vllm - 💡(How to fix) Fix [Bug]: vLLM only prints access logs, not performance statistics logs (v0.1.dev15830+g8d599d76a with deepseek-V4-flash) [2 comments, 2 participants]

Your current environment

Environment Information: • vLLM version: 0.1.dev15830+g8d599d76a

• Model: deepseek-V4-flash

• CUDA version: cu129

• GPU: H800

🐛 Describe the bug

Description:
When I run the vLLM API server with version 0.1.dev15830+g8d599d76a and the deepseek-V4-flash model, the logs only show basic access logs like

INFO: 127.0.0.1:32768 0 "POST v1/completions HTTP/1.1" 200 OK

But I'm not seeing the periodic performance statistics logs that include throughput, GPU KV cache usage, etc., for example:

13:57:45 Avg prompt throughput: 2598.6 tokens/s, Avg generation throughput: 1684.2 tokens/s, Running: 2 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.3%, Prefix cache hit rate: 96.9% 13:57:55 Avg prompt throughput: 83.6 tokens/s, Avg generation throughput: 1.8 tokens/s, Running: 11 reqs, Waiting: 0 reqs, GPU KV cache usage: 1.3%, Prefix cache hit rate: 96.9% 13:58:05 Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 11 reqs, Waiting: 0 reqs, GPU KV cache usage: 1.3%, Prefix cache hit rate: 96.9%

start code: VLLM_ENGINE_READY_TIMEOUT_S=1200 nohup vllm serve /mnt/algorithms/DeepSeek-V4-Flash --port 8000 --served-model-name DEEPSEEK-V4-284B-FLASH-TEST --trust-remote-code --kv-cache-dtype fp8 --block-size 256 --enable-expert-parallel --data-parallel-size 8 --compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE","custom_ops":["all"]}' --tokenizer-mode deepseek_v4 --tool-call-parser deepseek_v4 --enable-auto-tool-choice --reasoning-parser deepseek_v4 &

Steps to Reproduce:

1. Start vLLM API server with deepseek-V4-flash model
2. Send inference requests to the server
3. Check the console/log output

Expected Behavior: Periodic performance statistics logs should appear every few seconds showing throughput, GPU KV cache usage, request counts, etc.

Actual Behavior: Only basic HTTP access logs are printed, no performance statistics.

Additional Context: I've tried setting environment variables like VLLM_LOGGING_LEVEL=INFO and VLLM_LOG_STATS_INTERVAL=1, but still only get access logs. Are there specific configurations needed for this version or with the deepseek-V4-flash model to enable performance statistics logging?

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.