Gemma 4 E2B-it GGUFs hit shape mismatch

This looks like a launch-week compatibility bug in the GGUF ecosystem, not a hardware limitation. If the tensor layout in the file does not match what llama.cpp expects, no amount of extra GPU memory will make it load.

• –The error shows `blk.2.attn_q.weight` with the wrong dimensions, which is classic evidence of a model-format mismatch or stale conversion
• –Community chatter suggests some Gemma-4-E2B-IT GGUFs were broken or outdated, while reworked quants from other sources load successfully
• –Gemma-4-E2B-IT is being pushed hard for local multimodal use, so loader correctness matters as much as benchmark quality
• –For developers, the practical fix is to swap to a freshly rebuilt Gemma-4-E2B-IT GGUF and verify the exact quant/source, not just the llama.cpp version
• –This is a reminder that open-weight model releases can still be fragile at the packaging layer even when the model itself is fine