xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 05:55:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Docs] Rewrite Distributed Inference and Serving guide (#20593)

Browse Source 
Signed-off-by: Ricardo Decal <rdecal@anyscale.com>
Co-authored-by: Simon Mo <simon.mo@hey.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Ricardo Decal 2025-07-24 08:13:05 -07:00 committed by GitHub
parent cdb79ee63d
commit 684174115d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 79 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

131
docs/serving/distributed_serving.md
View File

@ -1,31 +1,38 @@
# Distributed Inference and Serving
# Distributed inference and serving
## How to decide the distributed inference strategy?
## Distributed inference strategies for a single-model replica
Before going into the details of distributed inference and serving, let's first make it clear when to use distributed inference and what are the strategies available. The common practice is:
To choose a distributed inference strategy for a single-model replica, use the following guidelines:
- **Single GPU (no distributed inference)**: If your model fits in a single GPU, you probably don't need to use distributed inference. Just use the single GPU to run the inference.
- **Single-Node Multi-GPU (tensor parallel inference)**: If your model is too large to fit in a single GPU, but it can fit in a single node with multiple GPUs, you can use tensor parallelism. The tensor parallel size is the number of GPUs you want to use. For example, if you have 4 GPUs in a single node, you can set the tensor parallel size to 4.
- **Multi-Node Multi-GPU (tensor parallel plus pipeline parallel inference)**: If your model is too large to fit in a single node, you can use tensor parallel together with pipeline parallelism. The tensor parallel size is the number of GPUs you want to use in each node, and the pipeline parallel size is the number of nodes you want to use. For example, if you have 16 GPUs in 2 nodes (8 GPUs per node), you can set the tensor parallel size to 8 and the pipeline parallel size to 2.
- **Single GPU (no distributed inference):** if the model fits on a single GPU, distributed inference is probably unnecessary. Run inference on that GPU.
- **Single-node multi-GPU using tensor parallel inference:** if the model is too large for a single GPU but fits on a single node with multiple GPUs, use *tensor parallelism*. For example, set `tensor_parallel_size=4` when using a node with 4 GPUs.
- **Multi-node multi-GPU using tensor parallel and pipeline parallel inference:** if the model is too large for a single node, combine *tensor parallelism* with *pipeline parallelism*. Set `tensor_parallel_size` to the number of GPUs per node and `pipeline_parallel_size` to the number of nodes. For example, set `tensor_parallel_size=8` and `pipeline_parallel_size=2` when using 2 nodes with 8 GPUs per node.
In short, you should increase the number of GPUs and the number of nodes until you have enough GPU memory to hold the model. The tensor parallel size should be the number of GPUs in each node, and the pipeline parallel size should be the number of nodes.
Increase the number of GPUs and nodes until there is enough GPU memory for the model. Set `tensor_parallel_size` to the number of GPUs per node and `pipeline_parallel_size` to the number of nodes.
After adding enough GPUs and nodes to hold the model, you can run vLLM first, which will print some logs like `# GPU blocks: 790`. Multiply the number by `16` (the block size), and you can get roughly the maximum number of tokens that can be served on the current configuration. If this number is not satisfying, e.g. you want higher throughput, you can further increase the number of GPUs or nodes, until the number of blocks is enough.
After you provision sufficient resources to fit the model, run `vllm`. Look for log messages like:
!!! note
There is one edge case: if the model fits in a single node with multiple GPUs, but the number of GPUs cannot divide the model size evenly, you can use pipeline parallelism, which splits the model along layers and supports uneven splits. In this case, the tensor parallel size should be 1 and the pipeline parallel size should be the number of GPUs.
```text
INFO 07-23 13:56:04 [kv_cache_utils.py:775] GPU KV cache size: 643,232 tokens
INFO 07-23 13:56:04 [kv_cache_utils.py:779] Maximum concurrency for 40,960 tokens per request: 15.70x
```
### Distributed serving of MoE (Mixture of Experts) models
The `GPU KV cache size` line reports the total number of tokens that can be stored in the GPU KV cache at once. The `Maximum concurrency` line provides an estimate of how many requests can be served concurrently if each request requires the specified number of tokens (40,960 in the example above). The tokens-per-request number is taken from the model configuration's maximum sequence length, `ModelConfig.max_model_len`. If these numbers are lower than your throughput requirements, add more GPUs or nodes to your cluster.
It is often advantageous to exploit the inherent parallelism of experts by using a separate parallelism strategy for the expert layers. vLLM supports large-scale deployment combining Data Parallel attention with Expert or Tensor Parallel MoE layers. See the page on [Data Parallel Deployment](data_parallel_deployment.md) for more information.
!!! note "Edge case: uneven GPU splits"
If the model fits within a single node but the GPU count doesn't evenly divide the model size, enable pipeline parallelism, which splits the model along layers and supports uneven splits. In this scenario, set `tensor_parallel_size=1` and `pipeline_parallel_size` to the number of GPUs. Furthermore, if the GPUs on the node do not have NVLINK interconnect (e.g. L40S), leverage pipeline parallelism instead of tensor parallelism for higher throughput and lower communication overhead.
## Running vLLM on a single node
### Distributed serving of *Mixture of Experts* (*MoE*) models
vLLM supports distributed tensor-parallel and pipeline-parallel inference and serving. Currently, we support [Megatron-LM's tensor parallel algorithm](https://arxiv.org/pdf/1909.08053.pdf). We manage the distributed runtime with either [Ray](https://github.com/ray-project/ray) or python native multiprocessing. Multiprocessing can be used when deploying on a single node, multi-node inference currently requires Ray.
It's often advantageous to exploit the inherent parallelism of experts by using a separate parallelism strategy for the expert layers. vLLM supports large-scale deployment combining Data Parallel attention with Expert or Tensor Parallel MoE layers. For more information, see [Data Parallel Deployment](data_parallel_deployment.md).
Multiprocessing will be used by default when not running in a Ray placement group and if there are sufficient GPUs available on the same node for the configured `tensor_parallel_size`, otherwise Ray will be used. This default can be overridden via the `LLM` class `distributed_executor_backend` argument or `--distributed-executor-backend` API server argument. Set it to `mp` for multiprocessing or `ray` for Ray. It's not required for Ray to be installed for the multiprocessing case.
## Single-node deployment
To run multi-GPU inference with the `LLM` class, set the `tensor_parallel_size` argument to the number of GPUs you want to use. For example, to run inference on 4 GPUs:
vLLM supports distributed tensor-parallel and pipeline-parallel inference and serving. The implementation includes [Megatron-LM's tensor parallel algorithm](https://arxiv.org/pdf/1909.08053.pdf).
The default distributed runtimes are [Ray](https://github.com/ray-project/ray) for multi-node inference and native Python `multiprocessing` for single-node inference. You can override the defaults by setting `distributed_executor_backend` in the `LLM` class or `--distributed-executor-backend` in the API server. Use `mp` for `multiprocessing` or `ray` for Ray.
For multi-GPU inference, set `tensor_parallel_size` in the `LLM` class to the desired GPU count. For example, to run inference on 4 GPUs:
```python
from vllm import LLM
@ -33,84 +40,96 @@ llm = LLM("facebook/opt-13b", tensor_parallel_size=4)
output = llm.generate("San Francisco is a")
```
To run multi-GPU serving, pass in the `--tensor-parallel-size` argument when starting the server. For example, to run API server on 4 GPUs:
For multi-GPU serving, include `--tensor-parallel-size` when starting the server. For example, to run the API server on 4 GPUs:
```bash
vllm serve facebook/opt-13b \
--tensor-parallel-size 4
```
You can also additionally specify `--pipeline-parallel-size` to enable pipeline parallelism. For example, to run API server on 8 GPUs with pipeline parallelism and tensor parallelism:
To enable pipeline parallelism, add `--pipeline-parallel-size`. For example, to run the API server on 8 GPUs with pipeline parallelism and tensor parallelism:
```bash
# Eight GPUs total
vllm serve gpt2 \
--tensor-parallel-size 4 \
--pipeline-parallel-size 2
```
## Running vLLM on multiple nodes
## Multi-node deployment
If a single node does not have enough GPUs to hold the model, you can run the model using multiple nodes. It is important to make sure the execution environment is the same on all nodes, including the model path, the Python environment. The recommended way is to use docker images to ensure the same environment, and hide the heterogeneity of the host machines via mapping them into the same docker configuration.
If a single node lacks sufficient GPUs to hold the model, deploy vLLM across multiple nodes. Multi-node deployments require Ray as the runtime engine. Ensure that every node provides an identical execution environment, including the model path and Python packages. Using container images is recommended because they provide a convenient way to keep environments consistent and to hide host heterogeneity.
The first step, is to start containers and organize them into a cluster. We have provided the helper script <gh-file:examples/online_serving/run_cluster.sh> to start the cluster. Please note, this script launches docker without administrative privileges that would be required to access GPU performance counters when running profiling and tracing tools. For that purpose, the script can have `CAP_SYS_ADMIN` to the docker container by using the `--cap-add` option in the docker run command.
### Ray cluster setup with containers
Pick a node as the head node, and run the following command:
The helper script `<gh-file:examples/online_serving/run_cluster.sh>` starts containers across nodes and initializes Ray. By default, the script runs Docker without administrative privileges, which prevents access to the GPU performance counters when profiling or tracing. To enable admin privileges, add the `--cap-add=CAP_SYS_ADMIN` flag to the Docker command.
Choose one node as the head node and run:
```bash
bash run_cluster.sh \
vllm/vllm-openai \
ip_of_head_node \
<HEAD_NODE_IP> \
--head \
/path/to/the/huggingface/home/in/this/node \
-e VLLM_HOST_IP=ip_of_this_node
-e VLLM_HOST_IP=<HEAD_NODE_IP>
```
On the rest of the worker nodes, run the following command:
On each worker node, run:
```bash
bash run_cluster.sh \
vllm/vllm-openai \
ip_of_head_node \
<HEAD_NODE_IP> \
--worker \
/path/to/the/huggingface/home/in/this/node \
-e VLLM_HOST_IP=ip_of_this_node
-e VLLM_HOST_IP=<WORKER_NODE_IP>
```
Then you get a ray cluster of **containers**. Note that you need to keep the shells running these commands alive to hold the cluster. Any shell disconnect will terminate the cluster. In addition, please note that the argument `ip_of_head_node` should be the IP address of the head node, which is accessible by all the worker nodes. The IP addresses of each worker node should be specified in the `VLLM_HOST_IP` environment variable, and should be different for each worker node. Please check the network configuration of your cluster to make sure the nodes can communicate with each other through the specified IP addresses.
Note that `VLLM_HOST_IP` is unique for each worker. Keep the shells running these commands open; closing any shell terminates the cluster. Ensure that all nodes can communicate with each other through their IP addresses.
!!! warning
It is considered best practice to set `VLLM_HOST_IP` to an address on a private network segment for the vLLM cluster. The traffic sent here is not encrypted. The endpoints are also exchanging data in a format that could be exploited to execute arbitrary code should a malicious party gain access to the network. Please ensure that this network is not reachable by any untrusted parties.
!!! warning "Network security"
For security, set `VLLM_HOST_IP` to an address on a private network segment. Traffic sent over this network is unencrypted, and the endpoints exchange data in a format that can be exploited to execute arbitrary code if an adversary gains network access. Ensure that untrusted parties cannot reach the network.
!!! warning
Since this is a ray cluster of **containers**, all the following commands should be executed in the **containers**, otherwise you are executing the commands on the host machine, which is not connected to the ray cluster. To enter the container, you can use `docker exec -it node /bin/bash`.
From any node, enter a container and run `ray status` and `ray list nodes` to verify that Ray finds the expected number of nodes and GPUs.
Then, on any node, use `docker exec -it node /bin/bash` to enter the container, execute `ray status` and `ray list nodes` to check the status of the Ray cluster. You should see the right number of nodes and GPUs.
!!! tip
Alternatively, set up the Ray cluster using KubeRay. For more information, see [KubeRay vLLM documentation](https://docs.ray.io/en/latest/cluster/kubernetes/examples/vllm-rayservice.html).
After that, on any node, use `docker exec -it node /bin/bash` to enter the container again. **In the container**, you can use vLLM as usual, just as you have all the GPUs on one node: vLLM will be able to leverage GPU resources of all nodes in the Ray cluster, and therefore, only run the `vllm` command on this node but not other nodes. The common practice is to set the tensor parallel size to the number of GPUs in each node, and the pipeline parallel size to the number of nodes. For example, if you have 16 GPUs in 2 nodes (8 GPUs per node), you can set the tensor parallel size to 8 and the pipeline parallel size to 2:
### Running vLLM on a Ray cluster
!!! tip
If Ray is running inside containers, run the commands in the remainder of this guide _inside the containers_, not on the host. To open a shell inside a container, connect to a node and use `docker exec -it <container_name> /bin/bash`.
Once a Ray cluster is running, use vLLM as you would in a single-node setting. All resources across the Ray cluster are visible to vLLM, so a single `vllm` command on a single node is sufficient.
The common practice is to set the tensor parallel size to the number of GPUs in each node, and the pipeline parallel size to the number of nodes. For example, if you have 16 GPUs across 2 nodes (8 GPUs per node), set the tensor parallel size to 8 and the pipeline parallel size to 2:
```bash
vllm serve /path/to/the/model/in/the/container \
--tensor-parallel-size 8 \
--pipeline-parallel-size 2
vllm serve /path/to/the/model/in/the/container \
--tensor-parallel-size 8 \
--pipeline-parallel-size 2
```
You can also use tensor parallel without pipeline parallel, just set the tensor parallel size to the number of GPUs in the cluster. For example, if you have 16 GPUs in 2 nodes (8 GPUs per node), you can set the tensor parallel size to 16:
Alternatively, you can set `tensor_parallel_size` to the total number of GPUs in the cluster:
```bash
vllm serve /path/to/the/model/in/the/container \
--tensor-parallel-size 16
```
To make tensor parallel performant, you should make sure the communication between nodes is efficient, e.g. using high-speed network cards like InfiniBand. To correctly set up the cluster to use InfiniBand, append additional arguments like `--privileged -e NCCL_IB_HCA=mlx5` to the `run_cluster.sh` script. Please contact your system administrator for more information on how to set up the flags. One way to confirm if the InfiniBand is working is to run vLLM with `NCCL_DEBUG=TRACE` environment variable set, e.g. `NCCL_DEBUG=TRACE vllm serve ...` and check the logs for the NCCL version and the network used. If you find `[send] via NET/Socket` in the logs, it means NCCL uses raw TCP Socket, which is not efficient for cross-node tensor parallel. If you find `[send] via NET/IB/GDRDMA` in the logs, it means NCCL uses InfiniBand with GPUDirect RDMA, which is efficient.
## Troubleshooting distributed deployments
### GPUDirect RDMA
To make tensor parallelism performant, ensure that communication between nodes is efficient, for example, by using high-speed network cards such as InfiniBand. To set up the cluster to use InfiniBand, append additional arguments like `--privileged -e NCCL_IB_HCA=mlx5` to the `run_cluster.sh` script. Contact your system administrator for more information about the required flags. One way to confirm if InfiniBand is working is to run `vllm` with the `NCCL_DEBUG=TRACE` environment variable set, for example `NCCL_DEBUG=TRACE vllm serve ...`, and check the logs for the NCCL version and the network used. If you find `[send] via NET/Socket` in the logs, NCCL uses a raw TCP socket, which is not efficient for cross-node tensor parallelism. If you find `[send] via NET/IB/GDRDMA` in the logs, NCCL uses InfiniBand with GPUDirect RDMA, which is efficient.
To enable GPUDirect RDMA with vLLM, specific configuration tweaks are needed. This setup ensures:
## Enabling GPUDirect RDMA
- `IPC_LOCK` Security Context: Add the `IPC_LOCK` capability to the container’s security context to lock memory pages and prevent swapping to disk.
- Shared Memory with `/dev/shm`: Mount `/dev/shm` in the pod spec to provide shared memory for IPC.
To enable GPUDirect RDMA with vLLM, configure the following settings:
When using Docker, you can set up the container as follows:
- `IPC_LOCK` security context: add the `IPC_LOCK` capability to the container's security context to lock memory pages and prevent swapping to disk.
- Shared memory with `/dev/shm`: mount `/dev/shm` in the pod spec to provide shared memory for interprocess communication (IPC).
If you use Docker, set up the container as follows:
```bash
docker run --gpus all \
@ -120,7 +139,7 @@ docker run --gpus all \
vllm/vllm-openai
```
When using Kubernetes, you can set up the pod spec as follows:
If you use Kubernetes, set up the pod spec as follows:
```yaml
...
@ -146,13 +165,21 @@ spec:
...
```
!!! warning
After you start the Ray cluster, you'd better also check the GPU-GPU communication between nodes. It can be non-trivial to set up. Please refer to the [sanity check script][troubleshooting-incorrect-hardware-driver] for more information. If you need to set some environment variables for the communication configuration, you can append them to the `run_cluster.sh` script, e.g. `-e NCCL_SOCKET_IFNAME=eth0`. Note that setting environment variables in the shell (e.g. `NCCL_SOCKET_IFNAME=eth0 vllm serve ...`) only works for the processes in the same node, not for the processes in the other nodes. Setting environment variables when you create the cluster is the recommended way. See <gh-issue:6803> for more information.
Efficient tensor parallelism requires fast inter-node communication, preferably through high-speed network adapters such as InfiniBand. To enable InfiniBand, append flags such as `--privileged -e NCCL_IB_HCA=mlx5` to `run_cluster.sh`. For cluster-specific settings, consult your system administrator.
!!! warning
Please make sure you downloaded the model to all the nodes (with the same path), or the model is downloaded to some distributed file system that is accessible by all nodes.
To confirm InfiniBand operation, enable detailed NCCL logs:
When you use huggingface repo id to refer to the model, you should append your huggingface token to the `run_cluster.sh` script, e.g. `-e HF_TOKEN=`. The recommended way is to download the model first, and then use the path to refer to the model.
```bash
NCCL_DEBUG=TRACE vllm serve ...
```
!!! warning
If you keep receiving the error message `Error: No available node types can fulfill resource request` but you have enough GPUs in the cluster, chances are your nodes have multiple IP addresses and vLLM cannot find the right one, especially when you are using multi-node inference. Please make sure vLLM and ray use the same IP address. You can set the `VLLM_HOST_IP` environment variable to the right IP address in the `run_cluster.sh` script (different for each node!), and check `ray status` and `ray list nodes` to see the IP address used by Ray. See <gh-issue:7815> for more information.
Search the logs for the transport method. Entries containing `[send] via NET/Socket` indicate raw TCP sockets, which perform poorly for cross-node tensor parallelism. Entries containing `[send] via NET/IB/GDRDMA` indicate InfiniBand with GPUDirect RDMA, which provides high performance.
!!! tip "Verify inter-node GPU communication"
After you start the Ray cluster, verify GPU-to-GPU communication across nodes. Proper configuration can be non-trivial. For more information, see [troubleshooting script][troubleshooting-incorrect-hardware-driver]. If you need additional environment variables for communication configuration, append them to `run_cluster.sh`, for example `-e NCCL_SOCKET_IFNAME=eth0`. Setting environment variables during cluster creation is recommended because the variables propagate to all nodes. In contrast, setting environment variables in the shell affects only the local node. For more information, see <gh-issue:6803>.
!!! tip "Pre-download Hugging Face models"
If you use Hugging Face models, downloading the model before starting vLLM is recommended. Download the model on every node to the same path, or store the model on a distributed file system accessible by all nodes. Then pass the path to the model in place of the repository ID. Otherwise, supply a Hugging Face token by appending `-e HF_TOKEN=<TOKEN>` to `run_cluster.sh`.
!!! tip
The error message `Error: No available node types can fulfill resource request` can appear even when the cluster has enough GPUs. The issue often occurs when nodes have multiple IP addresses and vLLM can't select the correct one. Ensure that vLLM and Ray use the same IP address by setting `VLLM_HOST_IP` in `run_cluster.sh` (with a different value on each node). Use `ray status` and `ray list nodes` to verify the chosen IP address. For more information, see <gh-issue:7815>.