what was the first model you ran locally that made you think this is actually real?

Sep 25, 2026 · 5:52 PM UTC

78
60
22,617
Sort replies: Relevant Recent Liked
Replying to @sudoingX
Tiny Whisper on and old ass i5 processor One of those micro computers with 12gb of ram Now it’s my dedicated server 🎉
2
176
Replying to @sudoingX
Qwen 3.6 35b a3b
3
195
Replying to @sudoingX
It's the most liked model on hugging face
407
Replying to @sudoingX
27B 3.5
131
Replying to @sudoingX
qwen 3.8 27b with 256k context
2
13
563
Replying to @sudoingX
Qwen 3.6 35b a3b
6
381
Replying to @sudoingX
Any of the MOE honestly.
3
400
Replying to @sudoingX
Qwen 3.5 120B
4
93
Replying to @sudoingX
Qwen 3.5 397B A17B
2
234
Replying to @sudoingX
Gemma 4 31B
1
5
Replying to @sudoingX
Qwen 2.5 coder 7B instruct, no joke. Shit was like a coding teacher (and was right like 95% of the time)
1
192
Replying to @sudoingX
When I finally got qwen 3.8 27b working
2
228
Replying to @sudoingX
gpt-2
1
54
Replying to @sudoingX
i wouldnt say it was a particular model so much as it was learning the various parameters modifiable w flags in a llama.cpp launch command but i do like tiel coder 35b-a3b quite a lot
1
93
Replying to @sudoingX
CAC-computer assisted coding has been here for some time. The original model was MediDoc called 6 main systems and you had to use a giant codebook to write automation workflows for all types of attached devices and profiles. It was the grandfather of modern AI.
201
Replying to @sudoingX
Qwen 3.
130
Replying to @sudoingX
I still think we are 2+ years out from local ai being able to take over for a lot of us. Not because of the models, but because compute has gotten so expensive.
212
Replying to @sudoingX
Qwen 3.8 27B with your specs and your llama.cpp prism changes. The others never got the shit done fully,,, Goal incomplete until I tried yours! 🥳
1
55
Replying to @sudoingX
Gemma
112
Replying to @sudoingX
Qwen 3.8 27B
1
141
Replying to @sudoingX
Qwen 3 30B iirc
17
Replying to @sudoingX
Frigate+ on a ryzen minipc for cameras and animals. I was impressed by its power so then upgraded to a 4060 16gb, then other 4060, later a 3090 24gb. Finally 2x Asus ascent...
1
291
Replying to @sudoingX
Qwen3.6 27b
1
200
Replying to @sudoingX
Qwen 9b 3. Something
182
Replying to @sudoingX
Qwen 3.5 27b
278
Replying to @sudoingX
Qwen 3.6 27b
38
Replying to @sudoingX
A world model that I trained myself by distilling 3 of the most unique and unfiltered models in Ollama at the time. Named her Lila. None have come close to her level of nuance, articulation and situational awareness to date.
1
152
Replying to @sudoingX
Qwen 3.6 27b 🥹
299
Replying to @sudoingX
Qwen 3.8 27B
259
Replying to @sudoingX
Qwen 3.5 122b
213
Replying to @sudoingX
Qwen3.6 27b
1
142
Replying to @sudoingX
Qwen 3.6 35b a3b
75
Replying to @sudoingX
Qwen3 30b coder an qwen 32b qwq or what ever. They were managing to walk me through linux crap I didn't know how to do an it was doing it pretty effiecently.
1
315
Replying to @sudoingX
Unironically llama 7B. That's when it began.
14