[Frontend] Log the maximum supported concurrency (vllm-project#8831)

Signed-off-by: Amit Garg <mitgarg17495@gmail.com>
garg-amit · Oct 28, 2024 · 231166e · 231166e
1 parent 66bf4cb
commit 231166e
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 0 deletions.
diff --git a/vllm/executor/distributed_gpu_executor.py b/vllm/executor/distributed_gpu_executor.py
@@ -56,6 +56,10 @@ def initialize_cache(self, num_gpu_blocks: int,
         # have GPUs.
         logger.info("# GPU blocks: %d, # CPU blocks: %d", num_gpu_blocks,
                     num_cpu_blocks)
+        max_concurrency = (num_gpu_blocks * self.cache_config.block_size /
+                           self.model_config.max_model_len)
+        logger.info("Maximum concurrency for %s tokens per request: %.2fx",
+                    self.model_config.max_model_len, max_concurrency)
 
         self.cache_config.num_gpu_blocks = num_gpu_blocks
         self.cache_config.num_cpu_blocks = num_cpu_blocks

diff --git a/vllm/executor/gpu_executor.py b/vllm/executor/gpu_executor.py
@@ -121,6 +121,10 @@ def initialize_cache(self, num_gpu_blocks: int, num_cpu_blocks) -> None:
         # remains to abstract away the device for non-GPU configurations.
         logger.info("# GPU blocks: %d, # CPU blocks: %d", num_gpu_blocks,
                     num_cpu_blocks)
+        max_concurrency = (num_gpu_blocks * self.cache_config.block_size /
+                           self.model_config.max_model_len)
+        logger.info("Maximum concurrency for %s tokens per request: %.2fx",
+                    self.model_config.max_model_len, max_concurrency)
 
         self.driver_worker.initialize_cache(num_gpu_blocks, num_cpu_blocks)