Add rule for detecting Stable Diffusion Web UI meta tensor corruption

[dhvll]

This PR adds a new Common Reliability Enumeration (CRE) rule to detect critical meta tensor corruption failures in Stable Diffusion Web UI applications. The rule identifies the specific error pattern NotImplementedError: Cannot copy out of meta tensor; no data! which causes complete service failure and prevents any image generation.

Root Cause

Meta tensor corruption occurs when PyTorch tensors lose their actual data while retaining only metadata (shape, dtype). This typically happens due to:

• Corrupted or incomplete model checkpoint files (safetensors/ckpt)
• PyTorch tensor corruption during model loading
• Device mismatch between CPU and GPU tensors
• Memory corruption during tensor operations

Error Pattern

NotImplementedError: Cannot copy out of meta tensor; no data!

Mitigation Strategies

Immediate Actions

1. Restart Stable Diffusion Web UI service to clear corrupted tensor states
2. Re-download and verify model checkpoint files
3. Check GPU memory and clear any corrupted tensor allocations

Preventive Measures

1. Implement model file integrity checks
2. Add tensor state validation before CUDA operations
3. Monitor GPU memory usage and tensor allocations

X Post Link

/fix #130
/claim #130

stable-diffusion.webm

Github repo Repo

preq playground

[dhvll]

@Lyndon-prequel can i get a review on this ?

[Lyndon-prequel]

@Lyndon-prequel can i get a review on this ?

@amanycodes

[amanycodes]

Hi @dhvll thanks for the submitting the CRE! Please add relevant tags to the rules/tags/tags.yaml to make the CI test pass. Also could you explain why you deleted multiple tags from the tags.yaml file?

[dhvll]

Because I was getting a merge error so.
Should I just get the copy of the latest tags.yaml file is it okay ?

[amanycodes]

mostly LGTM, please also add the applications field in the CRE yaml file, e.g:

applications:
        - name: "rabbitmq"
          version: "3.9.x"