rocm containers #11
rocm containers #11
Conversation
WalkthroughThis pull request updates documentation and overhauls the Docker setup for various speech-to-text services. The README table is reformatted and corrected, while new Dockerfiles and Docker Compose files are introduced or modified for ROCm and CUDA support. Multiple STT projects (FasterWhisper, MyNorthAI, Nemo, Hitz, Project AINA Whisper, and Whisper) now include tailored build arguments, metadata labels, environment variables, configuration file copies, and updated entrypoints. New JSON configuration files have also been added to define runtime settings for CPU and GPU deployments. Changes
Sequence Diagram(s)sequenceDiagram
participant Dev as Developer/CI
participant DC as Docker Compose
participant DF as Docker Build Process
participant CNT as Container Runtime
Dev->>DC: Trigger build/up (e.g., docker-compose up)
DC->>DF: Initiate build using updated Dockerfiles (with ARGs, ENV, configs)
DF->>DF: Execute steps: set base image, install packages, copy config files
DF->>CNT: Create container with defined entrypoint (STT server with specific plugin)
CNT-->>DC: Service starts, exposing ports and configured for GPU/CPU
Poem
📜 Recent review detailsConfiguration used: CodeRabbit UI 📒 Files selected for processing (27)
🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
CodeRabbit Configuration File (
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (12)
nemo/Dockerfile.cuda (1)
20-27: Improve Pip Install Efficiency
Consider appending the--no-cache-dirflag to your pip install commands. This prevents pip from caching intermediate files, which can effectively reduce the image size. For example:- RUN pip3 install aiohttp \ - && if [ "${ALPHA}" == "true" ]; then \ - pip3 install ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo --pre; \ - else \ - pip3 install ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo; \ - fi \ - && mkdir -p ${HOME}/flagged \ - && rm -rf ${HOME}/.cache/* + RUN pip3 install --no-cache-dir aiohttp \ + && if [ "${ALPHA}" == "true" ]; then \ + pip3 install --no-cache-dir ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo --pre; \ + else \ + pip3 install --no-cache-dir ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo; \ + fi \ + && mkdir -p ${HOME}/flagged \ + && rm -rf ${HOME}/.cache/*nemo/Dockerfile.rocm (2)
7-8: Review Label Description Consistency
While the title clearly indicates AMD ROCm support, the description still begins with “NVIDIA NeMo”. To avoid confusion, consider updating the description so it accurately reflects AMD ROCm support if that is the intent.
20-27: Optimize Pip Package Installation
Similar to the CUDA Dockerfile, using the--no-cache-dirflag for pip installs is recommended here to reduce the image footprint. For example:- RUN pip3 install aiohttp \ - && if [ "${ALPHA}" == "true" ]; then \ - pip3 install ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo --pre; \ - else \ - pip3 install ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo; \ - fi \ - && mkdir -p ${HOME}/flagged \ - && rm -rf ${HOME}/.cache/* + RUN pip3 install --no-cache-dir aiohttp \ + && if [ "${ALPHA}" == "true" ]; then \ + pip3 install --no-cache-dir ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo --pre; \ + else \ + pip3 install --no-cache-dir ovos-stt-http-server SpeechRecognition setuptools ovos-stt-plugin-nemo; \ + fi \ + && mkdir -p ${HOME}/flagged \ + && rm -rf ${HOME}/.cache/*base/Dockerfile.rocm (1)
21-30: Efficient System Setup and Cleanup
The RUN command chains package installation, user creation, and virtual environment setup with apt-get update and cleanup commands. Consider adding the--no-install-recommendsflag to the apt-get install command to limit unnecessary packages. Also, ensure that the use of${HOME}(e.g. in the cache removal command) is consistent with the subsequent usage of/home/${USER}.mynorthai/Dockerfile (1)
20-27: Optimize Pip Package Installation in MyNorthAI
As with the other Dockerfiles, adding the--no-cache-dirflag to pip install commands could help reduce the final image size. For example:- RUN pip3 install aiohttp \ - && if [ "${ALPHA}" == "true" ]; then \ - pip3 install ovos-stt-http-server SpeechRecognition git+https://github.com/TigreGotico/ovos-stt-plugin-whisper.git git+https://github.com/TigreGotico/ovos-stt-plugin-MyNorthAI.git torch --pre; \ - else \ - pip3 install ovos-stt-http-server SpeechRecognition ovos-stt-plugin-MyNorthAI; \ - fi \ - && mkdir -p ${HOME}/flagged \ - && rm -rf ${HOME}/.cache/* + RUN pip3 install --no-cache-dir aiohttp \ + && if [ "${ALPHA}" == "true" ]; then \ + pip3 install --no-cache-dir ovos-stt-http-server SpeechRecognition git+https://github.com/TigreGotico/ovos-stt-plugin-whisper.git git+https://github.com/TigreGotico/ovos-stt-plugin-MyNorthAI.git torch --pre; \ + else \ + pip3 install --no-cache-dir ovos-stt-http-server SpeechRecognition ovos-stt-plugin-MyNorthAI; \ + fi \ + && mkdir -p ${HOME}/flagged \ + && rm -rf ${HOME}/.cache/*fasterwhisper/Dockerfile.rocm (1)
20-27: Conditional Package Installation in RUN Command
The RUN block conditionally installs packages using anifconstruct dependent on theALPHAargument. This conditional installation of pre-release packages versus stable packages is well implemented.
Consider verifying that the pip installs do not require additional version pinning for production stability (optional improvement).docker-compose.rocm.yml (2)
15-33: Volume Definitions and Naming Consistency
All volumes are defined with the local driver to facilitate caching.
Note: The volume nameovos_stt_fasterwshiper_gradio_cache(lines 19-21) appears to have a typographical error. For consistency with the service name “fasterwhisper”, consider renaming it toovos_stt_fasterwhisper_gradio_cache.Proposed diff:
- ovos_stt_fasterwshiper_gradio_cache: - name: ovos_stt_fasterwshiper_gradio_cache + ovos_stt_fasterwhisper_gradio_cache: + name: ovos_stt_fasterwhisper_gradio_cache
88-114: Service Configuration: ovos_stt_fasterwhisper
The block forovos_stt_fasterwhisperis overall well structured.
Note: The same potential typo in the volume name appears again on line 113 withovos_stt_fasterwshiper_gradio_cache. Consistency across volume names is important.Proposed diff:
- - ovos_stt_fasterwshiper_gradio_cache:/home/${OVOS_USER}/gradio_cached_examples + - ovos_stt_fasterwhisper_gradio_cache:/home/${OVOS_USER}/gradio_cached_examplesdocker-compose.cuda.yml (2)
15-33: Volume Definitions Consistency
The volumes are defined similarly to the ROCm file.
Note: The volumeovos_stt_fasterwshiper_gradio_cacheis again present on lines 19-21; consider correcting it toovos_stt_fasterwhisper_gradio_cachefor consistency across configurations.Proposed diff:
- ovos_stt_fasterwshiper_gradio_cache: - name: ovos_stt_fasterwshiper_gradio_cache + ovos_stt_fasterwhisper_gradio_cache: + name: ovos_stt_fasterwhisper_gradio_cache
88-114: Service Configuration: ovos_stt_fasterwhisper (CUDA)
The fasterwhisper service is correctly set up.
Note: As with the ROCm file, the volume mount on line 113 still usesovos_stt_fasterwshiper_gradio_cache. Renaming it toovos_stt_fasterwhisper_gradio_cachewould enhance overall consistency.Proposed diff:
- - ovos_stt_fasterwshiper_gradio_cache:/home/${OVOS_USER}/gradio_cached_examples + - ovos_stt_fasterwhisper_gradio_cache:/home/${OVOS_USER}/gradio_cached_examplesmynorthai/Dockerfile.rocm (1)
20-27: Conditional RUN Command for Package Installation
The RUN command employs a conditional structure to install pre-release packages whenALPHAis true, including additional dependencies (e.g., plugin installations andtorchin pre-release mode). This is a thoughtful setup for environments requiring experimental features.mynorthai/Dockerfile.cuda (1)
20-27: Conditional Package Installation in RUN Command (CUDA)
The RUN block mirrors the ROCm version’s logic by conditionally installing pre-release packages (including installation oftorchin pre-release mode) whenALPHAis true. This setup is consistent and meets the requirements for enabling optional features.
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (10)
README.md(1 hunks)base/Dockerfile.rocm(1 hunks)docker-compose.cuda.yml(1 hunks)docker-compose.rocm.yml(1 hunks)fasterwhisper/Dockerfile.rocm(1 hunks)mynorthai/Dockerfile(1 hunks)mynorthai/Dockerfile.cuda(1 hunks)mynorthai/Dockerfile.rocm(1 hunks)nemo/Dockerfile.cuda(1 hunks)nemo/Dockerfile.rocm(1 hunks)
🔇 Additional comments (40)
nemo/Dockerfile.cuda (3)
1-2: Base Image and TAG Argument Configuration
The ARG declaration and FROM statement clearly set the base image using the default tagalpha. Verify that using thealphatag suits your release and testing process.
26-27: Consistent Use of Home Directory
The RUN command refers to${HOME}(e.g.${HOME}/flaggedand${HOME}/.cache), whereas the ENV settings later hardcode/home/${USER}. Ensure that the environment variable$HOMEis reliably set—or consider using/home/${USER}consistently throughout the Dockerfile.
29-33: Environment Setup and Entry Point
The ENV configurations for PATH and VIRTUAL_ENV, along with the ENTRYPOINT and EXPOSE directives, are well defined and clear.README.md (2)
62-62: Enhanced Table Header Formatting
The updated header row using| -----------improves the clarity and alignment of the table columns in the arguments section.
64-64: Correcting BUILD_DATE Default Value
The typo in the default value forBUILD_DATEhas been corrected fromunkowntounknown, ensuring the documentation accurately reflects the build argument's value.nemo/Dockerfile.rocm (3)
1-2: Base Image and TAG Setup for ROCm
The ARG declaration and FROM statement properly reference the ROCm base image (smartgic/ovos-stt-server-base-rocm:${TAG}). Confirm that the defaultalphatag is appropriate for your ROCm builds.
26-27: Ensure Consistent Home Directory Reference
As with the CUDA variant, verify that using${HOME}in this section aligns with subsequent hardcoded references such as/home/${USER}.
29-33: Final Environment Configuration
The PATH and VIRTUAL_ENV variables, as well as the ENTRYPOINT and EXPOSE directives, are configured appropriately for the ROCm container.base/Dockerfile.rocm (2)
1-13: Base Image and Metadata Labels
The Dockerfile effectively sets up the base image fromrocm/dev-ubuntu-24.04:latestand uses build arguments to assign clear metadata labels. The provided information is complete and well formatted.
32-37: Final Environment Configuration
Switching to the newly created user, setting the PATH and VIRTUAL_ENV, and establishing the WORKDIR are correctly handled, ensuring a secure and predictable runtime environment.mynorthai/Dockerfile (3)
1-2: Updated Base Image for MyNorthAI
The base image has been changed from the CUDA-enabled variant to the standard image. Please confirm that this change is intentional and that the standard base image meets your performance and compatibility requirements.
7-9: Accurate Metadata and Documentation
The label correction for the description now reads “MyNorthAI is a STT specialized with Portuguese language,” which corrects the previous spelling error. This improves the clarity of the image metadata.
29-34: Environment and Entrypoint Configuration
The ENV settings for PATH and VIRTUAL_ENV, along with the ENTRYPOINT and EXPOSE directives, are correctly configured to launch the MyNorthAI STT service.fasterwhisper/Dockerfile.rocm (7)
1-2: Base Image and Tag Argument Usage
The use of anARG TAGwith a default of"alpha"and its application in theFROMinstruction is clear and allows for flexible image versioning.
4-6: Build Metadata Arguments
DefiningARG BUILD_DATEandARG VERSIONwith reasonable defaults helps embed build-time information into the image.
7-13: Image Labeling for Metadata
The LABEL directives provide comprehensive metadata (title, description, version, creation date, documentation, source, vendor) which improves traceability and image documentation.
15-17: User and Feature Toggle Arguments
IntroducingARG ALPHA(defaulting tofalse) andARG USERallows conditionally installing pre-release packages and configuring user-specific paths. This approach is consistent with Docker best practices.
18-18: Shell Declaration
Switching the shell to Bash explicitly via theSHELLdirective is ideal for writing complex multi-line commands and ensures consistent behavior.
29-30: Environment Variables for Virtual Environment
Setting the PATH and VIRTUAL_ENV using theUSERargument provides a clear way to manage the Python virtual environment. Verify that${HOME}is configured as expected in the base image.
32-34: Entrypoint and Port Exposure
The ENTRYPOINT command correctly starts the STT server with the designated plugin, and exposing port 8080 aligns with the service requirements.docker-compose.rocm.yml (3)
1-14: YAML Header and Reusable Configuration Definitions
Version declaration, YAML anchors forx-podmanandx-loggingare well structured. This reuse of configuration improves maintainability across services.
35-61: Service Configuration: ovos_stt_mynorthai
Theovos_stt_mynorthaiservice block correctly uses thepodmananchor, sets container/hostname, restart policy, and mounts the appropriate volumes. Environment variables and port mappings are also properly configured.
62-87: Service Configuration: ovos_stt_nemo
The setup forovos_stt_nemois consistent with the previous service, with proper resource reservation (e.g., NVIDIA GPU) and volume mounts.docker-compose.cuda.yml (3)
1-14: YAML Header and Anchors for CUDA Setup
The header mirrors the rocM configuration accurately, with YAML anchors for podman and logging ensuring consistency.
35-61: Service Configuration: ovos_stt_mynorthai (CUDA)
The configuration for the MyNorthAI service under CUDA is well defined and mirrors the analogous ROCm setup (with image changes reflecting CUDA support).
62-87: Service Configuration: ovos_stt_nemo (CUDA)
The nemo service replicates the proper pattern for container configuration with GPU reservations and volume mounting.mynorthai/Dockerfile.rocm (7)
1-2: Base Image and Tag Argument
Using the ROCm-based image fromsmartgic/ovos-stt-server-base-rocmwith an argumentized tag allows for flexible versioning of the base image.
4-6: Build Information Arguments
IncludingARG BUILD_DATEandARG VERSIONensures that build metadata is injected into the image, aiding in traceability.
7-13: Labels for Metadata
The LABEL directives provide clear, useful metadata including a tailored title and description for the MyNorthAI image.
15-17: User and Feature Toggle Arguments
DefiningARG ALPHAandARG USERmaintains consistency with other Dockerfiles, enabling conditional dependency management and user-specific configurations.
18-18: Shell Configuration
Setting the shell to Bash is appropriate for the multi-line RUN command that follows.
29-30: Environment Variable Setup
The ENV directives for PATH and VIRTUAL_ENV construct the expected virtual environment path, ensuring that installed Python packages are accessible at runtime.
32-34: Entrypoint and Exposure
The ENTRYPOINT command correctly invokes the STT server with the MyNorthAI plugin, and exposing port 8080 makes the service available externally.mynorthai/Dockerfile.cuda (7)
1-2: CUDA Base Image and Tag Parameterization
Usingsmartgic/ovos-stt-server-base-cudawith an ARG for the tag provides consistency and flexibility for CUDA-based deployments.
4-6: Metadata Arguments
The arguments for build date and version are set appropriately to allow embedding of build information.
7-13: Image Metadata Labeling
The LABEL directives correctly document the image by including title, description (tailored for MyNorthAI and CUDA support), version, creation date, documentation link, source, and vendor information.
15-17: Feature-Control and User Parameters
The use ofARG ALPHAandARG USERhelps conditionally install additional packages and set the proper user context.
18-18: Shell Directive
Specifying the shell as Bash enables robust handling of the multi-line RUN command.
29-30: Environment Variable Definitions
Setting PATH and VIRTUAL_ENV ensures that the Python virtual environment is correctly configured for runtime command resolution.
32-34: Entrypoint and Port Exposure
The ENTRYPOINT consistently starts the STT server with MyNorthAI support and EXPOSE makes port 8080 available for external access.
|
@goldyfruit I will let you take over now, but the ROCM images are all working in my home server now :) |
Summary by CodeRabbit
Documentation
New Features