Optimize TrustyAI CPU image to use CPU-only PyTorch wheels instead of CUDA wheels

[coderabbitai]

Problem Description

The TrustyAI CPU Dockerfile () currently uses a that contains CUDA-enabled PyTorch wheels, even though this is intended to be a CPU-only image.

Current State

• File:
• Current PyTorch version:
• Current wheel source: CUDA-specific wheels from
• Impact: Unnecessarily large image size and potential compatibility issues

Affected Components

• TrustyAI CPU notebook image
• PyTorch dependency resolution
• Image build size and performance

Solution Options

Option 1: Regenerate lock with CPU-only PyTorch

• Modify to exclude CUDA index for CPU builds
• Regenerate with CPU-only PyTorch wheels
• Ensure torch version becomes (without suffix)

Option 2: Separate CPU/CUDA lock files

• Maintain separate files for CPU and CUDA variants
• Update Dockerfiles to use appropriate lock file

Option 3: Conditional lock generation

• Use build arguments or environment variables to generate appropriate locks

Acceptance Criteria

• TrustyAI CPU image uses CPU-only PyTorch wheels
• No CUDA-specific URLs or version suffixes in CPU image lock file
• Image size reduction verified
• Functional testing confirms no regression in CPU-only workflows
• Build process documentation updated if needed

Context

• PR: RHAIENG-304: add uv pyproject.tomls for multiple notebooks and runtimes across Jupyter, RStudio, and VSCode #2145 - Migration to uv pyproject.toml
• Review Comment: RHAIENG-304: add uv pyproject.tomls for multiple notebooks and runtimes across Jupyter, RStudio, and VSCode #2145 (comment)
• Previous Discussion: Issue was previously discussed with @harshad16 and @grdryn but deferred for future consideration

Additional Notes

This optimization would ensure CPU images only pull CPU-optimized wheels, reducing image size and avoiding unnecessary CUDA dependencies in CPU-only environments.