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

Github repo Repo

preq playground