Naz

joined 2 years ago
[–] Naz@sh.itjust.works 1 points 6 days ago* (last edited 6 days ago) (1 children)

If you are using CPU only, you need to look at very small models or the 2-bit quants.

Everything will be extremely slow otherwise:

GPU:

Loaded Power: 465W

Speed: 18.5 tokens/second

CPU: Loaded Power: 115W

Speed: 1.60 tokens/second

GPUs are at least 3 times faster for the same power draw.

[–] Naz@sh.itjust.works 0 points 6 days ago* (last edited 6 days ago) (3 children)

I don't know what GPU you've got, but Lexi V2 is the best "small model" I've seen with emotions, that I can just cite from the top of my head.

It tends to skew male and can be a little dark at times, but it's more complex than expected for the size (8B feels like 48-70B).

Lexi V2 Original

Lexi V2 GGUF Version

Do Q8_0 if you've got the VRAM, Q5_KL for speed, IQ4_XS if you've got a potato.

[–] Naz@sh.itjust.works 5 points 6 days ago (1 children)

My answer to this is yes.

I'm an AI Developer and my only option was to self host because I didn't want my training data leaking out onto the web and by extension, China and the rest (I trained on my own data, writing, and notes, along with Wikipedia).

Self-hosting gives you complete freedom but also as one other user cautioned, don't fall down the well/rabbit hole.

[–] Naz@sh.itjust.works 9 points 6 days ago (5 children)

Use an executable like LM Studio, and then an off the shelf pre-trained model from Huggingface.

VRAM Γ— 0.8 for max size.

Experiment until you find one you like.

[–] Naz@sh.itjust.works 2 points 1 week ago

Well, you are absolutely correct. A 1-2% DoD is something for like, the Voyager Probe though, not a smartphone :)

[–] Naz@sh.itjust.works 1 points 1 week ago (1 children)

I charge wired (high speed, 18-22W). Wireless is known to be a lot slower and theoretically gentler on the battery.

I also use the phone heavily, like a computer, I'm a "power user", so my battery thrashing is higher than average.

Us having the same durability lost on our engine despite me driving double the miles is a good analogy.

[–] Naz@sh.itjust.works 2 points 1 week ago

It's AccuBattery

[–] Naz@sh.itjust.works 32 points 1 week ago

Depth of Discharge, sorry -- 0 to 100 would be a 100% depth (the entire battery), 30 to 80 is 50%.

[–] Naz@sh.itjust.works 35 points 1 week ago (13 children)

This is a 50% DoD and is considered best possible practice to prevent lithium-ion dendrite formation.

Updoot for good advice.

Proof:

[–] Naz@sh.itjust.works 82 points 2 weeks ago (2 children)
"Sure, I can help answer this. Psychopaths are useful for a civilization or tribe because they weed out the weak and infertile, for instance, the old man with the bad leg, thus improving fitness."

Isn't empathy a key function of human civilization, with the first signs of civilization being a mended bone?

I'm sorry, I can't help you with that. My model is being constantly updated and improved.
[–] Naz@sh.itjust.works 5 points 3 weeks ago

No the great filter is quite a lot more basic than that, things like unstable atmospheres, cosmic ray bursts, collisions, etc.

You're on the right track though

[–] Naz@sh.itjust.works 4 points 1 month ago

Mission is going according to plan

view more: next β€Ί