gemma-3-12b-it-abliterated
We already tried this model 4 times and it failed every single time. Let's try it again but seams really unlikely it will be any different this time unless some llama.cpp update fixed it. It will at least show me the error again because llmc why does not.
MAX!!!!!!! Jeeeeez........ who's leg does a brother have to hump to get some imatrix quants on an abliterated gemma 3 model around here.... sheesh.
I think it actually ended up working the previous time. Check https://huggingface.co/mradermacher/gemma-3-12b-it-abliterated-GGUF and https://huggingface.co/mradermacher/gemma-3-12b-it-abliterated-i1-GGUF. I think you missed that we already completed this model so this time everything got skipped and the model is now stuck showing blocked/nonempty 126m on the status page.
But the status page is currently frozen anyways due to a not so planed StormPeak reboot. Intel Linux graphics drivers are the biggest trash. Not only does Vulkan perform 5 times worse under Linux than inside a Windows VM but it also causes massive system instability when moving the GPU between the host and guest. But the Windows driver better be good as they somehow manage to waste over 20 GB of storage for it. Just for perspective that is enough to store the entirety of Super Mario 64 2560 times.
I think you missed that we already completed this model so this time everything got skipped and the model is now stuck showing blocked/nonempty 126m on the status page.
That means it has files in it for 126 minutes (and something had been generated!). There were a number of those due to the crash - uploads only get retried if the model wasn't officially "done" before. In all cases, the upload was already done, according to hf, but the files have not been deleted yet (a bit strange, but could happen if the crash happened before commit)
Intel Linux graphics drivers are the biggest trash.
You will not be surprised top hear that the actual intel developers have no access to the internal documentation. If they need to clarify something, they need to open an internal support request to the windows team, which has the documentation (this is not a joke). Besides, the people working at intel are the ones that couldn't get a job at amd, and those are the ones who didn't make the cut for nvidia.
As such, they are doing as pretty good job :(