Running Local LLMs with Ollama on openSUSE Tumbleweed
-
The 30B-A3Bs I've tried have been suuuuuuuper repetitive. Do you have any specific settings to recommend to get them to work well?
wrote on last edited by [email protected]Random thing, I did not get a notification for this comment, I stumbled upon it. This happens all the time, and it makes me wonder how many replies I miss...
I don't run A3B specifically, but for Qwen3 32B Instruct I put something like "vary your prose; avoid repetitive vocabulary and sentence structure" in the system prompt, run at least 0.5 DRY, and maybe some dynamic sampler like mirostat if supported. Too much regular rep penalty makes it dumb, unfortunately.
But I have much better luck with base model derived models. Look up the finetunes you tried, and see if they were trained from A3B instruct or base. Qwen3 Instruct is pretty overtuned.
-
Random thing, I did not get a notification for this comment, I stumbled upon it. This happens all the time, and it makes me wonder how many replies I miss...
I don't run A3B specifically, but for Qwen3 32B Instruct I put something like "vary your prose; avoid repetitive vocabulary and sentence structure" in the system prompt, run at least 0.5 DRY, and maybe some dynamic sampler like mirostat if supported. Too much regular rep penalty makes it dumb, unfortunately.
But I have much better luck with base model derived models. Look up the finetunes you tried, and see if they were trained from A3B instruct or base. Qwen3 Instruct is pretty overtuned.
They may have been based on Instruct. It left such a bad impression, I didn't play around with them much. Good to know for the future, though. I haven't used DRY or mirostat really in the past, but I'll try them next time I look at the Qwen3s.
-
They may have been based on Instruct. It left such a bad impression, I didn't play around with them much. Good to know for the future, though. I haven't used DRY or mirostat really in the past, but I'll try them next time I look at the Qwen3s.
wrote on last edited by [email protected]Honestly I don’t use Qwen3 instruct unless it’s for code or “logic.” Even the 32B is soo dry and focused on that, and countering it with sampling dumbs it down.
Not sure if it’s too big, but I have been super impressed with Jamba 52B. It knows tons of fiction trivia and writing styles for such a “small” model, though I haven’t tried to manipulate its prompt for writing yet. And it’s an MoE model like A3B.
-
Honestly I don’t use Qwen3 instruct unless it’s for code or “logic.” Even the 32B is soo dry and focused on that, and countering it with sampling dumbs it down.
Not sure if it’s too big, but I have been super impressed with Jamba 52B. It knows tons of fiction trivia and writing styles for such a “small” model, though I haven’t tried to manipulate its prompt for writing yet. And it’s an MoE model like A3B.
Interesting. I hadn't heard of this one before, but the design sounds innovative. The biggest I run is 35B or 7x8B, but I'll have to try and check it out.
-
Interesting. I hadn't heard of this one before, but the design sounds innovative. The biggest I run is 35B or 7x8B, but I'll have to try and check it out.
wrote last edited by [email protected]Jamba is a killer model flying under the radar, though it does have a quirk I more recently discovered: no prompt caching in llama.cpp (yet).
If you have a 24GB GPU you can cram Nemotron 49B in it with no offloading, including the new reasoning version. It’s a monster at STEM stuff, and I can upload my special quantization (3bpw, with 4bpw KV heads, exllamav3) if you ask.
Qwen 30B coder is ridiculously fast for how smart it is at coding, just came out today…
TBH the last week or two has been nuts with new releases.
But FYI if you are looking for pure prose quality, I still use EVA Gutenberg 32B (based on Qwen 2.5 base) and Jonboro's brand new QWQ 32B fine tune, as new models have not surpassed them IMO. But for creative writing, I tend to write novel style instead of multi turn, so YMMV.
-
Jamba is a killer model flying under the radar, though it does have a quirk I more recently discovered: no prompt caching in llama.cpp (yet).
If you have a 24GB GPU you can cram Nemotron 49B in it with no offloading, including the new reasoning version. It’s a monster at STEM stuff, and I can upload my special quantization (3bpw, with 4bpw KV heads, exllamav3) if you ask.
Qwen 30B coder is ridiculously fast for how smart it is at coding, just came out today…
TBH the last week or two has been nuts with new releases.
But FYI if you are looking for pure prose quality, I still use EVA Gutenberg 32B (based on Qwen 2.5 base) and Jonboro's brand new QWQ 32B fine tune, as new models have not surpassed them IMO. But for creative writing, I tend to write novel style instead of multi turn, so YMMV.
I only have a 16GB card, and my CPU is new enough that it's better to offload some layers of all but 7-8B models, so I haven't tried exllama, but you're making me think I should, if only for comparison.
I like Qwen 2.5 based models in the 14B size range, but I don't think I tried the bigger ones. I tried the QWQ and didn't really like it, but I haven't seen this new one. You've given me a whole list of things to try, so thanks.
-
I only have a 16GB card, and my CPU is new enough that it's better to offload some layers of all but 7-8B models, so I haven't tried exllama, but you're making me think I should, if only for comparison.
I like Qwen 2.5 based models in the 14B size range, but I don't think I tried the bigger ones. I tried the QWQ and didn't really like it, but I haven't seen this new one. You've given me a whole list of things to try, so thanks.
wrote last edited by [email protected]16GB
Is it 3000 series or newer?
If so, with exllamav3, you can squeeze 32Bs in that 16GB card with relatively little loss. For instance: https://huggingface.co/turboderp/EXAONE-4.0-32B-exl3/tree/3.0bpw
The 3bpw weights are 13 GB, say another 1.5GB for some q5_q4 context, and you are looking at 14.5GB-15GB or so. It will be tight, but it will be leagues smarter than 14Bs.
24B Mistral models will fit much more easily. No need to CPU offload those on a 16GB card, you just need to be careful with your settings.
-
16GB
Is it 3000 series or newer?
If so, with exllamav3, you can squeeze 32Bs in that 16GB card with relatively little loss. For instance: https://huggingface.co/turboderp/EXAONE-4.0-32B-exl3/tree/3.0bpw
The 3bpw weights are 13 GB, say another 1.5GB for some q5_q4 context, and you are looking at 14.5GB-15GB or so. It will be tight, but it will be leagues smarter than 14Bs.
24B Mistral models will fit much more easily. No need to CPU offload those on a 16GB card, you just need to be careful with your settings.
I have an rx 6800. Looks like exllamav3 doesn't support AMD cards yet... I'll keep an eye on it, so I can try it when ROCm or Vulkan support is added.
-
I have an rx 6800. Looks like exllamav3 doesn't support AMD cards yet... I'll keep an eye on it, so I can try it when ROCm or Vulkan support is added.
wrote last edited by [email protected]Ah. You can still run them in exllamav2, but you're probably better off with ik_llama.cpp then:
https://github.com/ikawrakow/ik_llama.cpp
It supports special "KT" quantizations, aka trellis quants similar to exllamav3, and will work with vulkan (or rocm?) on your 6800.
Quantizing yourself is not too bad, but if you want, just ping me, and I can make some 16GB KT quants, or point you to how to do it yourself.
It's also a good candidate for Qwen3 30B with a little CPU offloading. ik_llama.cpp is specifically optimized for MoE offloading.
-
Ah. You can still run them in exllamav2, but you're probably better off with ik_llama.cpp then:
https://github.com/ikawrakow/ik_llama.cpp
It supports special "KT" quantizations, aka trellis quants similar to exllamav3, and will work with vulkan (or rocm?) on your 6800.
Quantizing yourself is not too bad, but if you want, just ping me, and I can make some 16GB KT quants, or point you to how to do it yourself.
It's also a good candidate for Qwen3 30B with a little CPU offloading. ik_llama.cpp is specifically optimized for MoE offloading.
I got both IK and Croco built and running on Vulkan, but they both seem abysmally slow compared to Kobold, whether I run one of the GGUFs I already had or a trellis quant I just made. It's using the GPU, so not sure why. I'll keep an eye on the projects and try them again in the future.
