Added support for lowvram argument for 'mps' devices.

[FromFirstByte]

Right now VRAM-related args are ignored if cpu_state == CPUState.MPS

If --lowvram arg passed it would prevent from loading 100% of weights to VRAM. With modified GGUF loader it allows to keep memory-mapped weights on disk significantly reducing RAM usage. Modified GGUF loader - https://github.com/FromFirstByte/ComfyUI-GGUF

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 2006597b-ce75-4183-8898-31d94434564a

📥 Commits

Reviewing files that changed from the base of the PR and between abbe30c and 489fc75.

📒 Files selected for processing (1)

• comfy/model_management.py

---

📝 Walkthrough

Walkthrough

In load_models_gpu, the fallback path for lowvram_model_memory now also applies when set_vram_to is LOW_VRAM, vram_state is SHARED, and force_full_load is false. The existing NO_VRAM case remains included.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the new low-VRAM support for MPS devices.
Description check	✅ Passed	The description matches the change by explaining that lowvram behavior now applies to MPS devices.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

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.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@comfy/model_management.py`:
- Around line 941-942: The new shared-lowvram fallback in the model memory
selection logic is applying too broadly and can affect CPU-loaded models. Update
the condition in the memory-setting block around the VRAMState check so the
shared-lowvram path only forces lowvram_model_memory when the current model’s
torch_dev is a non-CPU load device, preserving the existing behavior for CPU
loads. Keep the change localized to the lowvram memory decision in the relevant
model management function and avoid altering other VRAMState handling.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 19caa59a-e498-497a-940f-0b0aba97665f

📥 Commits

Reviewing files that changed from the base of the PR and between 64e1d74 and abbe30c.

📒 Files selected for processing (1)

• comfy/model_management.py