Posts
13158
Following
561
Followers
562
In serious need of a hug
Why do I want my liver to be eaten by a foxxy lady?
Typos are a spook

"Not the worst person on the instance"
@meso it's production code at my company lmao
3
0
3
@meso if i coupd i'd give you the rag tool i made for my finals
1
0
2
@meso vectorization is performed by a seperate specialised ai model
0
0
2
@meso retrieval augmented generation. You use vectorization to categorize parts of your text and can then draw up the closest matches and feed just those to an llm
2
0
3
@meso you'll prob need some kind of rag setup of you want to query over your entire traffic law tbh. Maybe there's some rag in a box thing but i'm not aware of any
2
0
1
repeated
I may no longer thirstpost on the other account, so let's do it here
4
1
7
Coax is magic
0
0
0
@chjara oh, it's osu. Thought for some reason that's a fedi profile in the background
1
0
1
Bicycle work is cool because i always do it better after the first time proving i'm not ocmpletely retarded and actually capable of learning
0
0
2
@meso oh yeah, imatrix quants are better than others
0
0
0
@meso usually llms are using 16bit weights nowadays but you can just use lower resolution
1
0
0
@meso as long as it's in vram you should see 50-100 tokens per sec ig
0
0
1
@meso just get a model that's smaller than the vram, usually you need about a gb less. Some 4bit quants of 12b and all of 8b models should fit
2
0
1
Therecs some steel wiring loose inside my frame now...
0
0
0
@meso sometimes, sometimes just a syllable or a single letter. Usually about 3-4 letters per token
1
0
1
@meso if you go to regular ram it's gonna get slow, like less than a token per second
1
1
1
@meso that's still enough for 4 bit quants of 12 or 8b models i think
1
0
1
Should've cut the frayed part before pulling...
0
0
0
Show older