🐛 fix torch memory profiling (vllm-project#9516)

Signed-off-by: Joe Runde <[email protected]>
Signed-off-by: Amit Garg <[email protected]>

tests/quantization/test_bitsandbytes.py

@@ -107,8 +107,7 @@ def validate_generated_texts(hf_runner,
                      quantization='bitsandbytes',
                      load_format='bitsandbytes',
                      tensor_parallel_size=vllm_tp_size,
-                     enforce_eager=False,
-                     gpu_memory_utilization=0.8) as llm:
+                     enforce_eager=False) as llm:
         vllm_outputs = llm.generate_greedy(prompts, 8)
         vllm_logs = log_generated_texts(prompts, vllm_outputs, "VllmRunner")
 

tests/worker/test_profile.py

@@ -54,16 +54,17 @@ def mock_mem_info():
         gpu_blocks, _ = worker.determine_num_available_blocks()
 
     # Peak vram usage by torch should be 0.7077 GiB
-    # Non-torch allocations should be 0.0079 GiB
+    # No memory should be allocated outside of torch
     # 9.0 GiB should be the utilization target
-    # 8.2843 GiB should be available for the KV cache
+    # 8.2923 GiB should be available for the KV cache
     block_size = CacheEngine.get_cache_block_size(
         engine_config.cache_config, engine_config.model_config,
         engine_config.parallel_config)
 
-    expected_blocks = (8.2843 * 1024**3) // block_size
+    expected_blocks = (8.2923 * 1024**3) // block_size
 
     # Check within a small tolerance for portability
     # Hardware, kernel, or dependency changes could all affect memory
-    # utilization
-    assert abs(gpu_blocks - expected_blocks) < 5
+    # utilization.
+    # A 10 block tolerance here should be about 6MB of wiggle room.
+    assert abs(gpu_blocks - expected_blocks) < 10

vllm/worker/worker.py

@@ -232,10 +232,11 @@ def determine_num_available_blocks(self) -> Tuple[int, int]:
         # gpu outside of `torch`. NCCL operations, for example, can use a few
         # GB during a forward pass
         torch.cuda.empty_cache()
-        # After emptying the torch cache, any other increase in gpu ram should
-        # be from non-torch allocations.
-        non_torch_allocations = free_memory_pre_profile - \
-            torch.cuda.mem_get_info()[0]
+        torch_allocated_bytes = torch.cuda.memory_stats(
+        )["allocated_bytes.all.current"]
+        total_allocated_bytes = torch.cuda.mem_get_info(
+        )[1] - torch.cuda.mem_get_info()[0]
+        non_torch_allocations = total_allocated_bytes - torch_allocated_bytes
         if non_torch_allocations > 0:
             peak_memory += non_torch_allocations
 
@@ -259,10 +260,12 @@ def determine_num_available_blocks(self) -> Tuple[int, int]:
         logger.info(
             "Memory profiling results: total_gpu_memory=%.2fGiB"
             " initial_memory_usage=%.2fGiB peak_torch_memory=%.2fGiB"
+            " memory_usage_post_profile=%.2fGib"
             " non_torch_memory=%.2fGiB kv_cache_size=%.2fGiB"
             " gpu_memory_utilization=%.2f", total_gpu_memory / (1024**3),
             (total_gpu_memory - free_memory_pre_profile) / (1024**3),
             (peak_memory - non_torch_allocations) / (1024**3),
+            total_allocated_bytes / (1024**3),
             non_torch_allocations / (1024**3),
             available_kv_cache_memory / (1024**3),
             self.cache_config.gpu_memory_utilization)