From 0b5b5d767e7fdc0b1070b37319de749e46a4d42a Mon Sep 17 00:00:00 2001
From: AlpinDale <52078762+AlpinDale@users.noreply.github.com>
Date: Wed, 9 Oct 2024 07:03:14 +0000
Subject: [PATCH] [Frontend] Log the maximum supported concurrency (#8831)

---
 vllm/executor/distributed_gpu_executor.py | 4 ++++
 vllm/executor/gpu_executor.py             | 4 ++++
 2 files changed, 8 insertions(+)

diff --git a/vllm/executor/distributed_gpu_executor.py b/vllm/executor/distributed_gpu_executor.py
index ad84422e..deb7cb1c 100644
--- a/vllm/executor/distributed_gpu_executor.py
+++ b/vllm/executor/distributed_gpu_executor.py
@@ -56,6 +56,10 @@ class DistributedGPUExecutor(GPUExecutor):
         # 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
index 2185c9cf..ed30d318 100644
--- a/vllm/executor/gpu_executor.py
+++ b/vllm/executor/gpu_executor.py
@@ -121,6 +121,10 @@ class GPUExecutor(ExecutorBase):
         # 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)