feat: basic vllm support for hf cached models #2262
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,148 @@ | ||
| /********************************************************************** | ||
| * Copyright (C) 2024 Red Hat, Inc. | ||
| * | ||
| * Licensed under the Apache License, Version 2.0 (the "License"); | ||
| * you may not use this file except in compliance with the License. | ||
| * You may obtain a copy of the License at | ||
| * | ||
| * http://www.apache.org/licenses/LICENSE-2.0 | ||
| * | ||
| * Unless required by applicable law or agreed to in writing, software | ||
| * distributed under the License is distributed on an "AS IS" BASIS, | ||
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| * See the License for the specific language governing permissions and | ||
| * limitations under the License. | ||
| * | ||
| * SPDX-License-Identifier: Apache-2.0 | ||
| ***********************************************************************/ | ||
|
|
||
| import { InferenceProvider } from './InferenceProvider'; | ||
| import type { TaskRegistry } from '../../registries/TaskRegistry'; | ||
| import type { PodmanConnection } from '../../managers/podmanConnection'; | ||
| import { type InferenceServer, InferenceType } from '@shared/src/models/IInference'; | ||
| import type { InferenceServerConfig } from '@shared/src/models/InferenceServerConfig'; | ||
| import type { ContainerProviderConnection, MountConfig } from '@podman-desktop/api'; | ||
| import * as images from '../../assets/inference-images.json'; | ||
| import { LABEL_INFERENCE_SERVER } from '../../utils/inferenceUtils'; | ||
| import { DISABLE_SELINUX_LABEL_SECURITY_OPTION } from '../../utils/utils'; | ||
| import { basename, dirname } from 'node:path'; | ||
| import { join as joinposix } from 'node:path/posix'; | ||
| import { getLocalModelFile } from '../../utils/modelsUtils'; | ||
|
|
||
| export class VLLM extends InferenceProvider { | ||
| constructor( | ||
| taskRegistry: TaskRegistry, | ||
| private podmanConnection: PodmanConnection, | ||
| ) { | ||
| super(taskRegistry, InferenceType.VLLM, 'vllm'); | ||
| } | ||
|
|
||
| dispose(): void {} | ||
|
|
||
| public enabled = (): boolean => true; | ||
|
|
||
| /** | ||
| * Here is an example | ||
| * | ||
| * podman run -it --rm | ||
| * -v C:\Users\axels\.cache\huggingface\hub\models--mistralai--Mistral-7B-v0.1:/cache/models--mistralai--Mistral-7B-v0.1 | ||
| * -e HF_HUB_CACHE=/cache | ||
| * localhost/vllm-cpu-env:latest | ||
| * --model=/cache/models--mistralai--Mistral-7B-v0.1/snapshots/7231864981174d9bee8c7687c24c8344414eae6b | ||
| * | ||
| * @param config | ||
| */ | ||
| override async perform(config: InferenceServerConfig): Promise<InferenceServer> { | ||
| if (config.modelsInfo.length !== 1) | ||
| throw new Error(`only one model is supported, received ${config.modelsInfo.length}`); | ||
|
|
||
| const modelInfo = config.modelsInfo[0]; | ||
| if (modelInfo.backend !== InferenceType.VLLM) { | ||
| throw new Error(`VLLM requires models with backend type ${InferenceType.VLLM} got ${modelInfo.backend}.`); | ||
| } | ||
|
|
||
| if (modelInfo.file === undefined) { | ||
| throw new Error('The model info file provided is undefined'); | ||
| } | ||
|
|
||
| console.log('[VLLM]', config); | ||
| console.log('[VLLM] modelInfo.file', modelInfo.file); | ||
|
|
||
| const fullPath = getLocalModelFile(modelInfo); | ||
|
|
||
| // modelInfo.file.path must be under the form $(HF_HUB_CACHE)/<repo-type>--<repo-id>/snapshots/<commit-hash> | ||
| const parent = dirname(fullPath); | ||
| const commitHash = basename(fullPath); | ||
| const name = basename(parent); | ||
| if (name !== 'snapshots') throw new Error('you must provide snapshot path for vllm'); | ||
| const modelCache = dirname(parent); | ||
|
|
||
| let connection: ContainerProviderConnection | undefined; | ||
| if (config.connection) { | ||
| connection = this.podmanConnection.getContainerProviderConnection(config.connection); | ||
| } else { | ||
| connection = this.podmanConnection.findRunningContainerProviderConnection(); | ||
| } | ||
|
|
||
| if (!connection) throw new Error('no running connection could be found'); | ||
|
|
||
| const labels: Record<string, string> = { | ||
| ...config.labels, | ||
| [LABEL_INFERENCE_SERVER]: JSON.stringify(config.modelsInfo.map(model => model.id)), | ||
| }; | ||
|
|
||
| const imageInfo = await this.pullImage(connection, config.image ?? images.vllm.default, labels); | ||
| // https://huggingface.co/docs/transformers/main/en/installation#offline-mode | ||
| // HF_HUB_OFFLINE in main | ||
| // TRANSFORMERS_OFFLINE for legacy | ||
| const envs: string[] = [`HF_HUB_CACHE=/cache`, 'TRANSFORMERS_OFFLINE=1', 'HF_HUB_OFFLINE=1']; | ||
|
|
||
| labels['api'] = `http://localhost:${config.port}/inference`; | ||
|
|
||
| const mounts: MountConfig = [ | ||
| { | ||
| Target: `/cache/${modelInfo.id}`, | ||
| Source: modelCache, | ||
| Type: 'bind', | ||
| }, | ||
| ]; | ||
|
|
||
| const containerInfo = await this.createContainer( | ||
| imageInfo.engineId, | ||
| { | ||
| Image: imageInfo.Id, | ||
| Detach: true, | ||
| Labels: labels, | ||
| HostConfig: { | ||
| AutoRemove: false, | ||
| Mounts: mounts, | ||
| PortBindings: { | ||
| '8000/tcp': [ | ||
| { | ||
| HostPort: `${config.port}`, | ||
| }, | ||
| ], | ||
| }, | ||
| SecurityOpt: [DISABLE_SELINUX_LABEL_SECURITY_OPTION], | ||
| }, | ||
| Env: envs, | ||
| Cmd: [`--model=${joinposix('/cache', modelInfo.id, 'snapshots', commitHash)}`], | ||
| }, | ||
| labels, | ||
| ); | ||
|
|
||
| return { | ||
| models: [modelInfo], | ||
| status: 'running', | ||
| connection: { | ||
| port: config.port, | ||
| }, | ||
| container: { | ||
| containerId: containerInfo.id, | ||
| engineId: containerInfo.engineId, | ||
| }, | ||
| type: InferenceType.VLLM, | ||
| labels: labels, | ||
| }; | ||
| } | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Highly recommend using the openshift AI vllm container images:
containers/ramalama#533
Ideally we'd just make ramalama the backend, and podman ai desktop wouldn't have to worry about the plumbing so much.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I am not finding any CPU image on the
quay.io/modh/vllmrepository; this is blockingThere was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Do the CUDA or ROCm images fallback to CPU inferencing? Worth checking out, I wouldn't be surprised if they did. Then we don't really need a new container image
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
To be honest with vllm in general, limited hardware support is an issue. It's why in RamaLama we offer switchable llama.cpp/vLLM
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No it just crashed. Here is the logs