[Feature] Per-model CPU execution

[lstein]

Implementation Complete ✅

This PR adds the ability to configure models to run on CPU instead of GPU on a per-model basis.

Changes Made

Backend:

• Added cpu_only field to MainModelDefaultSettings
• Modified ModelCache.put() to accept optional execution_device parameter
• Updated ModelLoader._load_and_cache() to pass device from model config
• Added helper method to determine execution device from model config

Frontend:

• New "CPU Only" toggle component in Model Manager Default Settings
• Translation strings for UI labels
• Form integration for saving the setting

How to Use

1. Open Model Manager
2. Select a main model
3. Go to Default Settings tab
4. Enable "CPU Only" toggle
5. Save settings
6. Model will now run on CPU instead of GPU

NOTE:

After debugging, this should be merged into remote.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>[enhancement]: Allow some models to run on CPU</issue_title>
<issue_description>### Is there an existing issue for this?

• I have searched the existing issues

Contact Details

No response

What should this feature add?

By default most models run in VRAM, but VRAM is limited. I would like the option to run a model on the CPU. This would take the form of a slider button in the model manager's Default Settings tab which allows you to set, on a per-model basis, the option to "Run model on CPU". There could also be a default setting for this defined in invokeai/backend/model_manager/configs/base.py.

Alternatives

No response

Additional Content

No response</issue_description>

<agent_instructions>Please make this pull request relative to the main branch.</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes [enhancement]: Allow some models to run on CPU #44

[lstein]

@copilot I'm getting this frontend linting error:

> @invoke-ai/invoke-ai-ui@0.0.1 lint:tsc /home/runner/work/InvokeAI/InvokeAI/invokeai/frontend/web
> tsc --noEmit

Error: src/features/modelManagerV2/hooks/useMainModelDefaultSettings.ts(45,58): error TS2339: Property 'cpu_only' does not exist on type '{ vae?: string | null | undefined; vae_precision?: "fp16" | "fp32" | null | undefined; scheduler?: "ddim" | "ddpm" | "deis" | "deis_k" | "lms" | "lms_k" | "pndm" | "heun" | "heun_k" | ... 22 more ... | undefined; ... 5 more ...; guidance?: number | ... 1 more ... | undefined; }'.
Error: src/features/modelManagerV2/hooks/useMainModelDefaultSettings.ts(46,47): error TS2339: Property 'cpu_only' does not exist on type '{ vae?: string | null | undefined; vae_precision?: "fp16" | "fp32" | null | undefined; scheduler?: "ddim" | "ddpm" | "deis" | "deis_k" | "lms" | "lms_k" | "pndm" | "heun" | "heun_k" | ... 22 more ... | undefined; ... 5 more ...; guidance?: number | ... 1 more ... | undefined; }'.
Error: src/features/modelManagerV2/subpanels/ModelPanel/MainModelDefaultSettings/DefaultCpuOnly.tsx(41,31): error TS2322: Type '"cpuOnly"' is not assignable to type 'Feature'.
 ELIFECYCLE  Command failed with exit code 2.
Error: Process completed with exit code 2.```

[Copilot]

@lstein I've opened a new pull request, #47, to work on those changes. Once the pull request is ready, I'll request review from you.

[lstein]

@copilot Can you make the front and backend changes needed to allow the user to change the cpu-only status of all the standalone encoder models? The best place to put the UI is in the model details pane for the various encoder models, using a layout similar to that used for Main models.

[Copilot]

@lstein I've opened a new pull request, #49, to work on those changes. Once the pull request is ready, I'll request review from you.