Fix MiniMax H3 audio VAE DynamicVRAM thrashing

[Deno2026]

Summary

• fully load the MiniMax H3 audio VAE instead of using DynamicVRAM weight streaming
• keep global DynamicVRAM behavior and the MiniMax H3 video VAE unchanged

Problem

MiniMaxH3AudioVAE is about 577 MB, but it currently inherits disable_offload = False. That selects CoreModelPatcher, so its weights are streamed during decode. In the reproduced Windows/NVIDIA path this caused repeated VRAM transfers and made a 5-second audio decode take about 153 seconds.

This also avoids the exact 576MB Staged, 401 weights DynamicVRAM path reported in #15276 and the broader host-buffer failures discussed in #15255.

Change

Set the VAE-owned disable_offload flag when the MiniMax H3 audio checkpoint is detected. This selects the regular ModelPatcher and the existing full-load decode path, matching other audio VAE integrations that disable offloading.

The tradeoff is keeping about 577 MB resident while the audio VAE is loaded. Global DynamicVRAM remains enabled for larger models.

Validation

• python -m py_compile comfy/sd.py
• git diff --check
• loaded minimax_h3_audio_vae_fp32.safetensors through comfy.sd.VAE
• verified disable_offload == True, ModelPatcher, and is_dynamic() == False
• decoded a [1, 32, 2, 200] 5-second latent to finite [1, 160000, 2] audio
• RTX PRO 6000 Blackwell, PyTorch 2.9.1+cu130: 0.447 s decode, 686.83 MiB CUDA peak
• ComfyUI queue smoke test logged loaded completely; 577.08 MB loaded, full load: True and completed the prompt in 0.70 s

[github-actions]

✅ All contributors have signed the CLA. Thank you! This PR is ready to be merged.
_{Posted by the CLA Assistant Lite bot.}

[Deno2026]

I have read and agree to the Contributor License Agreement

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 0b374570-59de-438f-a8a7-ec34ce6a1fa4

📥 Commits

Reviewing files that changed from the base of the PR and between 88fec4b and db7f6cf.

📒 Files selected for processing (1)

• comfy/sd.py

📜 Recent review details

🧰 Additional context used

📓 Path-based instructions (6)

**/*

📄 CodeRabbit inference engine (AGENTS.md)

**/*: Keep changes small, direct, and limited to the narrowest necessary code path and smallest number of files.
Prefer practical fixes, minimal dependencies, and existing repository patterns; remove obsolete, dead, unreachable, or unused code.
Preserve existing APIs, node names, model-loading behavior, file layout, and workflow compatibility unless replacement is explicitly intended.
Core ComfyUI must not add outbound internet requests, telemetry, tracking, reporting, remote configuration, or background network activity. User-authorized model downloads are limited to the requested artifact and must exclude telemetry and unrelated metadata.

Files:

• comfy/sd.py

**/*.py

📄 CodeRabbit inference engine (AGENTS.md)

**/*.py: Keep state and capability flags on the object that owns the behavior. Prefer explicit parent-owned attributes over probing child objects with getattr; use child checks only when the child owns the delegated behavior.
Preserve shared method signatures, argument order, return shapes, side effects, and error behavior unless every affected caller and interface is intentionally updated.
Do not add unused compatibility parameters, flags, attributes, constructor options, or model-specific options to shared helpers; keep one-off behavior at the integration boundary.
Normalize third-party return conventions at integration boundaries so core code receives the expected type and shape; avoid undocumented caller-side unwrapping.
Do not add torch.no_grad, torch.inference_mode, or inference-mode wrappers. Do not add model freeze/unfreeze toggles; only disable globally enabled inference mode when a training path requires gradients.
Remove inference-only training behavior such as dropout while preserving checkpoint and state-dict compatibility; use nn.Identity when deleting a module would alter keys or ordering.
Keep imports at module scope except established optional-backend probes or imports required to avoid cycles; avoid unnecessary try/except blocks and use specific exceptions with useful fallbacks.
Do not add workarounds for unsupported library versions, especially PyTorch exception-and-float-cast retries, unless a comment names the exact versions still requiring them.
Let unsupported model formats, invalid quantization metadata, and bad states fail with clear errors instead of silently degrading output.
Match local style, keep comments sparse and useful, and remove comments that merely restate obvious code.
Treat dtype, device placement, VRAM use, and offloading as correctness concerns across CPU, CUDA, ROCm, MPS, DirectML, XPU, NPU, and low-VRAM environments.
Prefer existing ComfyUI and Comfy Kitchen operations, quantization helpers, cast/offload helpe...

Files:

• comfy/sd.py

**/*.{py,json}

📄 CodeRabbit inference engine (AGENTS.md)

Treat legacy combo, io.Combo, and io.DynamicCombo values affecting filesystem access as untrusted; revalidate them at load/save boundaries with folder_paths, containment checks, or fixed allowlists.

Files:

• comfy/sd.py

**/*.{py,md,txt,json}

📄 CodeRabbit inference engine (AGENTS.md)

Keep warning and info messages short and actionable, remove noisy or misleading logging, and make documentation edits concise, factual, and tied to changed behavior.

Files:

• comfy/sd.py

**

⚙️ CodeRabbit configuration file

**: IMPORTANT: Only comment on issues directly introduced by this PR's code changes.
Treat AGENTS.md as mandatory repository policy, not optional style guidance.
Flag PR changes that violate AGENTS.md even when the code is otherwise functional.
In particular, enforce architecture boundaries, dtype/device/memory rules,
interface contracts, import style, no unnecessary try/except blocks, no inline
imports, no outbound internet paths in core ComfyUI, and narrow scoped fixes.
Prefer direct findings over suggestions when a rule is violated. Only ignore
AGENTS.md when it clearly conflicts with a newer explicit maintainer instruction
in the PR.
Do NOT flag pre-existing issues in code that was merely moved, re-indented,
de-indented, or reformatted without logic changes. If code appears in the diff
only due to whitespace or structural reformatting (e.g., removing a with: block),
treat it as unchanged. Contributors should not feel obligated to address
pre-existing issues outside the scope of their contribution.

Files:

• comfy/sd.py

comfy/**

⚙️ CodeRabbit configuration file

comfy/**: Core ML/diffusion engine. Focus on:

• Backward compatibility (breaking changes affect all custom nodes)
• Memory management and GPU resource handling
• Performance implications in hot paths
• Thread safety for concurrent execution

Files:

• comfy/sd.py

🧠 Learnings (3)

📚 Learning: 2026-02-21T14:01:41.482Z

Learnt from: pythongosssss
Repo: Comfy-Org/ComfyUI PR: 12555
File: comfy_extras/nodes_glsl.py:719-724
Timestamp: 2026-02-21T14:01:41.482Z
Learning: In PyOpenGL, bare Python scalars can be accepted for 1-element array parameters by NumberHandler. This means you can pass an int/float directly to OpenGL texture deletion (e.g., glDeleteTextures(tex)) without wrapping in a list. Verify function-specific expectations and ensure types match what the OpenGL call expects; use explicit lists only when the API requires an array.

Applied to files:

• comfy/sd.py

📚 Learning: 2026-05-13T12:31:45.069Z

Learnt from: rattus128
Repo: Comfy-Org/ComfyUI PR: 13802
File: comfy/pinned_memory.py:19-30
Timestamp: 2026-05-13T12:31:45.069Z
Learning: When reviewing code that uses comfy/pinned_memory.py’s `HostBuffer.extend(size=..., reallocate=...)`: by default (`reallocate` is not True / False), `extend(size=...)` is a *relative increment* that grows the buffer by `size` bytes—so slicing like `[offset:offset+size]` after `hostbuf.extend(size=size)` is correct and the argument should not be rewritten to `offset + size`. Only in the single-segment reallocation mode (`reallocate=True`, e.g., as used by `resize_pin_buffer()` in `comfy/model_management.py`) should `size` be treated as an *absolute target* and the call/arguments should be checked accordingly.

Applied to files:

• comfy/sd.py

📚 Learning: 2026-08-06T22:18:59.719Z

Learnt from: kijai
Repo: Comfy-Org/ComfyUI PR: 15362
File: comfy/ldm/wan/model_animate2.py:186-223
Timestamp: 2026-08-06T22:18:59.719Z
Learning: When reviewing ComfyUI quantization code, treat `comfy.quant_ops.TensorWiseINT8Layout` and `comfy.quant_ops.TensorCoreConvRotW4A4Layout` as re-exports from `comfy_kitchen`. Validate their behavior against the re-exported `comfy_kitchen` implementations rather than assuming they are local fallback classes.

Applied to files:

• comfy/sd.py

🔇 Additional comments (1)

comfy/sd.py (1)

981-981: LGTM!

---

📝 Walkthrough

Walkthrough

The MiniMax H3 audio VAE configuration now sets disable_offload to True. This selects the non-offloading model patcher path.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the MiniMax H3 audio VAE DynamicVRAM issue that this change fixes.
Description check	✅ Passed	The description directly explains the offloading change, its purpose, scope, tradeoff, and validation results.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Create stacked PR
• Commit on current branch

---

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.}

[rattus128]

what are the repeated VRAM transfers? This doesnt add up. CoreModelPatcher does not imply a repeating transfer (aka async offload). Something else must be going on.