (MULTIMODAL_REGISTRY, BatchedTensorInputs,
                              MultiModalRegistry)
 from vllm.sampling_params import SamplingParams
 from vllm.sequence import IntermediateTensors, SequenceGroupMetadata
-from vllm.utils import DeviceMemoryProfiler, make_tensor_with_pad
+from vllm.utils import DeviceMemoryProfiler, GiB_bytes, make_tensor_with_pad
 from vllm.worker.model_runner import AttentionMetadata, SamplingMetadata
 from vllm.worker.model_runner_base import (
     ModelRunnerBase, ModelRunnerInputBase, ModelRunnerInputBuilderBase,
@@ -422,8 +422,8 @@ class XPUModelRunner(ModelRunnerBase[ModelInputForXPUWithSamplingMetadata]):
             self.model = get_model(vllm_config=self.vllm_config)
 
         self.model_memory_usage = m.consumed_memory
-        logger.info("Loading model weights took %.4f GB",
+        logger.info("Loading model weights took %.4f GiB",
-                    self.model_memory_usage / float(2**30))
+                    self.model_memory_usage / GiB_bytes)
 
     def get_model(self) -> nn.Module:
         return self.model