Safetensors import of Gemma3 fails quantization > 0.6.5

### What is the issue? As an avid user of the Gemm3 model I have been following this rather closely but, will admit some of the chatter is difficult to follow. The issue I found is that with Ollama v0.6.5 I was able to run the following command and load a Gemma3 Safetensors model into the library. (12b Int) ollama create Gemma3 -f Modelfile --quantize q8_0 This allowed me to use the image recognition features and was generally a good time. When upgrading to Ollama v0.6.6 (and later) the same command fails ``` converting model quantizing F16 model to Q8_0 Error: llama_model_quantize: 1 ``` I have included some relevant logs though I feel it is rather easy to reproduce. ### Relevant log output ```shell gathering model components copying file sha256:4667f2089529e8e7657cfb6d1c19910ae71ff5f28aa7ab2ff2763330affad795 100% copying file sha256:4847447e92599833e8dbaa3067cd201c3bb5c052efa91f11ba891e43234f7832 100% copying file sha256:891bd54eed03cba9ee1e705533a02a8217fcc29f356e4a1f53e5fd0d178883ad 100% copying file sha256:fe16baf728db49457cde32802cd7efc0ac8a7a9877dbe22fe3322b2d9dc6ccd9 100% copying file sha256:fd9324becc53c4be610db39e13a613006f09fd6ef71a95fb6320dc33157490a3 100% copying file sha256:ed14bd4908c98fed9f61e8cd410167e0846de9abd78e0452ab092072e5d9252d 100% copying file sha256:bfe25c2735e395407beb78456ea9a6984a1f00d8c16fa04a8b75f2a614cf53e1 100% copying file sha256:788cc42a1a92835df62d9a3791f47105f63504c7c404637a73288e9b11bc7b82 100% copying file sha256:39172c4124d3470341bbbb25f2926fd97edf68f0fe3a9fa4cde6acb9b7ed2cc6 100% copying file sha256:2f7b0adf4fb469770bb1490e3e35df87b1dc578246c5e7e6fc76ecf33213a397 100% copying file sha256:f688d6bb20c5017601c4011de7ca656da8485b540b05013efdaf986c0fcc918d 100% copying file sha256:50b2f405ba56a26d4913fd772089992252d7f942123cc0a034d96424221ba946 100% copying file sha256:1299c11d7cf632ef3b4e11937501358ada021bbdf7c47638d13c0ee982f2e79c 100% copying file sha256:8bc75a29a730c9e743cad013feda3b0991a913fafe787c58a1c6e20…