睡眠模式指南

睡眠模式指南#

概述#

Sleep Mode 是一个用于卸载模型权重并清除 NPU 内存中 KV 缓存的 API。此功能对于强化学习（RL）后训练任务尤其重要，特别是在 PPO、GRPO 或 DPO 等在线算法中。在训练过程中，策略模型通常会使用像 vLLM 这样的推理引擎进行自回归生成，然后进行前向和反向传播以进行优化。

由于生成和训练阶段可能采用不同的模型并行策略，因此在训练过程中及时释放 KV 缓存，甚至卸载存储在 vLLM 内的模型参数变得至关重要。这可以确保内存的高效利用，并避免 NPU 上的资源争用。

快速上手#

With enable_sleep_mode=True, the way we manage memory (malloc, free) in vllm is under a specific memory pool. During model loading and KV cache initialization, we tag the memory as a map: {"weight": data, "kv_cache": data}.

The engine (v0/v1) supports two sleep levels to manage memory during idle periods:

一级睡眠
- 操作：卸载模型权重并清除KV缓存。
- 内存：模型权重被移动到CPU内存；KV缓存被清除。
- 用例：适用于之后需要重复使用同一个模型的情况。
- 注意：请确保有足够的CPU内存来存储模型权重。
二级睡眠
- 操作：同时丢弃模型权重和KV缓存。
- Memory: The content of both the model weights and KV cache is forgotten.
- 用例：当切换到不同的模型或更新当前模型时非常理想。

Since this feature uses the low-level API AscendCL, in order to use sleep mode, you should follow the installation guide and build from source. If you are using v0.7.3, remember to set export COMPILE_CUSTOM_KERNELS=1. For the latest version (v0.11.0), the environment variable COMPILE_CUSTOM_KERNELS will be set to 1 by default while building from source.

用法#

以下是如何使用睡眠模式的一个简单示例。

Offline inference:

import os

import torch
from vllm import LLM, SamplingParams
from vllm.utils import GiB_bytes


os.environ["VLLM_USE_MODELSCOPE"] = "True"
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"

if __name__ == "__main__":
    prompt = "How are you?"

    free, total = torch.npu.mem_get_info()
    print(f"Free memory before sleep: {free / 1024 ** 3:.2f} GiB")
    # record npu memory use baseline in case other process is running
    used_bytes_baseline = total - free
    llm = LLM("Qwen/Qwen2.5-0.5B-Instruct", enable_sleep_mode=True)
    sampling_params = SamplingParams(temperature=0, max_tokens=10)
    output = llm.generate(prompt, sampling_params)

    llm.sleep(level=1)

    free_npu_bytes_after_sleep, total = torch.npu.mem_get_info()
    print(f"Free memory after sleep: {free_npu_bytes_after_sleep / 1024 ** 3:.2f} GiB")
    used_bytes = total - free_npu_bytes_after_sleep - used_bytes_baseline
    # now the memory usage should be less than the model weights
    # (0.5B model, 1GiB weights)
    assert used_bytes < 1 * GiB_bytes

    llm.wake_up()
    output2 = llm.generate(prompt, sampling_params)
    # cmp output
    assert output[0].outputs[0].text == output2[0].outputs[0].text

Online serving:

备注

Considering there may be a risk of malicious access, please make sure you are under a dev-mode, and explicit specify the dev environment VLLM_SERVER_DEV_MODE to expose these endpoints (sleep/wake up).

export VLLM_SERVER_DEV_MODE="1"
export VLLM_WORKER_MULTIPROC_METHOD="spawn"
export VLLM_USE_MODELSCOPE="True"

vllm serve Qwen/Qwen2.5-0.5B-Instruct --enable-sleep-mode

# after serveing is up, post these endpoints

# sleep level 1
curl -X POST http://127.0.0.1:8000/sleep \
    -H "Content-Type: application/json" \
    -d '{"level": "1"}'

curl -X GET http://127.0.0.1:8000/is_sleeping

# sleep level 2
curl -X POST http://127.0.0.1:8000/sleep \
    -H "Content-Type: application/json" \
    -d '{"level": "2"}'

# wake up
curl -X POST http://127.0.0.1:8000/wake_up

# wake up with tag, tags must be in ["weights", "kv_cache"]
curl -X POST "http://127.0.0.1:8000/wake_up?tags=weights"

curl -X GET http://127.0.0.1:8000/is_sleeping

# after sleep and wake up, the serving is still available
curl http://localhost:8000/v1/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "Qwen/Qwen2.5-0.5B-Instruct",
        "prompt": "The future of AI is",
        "max_tokens": 7,
        "temperature": 0
    }'

睡眠模式指南

目录

睡眠模式指南#

概述#

快速上手#

用法#