https://huggingface.co/ConicCat/Qwen3.5-Antirep-27B
https://huggingface.co/ConicCat/Qwen3.5-Antirep-27B
Another shameless self request.
Pretty much a finetune of Qwen using dpo to remove th multi turn repetition and looping issues which consistently plague Qwen models.
Thanks for all of your Quants!
It's queued!
You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Qwen3.5-Antirep-27B-GGUF for quants to appear.
Is it good? 👀
https://github.com/ggml-org/llama.cpp/pull/20119/changes
I looked into it a bit and the text version of 3.5 doesn't convert with lcpp as it was overlooked in the 3.5 pr, although there is a pr in the works for it now.
As for whether the model is good, I've had good luck with the same recipe on Mistral Small 3.
Unfortunately, I'm nowhere near vram rich enough to run the bf16 checkpoint to test w/o quantization, which, well ^^^
we are providing only quants only from released llama cpp version, so until this pr is not merged, I cant do anything about it. as soon as it is merged I ask nico to update llama cpp so I can queue it.
So please remind me when it is merged =)
sure =)
I asked nico to update, so when the model reaches the processing stage llama cpp will already be updated
You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Qwen3.5-Antirep-27B-GGUF for quants to appear.