Update NVIDIA PyTorch to 2.9.1 cu130

[benceruleanlu]

┆Issue is synchronized with this Notion page by Unito

[coderabbitai]

📝 Walkthrough

Walkthrough

CUDA 130 upgrade: Updates PyTorch wheel index URLs from cu129 to cu130 and bumps package versions including PyTorch, torchaudio, torchvision, and related dependencies to cu130-compatible releases across configuration and code constants files.

Changes

Cohort / File(s)	Summary
PyTorch wheel mirror updates assets/override.txt, src/constants.ts	Updated PyTorch wheel index URLs from cu129 to cu130 in override configuration and TorchMirrorUrl enum constants for both standard and nightly CUDA mirrors.
Windows NVIDIA compiled requirements assets/requirements/windows_nvidia.compiled	Bumped package versions for cu130 compatibility: PyTorch, torchaudio, and torchvision to 2.9.1/0.24.1+cu130; comfyui-manager to 4.0.4; comfyui-workflow-templates-core to 0.3.61, media-api to 0.3.34, and media-other to 0.3.62; updated dependencies (mpmath, networkx, numpy, pillow, etc.) to cu130-compatible builds.

✨ Finishing touches

• 📝 Generate docstrings

---

📜 Recent review details

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 2f6f2c9 and cb7c469.

📒 Files selected for processing (3)

• assets/override.txt
• assets/requirements/windows_nvidia.compiled
• src/constants.ts

🧰 Additional context used

📓 Path-based instructions (4)

**/*.{ts,mts,cts}

📄 CodeRabbit inference engine (.cursor/rules/typescript.mdc)

**/*.{ts,mts,cts}: Use features available in TypeScript 5.x
Use ESM only (type: module). Avoid CommonJS (require, module.exports)
Target ESNext runtime APIs and syntax. Prefer top-level await only when it improves clarity
Code must compile with strict mode and noImplicitAny enabled, with zero implicit any
Use unknown instead of any. Narrow with type guards. any is allowed only when interfacing with untyped code and must be localized and justified
Use interface for public object shapes intended for extension/implementation
Use type for unions, mapped/conditional types, function types, and when composition is needed
Model domain variants with a discriminant field (e.g., kind: 'X' | 'Y'). Use exhaustive switch statements
Prefer as const objects and string/number literal unions over enum
Prefer readonly properties and ReadonlyArray<T> where appropriate. Do not mutate function parameters
Prefer T[] over Array<T> for readability
Prefer Record<Key, T> for simple maps; use Map/Set when iteration semantics or key types require it
Avoid non-null assertions (!). Use optional chaining, nullish coalescing, and explicit guards
Use satisfies to enforce object literal conformance without widening values
Prefer named exports. Default exports are allowed only when a module clearly has a single primary export
Use import type { X } from '…' for type-only imports. Keep value and type imports separated when helpful for clarity and bundling
Exported functions, classes, and public APIs must have explicit parameter and return types. Internal locals can rely on inference
Prefer arrow functions for local functions and callbacks. Use function declarations when hoisting or this binding is required
Modules should be pure. Avoid executing logic at import time unless necessary (e.g., polyfills)
Prefer async/await over raw then/catch chains. Keep linear flow and use try/catch for failures
Either await promises or deliberately mark them...

Files:

• src/constants.ts

src/**/*.{ts,mts,cts}

📄 CodeRabbit inference engine (.cursor/rules/typescript.mdc)

Centralize reusable domain types in src/** where discoverable. Avoid ad-hoc inline types for shared structures

Files:

• src/constants.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

**/*.{ts,tsx}: TypeScript: Do not use the any type; use unknown when the type is genuinely unknown
Use JSDoc format for method documentation with common tags @param and @return (omit @return for void return type); use {@link } to reference symbols

Files:

• src/constants.ts

**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use ESLint and Prettier for code formatting; run yarn lint and yarn format before committing

Files:

• src/constants.ts

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (4)

• GitHub Check: Agent
• GitHub Check: build-apple-debug-all / build-macos-debug
• GitHub Check: lint-and-format (windows-latest)
• GitHub Check: build-and-test-e2e-windows / integration-windows-test

🔇 Additional comments (2)

src/constants.ts (1)

172-182: TorchMirrorUrl now correctly targets cu130 CUDA mirrors

The enum update from cu129 to cu130 is complete. Deprecated constants below (CUDA_TORCH_URL, NIGHTLY_CPU_TORCH_URL, DEFAULT_PYPI_INDEX_URL) correctly reference the enum and will pick up the new values automatically. No stale cu129 references remain in the codebase.

assets/requirements/windows_nvidia.compiled (1)

2-332: Compiled Windows NVIDIA requirements are consistent with cu130 torch stack

The uv header, global --extra-index-url, and pinned torch packages are all correctly aligned with cu130:

• torch==2.9.1+cu130
• torchaudio==2.9.1+cu130
• torchvision==0.24.1+cu130

No stale cu129 references remain in the compiled file.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Copilot]

Pull request overview

This pull request updates NVIDIA PyTorch packages from version 2.8.0 with CUDA 12.9 support to version 2.9.1 with CUDA 13.0 support for Windows systems with NVIDIA GPUs.

• Updated PyTorch and related packages (torch, torchaudio, torchvision) to newer versions with CUDA 13.0 support
• Updated CUDA mirror URLs from cu129 to cu130 across configuration files
• Regenerated Windows NVIDIA requirements file with updated dependency versions

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
src/constants.ts	Updated CUDA mirror URLs from cu129 to cu130 for both regular and nightly builds
assets/override.txt	Updated extra-index-url to point to cu130 PyTorch wheel repository
assets/requirements/windows_nvidia.compiled	Updated PyTorch packages to 2.9.1+cu130, torchaudio to 2.9.1+cu130, torchvision to 0.24.1+cu130, and regenerated all dependency references to use cu130 mirror

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[coderabbitai]

Actionable comments posted: 0

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

assets/override.txt (1)

1-6: cu129 references in macos.compiled still need to be updated to cu130

While assets/override.txt correctly targets cu130, assets/requirements/macos.compiled still contains cu129 references:

• Line 4: --extra-index-url https://download.pytorch.org/whl/cu129
• Multiple dependency comments (lines 84, 94, 117, 140, 148, 157, 168, 259, 324) reference cu129

This inconsistency must be resolved. Update macos.compiled to target cu130 to match the override and ensure consistent dependency resolution.

📜 Review details

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 2f6f2c9 and cb7c469.

📒 Files selected for processing (3)

• assets/override.txt
• assets/requirements/windows_nvidia.compiled
• src/constants.ts

🧰 Additional context used

📓 Path-based instructions (4)

**/*.{ts,mts,cts}

📄 CodeRabbit inference engine (.cursor/rules/typescript.mdc)

**/*.{ts,mts,cts}: Use features available in TypeScript 5.x
Use ESM only (type: module). Avoid CommonJS (require, module.exports)
Target ESNext runtime APIs and syntax. Prefer top-level await only when it improves clarity
Code must compile with strict mode and noImplicitAny enabled, with zero implicit any
Use unknown instead of any. Narrow with type guards. any is allowed only when interfacing with untyped code and must be localized and justified
Use interface for public object shapes intended for extension/implementation
Use type for unions, mapped/conditional types, function types, and when composition is needed
Model domain variants with a discriminant field (e.g., kind: 'X' | 'Y'). Use exhaustive switch statements
Prefer as const objects and string/number literal unions over enum
Prefer readonly properties and ReadonlyArray<T> where appropriate. Do not mutate function parameters
Prefer T[] over Array<T> for readability
Prefer Record<Key, T> for simple maps; use Map/Set when iteration semantics or key types require it
Avoid non-null assertions (!). Use optional chaining, nullish coalescing, and explicit guards
Use satisfies to enforce object literal conformance without widening values
Prefer named exports. Default exports are allowed only when a module clearly has a single primary export
Use import type { X } from '…' for type-only imports. Keep value and type imports separated when helpful for clarity and bundling
Exported functions, classes, and public APIs must have explicit parameter and return types. Internal locals can rely on inference
Prefer arrow functions for local functions and callbacks. Use function declarations when hoisting or this binding is required
Modules should be pure. Avoid executing logic at import time unless necessary (e.g., polyfills)
Prefer async/await over raw then/catch chains. Keep linear flow and use try/catch for failures
Either await promises or deliberately mark them...

Files:

• src/constants.ts

src/**/*.{ts,mts,cts}

📄 CodeRabbit inference engine (.cursor/rules/typescript.mdc)

Centralize reusable domain types in src/** where discoverable. Avoid ad-hoc inline types for shared structures

Files:

• src/constants.ts

**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

**/*.{ts,tsx}: TypeScript: Do not use the any type; use unknown when the type is genuinely unknown
Use JSDoc format for method documentation with common tags @param and @return (omit @return for void return type); use {@link } to reference symbols

Files:

• src/constants.ts

**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (AGENTS.md)

Use ESLint and Prettier for code formatting; run yarn lint and yarn format before committing

Files:

• src/constants.ts

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (4)

• GitHub Check: Agent
• GitHub Check: build-apple-debug-all / build-macos-debug
• GitHub Check: lint-and-format (windows-latest)
• GitHub Check: build-and-test-e2e-windows / integration-windows-test

🔇 Additional comments (2)

src/constants.ts (1)

172-182: TorchMirrorUrl now correctly targets cu130 CUDA mirrors

The enum update from cu129 to cu130 is complete. Deprecated constants below (CUDA_TORCH_URL, NIGHTLY_CPU_TORCH_URL, DEFAULT_PYPI_INDEX_URL) correctly reference the enum and will pick up the new values automatically. No stale cu129 references remain in the codebase.

assets/requirements/windows_nvidia.compiled (1)

2-332: Compiled Windows NVIDIA requirements are consistent with cu130 torch stack

The uv header, global --extra-index-url, and pinned torch packages are all correctly aligned with cu130:

• torch==2.9.1+cu130
• torchaudio==2.9.1+cu130
• torchvision==0.24.1+cu130

No stale cu129 references remain in the compiled file.