Thread 106187009

13 posts 4 images /g/

Anonymous 8/8/2025, 8:19:26 AM No.106187009 >>106187242 >>106188127 >>106189055 >>106190320 >>106192534

1753983366768197.png md5: bc63ed2e... 🔍

i like gpt-oss:20b, pretty decent for it's parameters

Anonymous 8/8/2025, 9:03:56 AM No.106187242

>>106187009 (OP)
>135tok/s
Yeah, it's pretty nice.

Anonymous 8/8/2025, 9:50:14 AM No.106187502

hart

Anonymous 8/8/2025, 9:55:42 AM No.106187533 >>106187552

none of this shit from hugging face is up to scratch. it's a total waste of a time compared to the premium stuff.

Anonymous 8/8/2025, 9:58:54 AM No.106187552 >>106188175

>>106187533
The VRAM moat is real.

Anonymous 8/8/2025, 11:44:09 AM No.106188122

>mfw the 2020 dodge charger is a four door

Anonymous 8/8/2025, 11:44:50 AM No.106188127

>>106187009 (OP)
>>>/vt/ go back to your containment board

Anonymous 8/8/2025, 11:50:39 AM No.106188175

>>106187552
part of it is VRAM, but a large part of it is the extra, non-model things they don't talk about that actually QA the output before sending it to you.

Anonymous 8/8/2025, 12:10:17 PM No.106188302

whats that child doing under my desk?

Anonymous 8/8/2025, 2:06:21 PM No.106189055

>>106187009 (OP)
It's literally just a shitty reasoning finetune on top of 1.5 generation old weights. Llama-2-13B-Chat mogs on it.

Anonymous 8/8/2025, 4:22:08 PM No.106190320 >>106190724

>>106187009 (OP)
>i like the thing
enough to make an entire thread? wow

Anonymous 8/8/2025, 5:00:36 PM No.106190724

>>106190320
It's a slide thread
>Lust inducing picture
>time wasting question

Anonymous 8/8/2025, 7:57:57 PM No.106192534

1749309678773725.jpg md5: ec4fc268... 🔍

>>106187009 (OP)
The reasoning is much efficient than Qwen3. They take for-fucking-ever.