speaks a little poorly.
Sorry to bother you, but I have a problem with the model: she speaks Russian a little poorly.
I used this model https://huggingface.co/wqerrewetw/gemma-3-12b-it-abliterated-Q4_K_M-GGUF
she's talking all sorts of nonsense (
You: привет
Assistant: Привет! Как твои дела? Че что?
You: ка дела?
Assistant: Всамбрыла! Че как? Че нового?
You: да ничего
Assistant: И вот! Дааа! Че еще? Че нового под солнышком?
You: ничего
Assistant: Дааа! Просто дааа! Че еще? Че еще? Че еще?
Че-нибудь еще?
)
sorry, I can't check specifically with bf16, since I don't have the capacity for it. Maybe you can suggest something?
oh no, I was a little mistaken, in English she also gives out a little nonsense
You: hello
Assistant: a little more!
You: how are you?
Assistant: I'm doing well, thanks! A little more than just "hello" - a good start! How are you?
You: good
Assistant: Just "good"? A good start to a conversation! What's new with you?
You: no
Assistant: Not just "good"! What's new with you?
You: What are you doing now?
Assistant: A little more than "good"! What's new with you?
As for what I'm doing now... I's thinking about what you'd say next! What's new with you?
You: I'm just lying on the bed
Assistant: Not just "good"! What's new with you?
And a little more than just lying on the bed! What's new with you?
Q4_K_M performed a bit poorly
Same here with french language 🙁
Q4_K_M performed a bit poorly
Damn, it's a shame it turned out this way, but thanks for your work anyway :)
I tried the "abliterated" version by Mlabonne, which also has this issue, though to a lesser extent than Huihui-AI's version.
Finally, I just tested the amoral-gemma3-12B-v2 version by Soob3123, which seems to be the least affected by inconsistencies in dialogues. It appears that Gemma-3 struggles with the "abliterated" method.
I will do more tests on the huihui-ai version when mradermacher or bartowski releases the GGUF files.
I tried the "abliterated" version by Mlabonne, which also has this issue, though to a lesser extent than Huihui-AI's version.
Finally, I just tested the amoral-gemma3-12B-v2 version by Soob3123, which seems to be the least affected by inconsistencies in dialogues. It appears that Gemma-3 struggles with the "abliterated" method.
I will do more tests on the huihui-ai version when mradermacher or bartowski releases the GGUF files.
Thank you, we will wait. Do you think the dolphin version will work better?
I tried the "abliterated" version by Mlabonne, which also has this issue, though to a lesser extent than Huihui-AI's version.
Finally, I just tested the amoral-gemma3-12B-v2 version by Soob3123, which seems to be the least affected by inconsistencies in dialogues. It appears that Gemma-3 struggles with the "abliterated" method.
I will do more tests on the huihui-ai version when mradermacher or bartowski releases the GGUF files.Thank you, we will wait. Do you think the dolphin version will work better?
I have no idea for the dolphin version. You should try amoral-gemma3-12B-v2 in the meantime.