LGAI-EXAONE

/

EXAONE-Deep-7.8B-GGUF

like 21

Follow

LG AI Research 487

same

Hello, thank you for asking.

We have updated the guide for LM-Studio users in our GitHub README. Please refer to this section for instructions.

If you encounter any issues while following the guide, please let us know.

Very impressive, even the 2.4B is on par with llama 3 8B from my several random questions. I wouldn't believe that's an LLM less than 3B before I tried this.

Very impressive, even the 2.4B is on par with llama 3 8B from my several random questions. I wouldn't believe that's an LLM less than 3B before I tried this.

Using example question in lm studio "What is the capital of France?", it thought 2:18 min, and the answer is too long + it continues thinking inside answer...So something is not right.
I did everything according to readme
Set "Reasoning Section Parsing" as <thought> and </thought>
Set "Template (Jinja)" as that of EXAONE 3.5.
Temp 0.6

Very impressive, even the 2.4B is on par with llama 3 8B from my several random questions. I wouldn't believe that's an LLM less than 3B before I tried this.

Using example question in lm studio "What is the capital of France?", it thought 2:18 min, and the answer is too long + it continues thinking inside answer...So something is not right.
I did everything according to readme
Set "Reasoning Section Parsing" as <thought> and </thought>
Set "Template (Jinja)" as that of EXAONE 3.5.
Temp 0.6

Those are mine. There is no reasoning process at all. You should disable the repeat penalty. It is very verbose, like 2k tokens for an easy question, but it's normal for thinking models.

Hi urtuuuu,

EXAONE-Deep-7.8B-GGUF seems to work well for the question, "What is the capital of France?".
Here are my output and configuration of LM Studio.

Looks like it works now, thanks :) But still, not just any question will make it think right?