Zengtudor/Wan2.1
1
0
Fork 0
mirror of https://github.com/Wan-Video/Wan2.1.git synced 2025-11-04 06:15:17 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge 0543adc653 into 7c81b2f27d

Browse Source
This commit is contained in:
ariaemaan 2025-07-18 15:23:25 +08:00 committed by GitHub
commit 90eba46893
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 234 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

143
DEPLOY_GCLOUD.md Normal file
View File

@ -0,0 +1,143 @@
# Deploying to Google Cloud
This guide provides instructions on how to build and deploy this application to Google Cloud using Docker, Cloud Build, and Cloud Run or Vertex AI Endpoints.
## Prerequisites
1. **Google Cloud Project:** You have a Google Cloud Project with billing enabled.
2. **Enable APIs:** Ensure the following APIs are enabled in your project:
* Cloud Build API
* Artifact Registry API
* Cloud Run API
* Vertex AI API (if deploying to Vertex AI Endpoints)
* Identity and Access Management (IAM) API
3. **`gcloud` CLI:** You have the [Google Cloud CLI](https://cloud.google.com/sdk/docs/install) installed and configured.
4. **Permissions:** You have sufficient permissions to create Artifact Registry repositories, trigger Cloud Builds, and deploy to Cloud Run/Vertex AI. Roles like `Artifact Registry Administrator`, `Cloud Build Editor`, `Cloud Run Admin`, and `Vertex AI Administrator` are typically needed.
5. **Docker configured for gcloud:** Configure Docker to authenticate with Artifact Registry:
```bash
gcloud auth configure-docker ${_REGION}-docker.pkg.dev
```
Replace `${_REGION}` with the region of your Artifact Registry (e.g., `us-central1`). You'll define this region in the Cloud Build substitutions.
## Step 1: Configure Cloud Build Substitutions
The `cloudbuild.yaml` file uses substitutions for your project ID, Artifact Registry region, repository name, and image name. You need to set these. The easiest way for a manual build is via the `gcloud` command.
**Important:** Before running the build, decide on:
* `YOUR_GCP_PROJECT_ID`: Your actual Google Cloud Project ID.
* `YOUR_ARTIFACT_REGISTRY_REGION`: The Google Cloud region for your Artifact Registry (e.g., `us-central1`).
* `YOUR_ARTIFACT_REGISTRY_REPO`: The name for your Artifact Registry repository (e.g., `wan-video-repo`).
* `YOUR_IMAGE_NAME`: The name for your Docker image (e.g., `wan-video-generator`).
The `cloudbuild.yaml` has default values for `_REGION`, `_REPOSITORY`, and `_IMAGE_NAME`. You can either:
* Modify these defaults directly in `cloudbuild.yaml`.
* Override them when submitting the build using the `--substitutions` flag (recommended for flexibility).
Example of overriding:
`--substitutions=PROJECT_ID="your-gcp-project-id",_REGION="us-central1",_REPOSITORY="wan-video-repo",_IMAGE_NAME="wan-video-generator"`
## Step 2: Create an Artifact Registry Repository
If you haven't already, create a Docker repository in Artifact Registry:
```bash
gcloud artifacts repositories create YOUR_ARTIFACT_REGISTRY_REPO --repository-format=docker --location=YOUR_ARTIFACT_REGISTRY_REGION --description="Docker repository for Wan video generator"
```
Ensure `YOUR_ARTIFACT_REGISTRY_REPO` and `YOUR_ARTIFACT_REGISTRY_REGION` match the values you'll use for the `_REPOSITORY` and `_REGION` substitutions in Cloud Build.
## Step 3: Build and Push the Docker Image
Submit the build to Google Cloud Build:
```bash
# Ensure you are in the root directory of the project where cloudbuild.yaml is located.
# Replace placeholders with your actual values.
gcloud builds submit --config cloudbuild.yaml --substitutions=PROJECT_ID="YOUR_GCP_PROJECT_ID",_REGION="YOUR_ARTIFACT_REGISTRY_REGION",_REPOSITORY="YOUR_ARTIFACT_REGISTRY_REPO",_IMAGE_NAME="YOUR_IMAGE_NAME" .
```
This command uses `cloudbuild.yaml` to build your Docker image and push it to your Artifact Registry.
## Step 4: Deploy the Container
You can deploy the container to Cloud Run or Vertex AI Endpoints.
### Option A: Deploy to Cloud Run
Cloud Run is suitable for web applications like Gradio interfaces.
1. **Basic Deployment (CPU only):**
```bash
# Replace placeholders with your values.
# YOUR_CLOUD_RUN_REGION should be a region where Cloud Run is available (e.g., us-central1).
gcloud run deploy YOUR_IMAGE_NAME --image YOUR_ARTIFACT_REGISTRY_REGION-docker.pkg.dev/YOUR_GCP_PROJECT_ID/YOUR_ARTIFACT_REGISTRY_REPO/YOUR_IMAGE_NAME:latest --platform managed --region YOUR_CLOUD_RUN_REGION --allow-unauthenticated --port 7860 --set-env-vars GRADIO_APP_SCRIPT="gradio/t2v_14B_singleGPU.py" --memory=4Gi --cpu=2
# Adjust memory, CPU, and other flags as needed.
# For larger models, you will need significantly more memory and potentially more CPU.
```
* The `--port 7860` matches the `EXPOSE 7860` in the Dockerfile.
* Use `--set-env-vars` to specify which Gradio app to run via the `GRADIO_APP_SCRIPT` environment variable.
* The image path format is `REGION-docker.pkg.dev/PROJECT_ID/REPOSITORY/IMAGE_NAME:latest`.
2. **Deployment with GPU:**
Cloud Run supports GPUs (check availability and machine types in your region).
```bash
# Replace placeholders. YOUR_CLOUD_RUN_REGION must support GPUs.
gcloud beta run deploy YOUR_IMAGE_NAME-gpu --image YOUR_ARTIFACT_REGISTRY_REGION-docker.pkg.dev/YOUR_GCP_PROJECT_ID/YOUR_ARTIFACT_REGISTRY_REPO/YOUR_IMAGE_NAME:latest --platform managed --region YOUR_CLOUD_RUN_REGION --allow-unauthenticated --port 7860 --set-env-vars GRADIO_APP_SCRIPT="gradio/t2v_14B_singleGPU.py" --memory=16Gi --cpu=4 --execution-environment gen2 --args=--machine-type=a2-highgpu-1g # Or other suitable GPU machine types.
# Consult Cloud Run documentation for current GPU options.
```
3. **Considerations for Cloud Run:**
* **Timeout:** Default request timeout is 5 minutes. Increase if needed (max 60 minutes for Gen2).
* **Concurrency:** Adjust based on instance capacity.
* **Model Loading & Storage:**
* Models are currently packaged in the Docker image. This increases image size.
* For very large models, consider downloading them at startup from Google Cloud Storage (GCS) into the container. This would require modifying the Docker `CMD` or `ENTRYPOINT`.
### Option B: Deploy to Vertex AI Endpoints
Vertex AI Endpoints are better for dedicated ML model serving and offer more powerful hardware options.
1. **Create an Endpoint:**
```bash
# Replace placeholders. YOUR_VERTEX_AI_REGION is e.g., us-central1.
gcloud ai endpoints create --project=YOUR_GCP_PROJECT_ID --region=YOUR_VERTEX_AI_REGION --display-name="wan-video-endpoint"
```
Note the `ENDPOINT_ID` from the output.
2. **Deploy the model (container) to the Endpoint:**
```bash
# Replace placeholders.
# ENDPOINT_ID is from the previous command.
# MACHINE_TYPE can be n1-standard-4, or a GPU type like a2-highgpu-1g.
gcloud ai models deploy wan-video-model --project=YOUR_GCP_PROJECT_ID --region=YOUR_VERTEX_AI_REGION --endpoint=ENDPOINT_ID --display-name="v1" --container-image-uri="YOUR_ARTIFACT_REGISTRY_REGION-docker.pkg.dev/YOUR_GCP_PROJECT_ID/YOUR_ARTIFACT_REGISTRY_REPO/YOUR_IMAGE_NAME:latest" --machine-type=MACHINE_TYPE # --container-env-vars="GRADIO_APP_SCRIPT=gradio/t2v_14B_singleGPU.py" # If serving Gradio UI
# --container-ports=7860 # If serving Gradio UI
# For dedicated prediction (non-Gradio), you'd typically implement /predict and /health routes.
# --container-predict-route="/predict"
# --container-health-route="/health"
```
* Vertex AI is more commonly used for direct prediction endpoints. If serving a Gradio UI, ensure networking and port configurations are appropriate. Accessing a UI might require additional setup (e.g., IAP).
## Step 5: Accessing your Deployed Application
* **Cloud Run:** The `gcloud run deploy` command will output a service URL.
* **Vertex AI Endpoint:** Typically accessed programmatically via SDK or REST API.
## Step 6: Checking Logs for Troubleshooting
* **Cloud Build Logs:** Google Cloud Console > Cloud Build > History.
* **Cloud Run Logs:** Google Cloud Console > Cloud Run > Select Service > Logs tab.
* **Vertex AI Endpoint Logs:** Google Cloud Console > Vertex AI > Endpoints > Select Endpoint > View Logs. Also available in Cloud Logging.
Look for errors related to dependency installation, model loading, resource limits (memory/CPU), or port configurations.
## Choosing Machine Types and Resources
* **CPU/Memory:** The 14B models are very demanding. Start with high CPU/memory (e.g., 4+ vCPUs, 16GB+ RAM) and scale up.
* **GPU:** Essential for 14B models and highly recommended for 1.3B models.
* **Cloud Run:** Gen2 execution environment with GPU machine types.
* **Vertex AI:** Offers a wide variety of GPUs (T4, V100, A100).
* **Model Sizes & Compatibility:**
* The `nvcr.io/nvidia/pytorch:24.03-py3` base image uses CUDA 12.1. Ensure chosen GPUs are compatible (e.g., Ampere, Hopper).
* For 14B models, you'll likely need GPUs with large VRAM (e.g., A100 40GB or 80GB). Check the model's specific requirements.
This guide provides a starting point. You may need to adjust configurations based on the specific model you are deploying and your performance requirements.

48
Dockerfile Normal file
View File

@ -0,0 +1,48 @@
# Use a PyTorch base image with CUDA support
# Check NVIDIA's NGC catalog or Docker Hub for suitable images
# Example: nvcr.io/nvidia/pytorch:24.03-py3 (PyTorch 2.4.0, Python 3.10, CUDA 12.1)
# Ensure the Python version in the base image is compatible with project dependencies.
# Python 3.10 is generally a safe bet for recent PyTorch versions.
FROM nvcr.io/nvidia/pytorch:24.03-py3
# Set the working directory
WORKDIR /app
# Install essential build tools and Python headers (if needed for some packages)
# RUN apt-get update && apt-get install -y --no-install-recommends # build-essential # python3-dev # && rm -rf /var/lib/apt/lists/*
# Copy the requirements file first to leverage Docker layer caching
COPY requirements.txt .
# Install Python dependencies
# Upgrading pip, setuptools, and wheel first can prevent some build issues.
RUN pip install --upgrade pip setuptools wheel
# Special handling for flash-attn as per INSTALL.md
# Attempting the no-build-isolation method first.
# Ensure that the PyTorch and CUDA versions are compatible with flash-attn
RUN pip install flash-attn --no-build-isolation
# Install other dependencies
# Using --no-cache-dir to reduce image size
RUN pip install --no-cache-dir -r requirements.txt
# Copy the rest of the application code
COPY . .
# Make generate.py executable (if it's intended to be run directly and has a shebang)
# RUN chmod +x generate.py
# Set environment variables (optional, can be overridden at runtime)
# Example: Set a default Gradio app to run
ENV GRADIO_APP_SCRIPT gradio/t2v_14B_singleGPU.py
# ENV MODEL_CKPT_DIR ./Wan2.1-T2V-14B # Or a path inside the container where models will be mounted/downloaded
# Expose the default Gradio port (usually 7860)
EXPOSE 7860
# Default command to run the Gradio application
# This assumes the Gradio apps are launched with `python <script_name>`
# Users might need to adjust this based on how they want to run their app.
# Using `sh -c` to allow environment variable substitution in the command.
CMD ["sh", "-c", "python $GRADIO_APP_SCRIPT"]

43
cloudbuild.yaml Normal file
View File

@ -0,0 +1,43 @@
steps:
- name: 'gcr.io/cloud-builders/docker'
args: [
'build',
'-t',
'${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_IMAGE_NAME}:${COMMIT_SHA}',
'-t',
'${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_IMAGE_NAME}:latest',
'.'
]
id: 'Build Docker image'
- name: 'gcr.io/cloud-builders/docker'
args: ['push', '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_IMAGE_NAME}:${COMMIT_SHA}']
id: 'Push image to Artifact Registry (tagged with commit SHA)'
- name: 'gcr.io/cloud-builders/docker'
args: ['push', '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_IMAGE_NAME}:latest']
id: 'Push image to Artifact Registry (tagged as latest)'
# Substitutions (users should replace these with their actual values or set them during build trigger)
# _REGION: The region of your Artifact Registry repository (e.g., us-central1)
# _REPOSITORY: The name of your Artifact Registry repository (e.g., my-app-repo)
# _IMAGE_NAME: The name for your Docker image (e.g., wan-video-generator)
# Default values for substitutions (can be overridden)
substitutions:
_REGION: 'us-central1' # Replace with your Artifact Registry region
_REPOSITORY: 'my-app-images' # Replace with your Artifact Registry repository name
_IMAGE_NAME: 'wan-video-service' # Replace with your desired image name
# Optional: specify image to be pushed to Artifact Registry
# This allows the image to be displayed in the build summary in Google Cloud Console
images:
- '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_IMAGE_NAME}:${COMMIT_SHA}'
- '${_REGION}-docker.pkg.dev/${PROJECT_ID}/${_REPOSITORY}/${_IMAGE_NAME}:latest'
options:
logging: CLOUD_LOGGING_ONLY
# Optional: Specify machine type for faster builds (e.g., for GPU builds if needed by Dockerfile)
# machineType: 'E2_HIGHCPU_8' # or 'N1_HIGHCPU_32'
# For GPU builds during the Docker build process (if Dockerfile uses GPU):
# machineType: 'N1_HIGHCPU_8_CLOUD_GPU_P100_1' # Example, check available types