# Ten measured notes for diffusers, with runnable checks

> Source: <https://discuss.huggingface.co/t/ten-measured-notes-for-diffusers-with-runnable-checks/192226#post_2>
> Published: 2026-11-21 10:47:27+00:00

**Recommendation:** Run one-PR-at-a-time GPU checks for offloading fixes and apply both #15330 and #15333, since a repeated-call test with #15333 alone can mask a missing GPU onload.
More in the playbook "Inference and self-hosting": <https://wpnews.pro/playbooks/inference-serving>

I went ahead and tried running these on my side as well:

I used the ten [Diffusers rows in the Corrections Desk](https://huggingface.co/datasets/nootxlm/corrections-desk) as independent-reproduction targets. I started with small CPU models, then used an NVIDIA L4 for the offloading and image-pipeline checks. I also tested three proposed PR patches against **the same pinned development commit**, in isolated processes. In those controls, each of the ten rows had an independently observed matching failure or behavior. This is a second set of observations, not a claim that every production checkpoint, hardware backend, or released Diffusers version has been tested.

The most useful result beyond the individual reproductions was the interaction between the two group-offloading fixes. **Restoring a buffer-only module before a repeated forward** and **keeping a buffer’s value available when it is read without any forward** are different requirements. My one-PR-at-a-time GPU checks distinguish them. [#15330](https://github.com/huggingface/diffusers/pull/15330) and [#15333](https://github.com/huggingface/diffusers/pull/15333) each repair a different control; applying both passed both. A passing repeated-call test with #15333 alone can obscure the missing GPU onload if its forward happens to call `.to(input.device)`.

The ten Desk rows cover **nine GitHub issues**; HF-320 and HF-359 are two observations of [#15313](https://github.com/huggingface/diffusers/issues/15313). The table distinguishes my own run from what is reported in the Desk or an issue.

| Desk row | Issue | My independent result | Important limit | 
|---|---|---|---|
| **HF-320** | [#15313](https://github.com/huggingface/diffusers/issues/15313) | Registered `MyToken` , three vectors: prompt conversion yielded only`mytoken` . | My run: Diffusers 0.43.0 / Transformers 5.22.2, **not** the Desk’s 0.35.1 / 5.23.0 tuple. | 
| **HF-353** | [#15020](https://github.com/huggingface/diffusers/issues/15020) | Tiny T5 LoRA saved by Diffusers failed on reload; separate format/rank diagnostics exposed both gates. Mixed-rank patched replay matched all 16 weights and its forward output exactly. | The successful path used **temporary diagnostic monkeypatches** , not an upstream fix. | 
| **HF-359** | [#15313](https://github.com/huggingface/diffusers/issues/15313) | Token registered `neg` expanded to three tokens as`neg` ; prompt`Neg` remained one. | Tokenizer/prompt construction, not rendered-image quality. | 
| **HF-380** | [#15329](https://github.com/huggingface/diffusers/issues/15329) | A twice-invoked buffer-only module with disk offload plus streams returned a correct first result and nonfinite later results. | Actual Diffusers CUDA hooks on a **tiny module** , not gated LTX-2.5. | 
| **HF-397** | [#15337](https://github.com/huggingface/diffusers/issues/15337) | Real Z-Image-Turbo img2img failed on a requested size differing from its input; tiny img2img/inpaint did too. #15353 fixed the tiny GPU size controls. | No post-patch **pretrained** image-quality run. | 
| **HF-465** | [#15344](https://github.com/huggingface/diffusers/issues/15344) | The public HunyuanVideo musubi LoRA failed in the old converter; code containing the merged fix yielded 1,040 finite converted tensors. | Converter only, no full video model or generation. | 
| **HF-474** | [#15339](https://github.com/huggingface/diffusers/issues/15339) | One-block MagCache executed `[yes, no, no, no]` ; two-block control executed`[yes, no, yes, no]` . | Tiny random Wan model, not a production cache benchmark or repair test. | 
| **HF-484** | [#15332](https://github.com/huggingface/diffusers/issues/15332) | Direct buffer reads preserved values in **8/8** memory-offload checks and**0/8** disk-offload checks. | Tiny hooks, not a full FLUX.2 Klein render. | 
| **HF-511** | [#15342](https://github.com/huggingface/diffusers/issues/15342) | Red versus blue reference affected tiny Cosmos 3 output with internal noise, but the effect became **zero** with explicit raw latents, in image and video modes. | Tiny standard pipeline, not full pretrained or modular models. | 
| **HF-512** | [#15343](https://github.com/huggingface/diffusers/issues/15343) | In text-only Cosmos 3, supplied inactive sound/action latents raised the reported `AttributeError` /`TypeError` . | Active sound/action modes were not tested. | 

I found the cheapest informative comparison was usually **one positive control followed by one changed condition**. Use the [Desk checks](https://huggingface.co/datasets/nootxlm/corrections-desk/tree/main/checks) as entry points, then specify the imported Diffusers version or commit, dependency versions, device/dtype, weights, input dimensions, seed, and the **first diverging operation**. Some standalone checks deliberately model a reported code or shape condition without running an entire library pipeline. That is valuable, but it should not be labeled an end-to-end run. Likewise, an import error in a notebook is not evidence for or against the target library defect.

The fastest oracles were saved adapter keys/ranks, tokenized prompt output, buffer **value and device** on calls 1-3 (memory/disk, stream on/off), image/mask/noise shapes, real cache block executions, and Cosmos outputs under fixed noise. Those comparisons are more diagnostic than screenshots or timing. The [Diffusers reproducibility guide](https://huggingface.co/docs/diffusers/using-diffusers/reusing_seeds) helps with generators, but a shared seed does not guarantee identical pixels across hardware.

On an NVIDIA L4 I started each comparison process from the same [Diffusers `e84bda1` source](https://github.com/huggingface/diffusers/commit/e84bda1c0005da3f48cc4c899635438cb0aa13c1) (`0.44.0.dev0`), with only the indicated PR code added. This holds the development source constant rather than conflating a 0.43.0 wheel with a later checkout.

| Code variant | Repeat buffer-only module; forward moves the buffer to input device | Repeat module; subsequent computation needs buffer already on CUDA | Direct read outside owning module’s forward after disk offload | 
|---|---|---|---|
| Pinned base | First call correct; later nonfinite | Second call device error | Saved value lost | 
| **#15330 alone** | All three calls correct | All three correct | Value still lost | 
| **#15333 alone** | All three correct in this *auto-move* path | Second call still has device error | Saved value retained | 
| **Both** | All three correct | All three correct | Saved value retained | 

#15333 does **not** replace #15330: preserving a CPU buffer can make `.to(device)` succeed, but does not onload the registered buffer for a later CUDA operation. And onloading a module does not protect a buffer read without that module’s forward.

As checked on **November 21, 2026**, [#15330](https://github.com/huggingface/diffusers/pull/15330), [#15333](https://github.com/huggingface/diffusers/pull/15333), [#15353](https://github.com/huggingface/diffusers/pull/15353), and [#15335](https://github.com/huggingface/diffusers/pull/15335) were open. HunyuanVideo’s [#15346](https://github.com/huggingface/diffusers/pull/15346) merged on November 19. A PR being merged is not proof that a particular published wheel contains it.
