segmondy

15 ms·

361.
by segmondy 8mo ago
There's a very big difference between xenophobia and racism. Racism is much worse.
362.
by segmondy 8mo ago
Interesting ... the police in this case are claiming to be the owners of the camera. https://oaklandcounty115.com/2026/03/03/clarkston-man-accuse...
363.
by segmondy 8mo ago
It's already bad enough that people are saying there's too much code to read and review. You want to add session to it? Running it again, might not yield the same output. These models are non deterministic and models are often
364.
by segmondy 8mo ago
Don't trust Google, will they send the data to their servers to "improve the service"?
365.
by segmondy 8mo ago
What is the goal?
366.
by segmondy 8mo ago
It's right there https://unsloth.ai/docs/models/qwen3.5/gguf-benchmarks I looked at the weights before. It's not PR fluff, they made it clear by showing how it really affected various tensors terr
367.
by segmondy 8mo ago
each layer is made up of various weights, the weights are adjusted to quant it. a pure q8 will have all the weights as q8, or a q4 the same. but some are kept as f32, etc. here's an example of q3_k_xl - https://huggingfa
368.
by segmondy 8mo ago
It's probably how they are coping with the cognitive dissonance. I certainly feel for them, I don't know that I could easily walk away from a big pay package either without backup options when I have family to support and I'
369.
by segmondy 8mo ago
My point being that no one should be buying expensive GPUs when you can pick up a few used ones to get started. But for the sake of discussion let's say you do get a blackwell pro 6000 that's now going for $10,000. I can assure
370.
by segmondy 8mo ago
llama.cpp is designed for partial offloading, the most important part of the model will be loaded into the GPU and the rest on system ram. I run 500B+ models such as DeepSeek/KimiK2.5/GLM-5 without having that much GPU vram.
371.
by segmondy 8mo ago
GPUs are not going obsolete anytime soon. the nvidia p40/p100 launched in 2016, 10 years ago and is popular in the local space. My first set of GPUs were a bunch of P40s from 3 years ago for $150 a piece. They at one point went up
372.
by segmondy 8mo ago
the macs outperform it and I figure it's a better general purpose computer than strix halo. if budget is a problem, then a strix halo is a decent alternative.
373.
by segmondy 8mo ago
We are not in this thread because of finances but because of safety from oppressive governments and bad big corps. It's for you to decide the price of your own safety.
374.
by segmondy 8mo ago
qwen3-8b is good and if you are doing tab completion then it's more than adequate. you can get basic agentic with it, but if you really want to use a serious agent and do some serious work, then at the very least qwen3.5-27B if you ha
375.
by segmondy 8mo ago
use llama.cpp, you will be surprised how fast a model like qwen3.5-35b-a3b will run. that a3b means only 3B active parameter, so while infering the entire 3B will be in your GPU and you will get amazing performance. for your system, you s
376.
by segmondy 8mo ago
if you can run Q8, go for it, always go for the best. matters a lot with vision models, never quantizie your kv cache, those always at f16. you can always try evals and see if you have a q6 or q4 that can perform better than your q8. for
377.
by segmondy 8mo ago
You can't be this naive?
378.
by segmondy 8mo ago
10 years is a long time, but 10 years after a war is not a long time. Damages to building still remains, mines and plenty of unexploded ordinances will remain, and psychological scars are still very strong.
379.
by segmondy 8mo ago
This is a good time to promote running your own models. I have been running my own models locally and I would wager a local model will meet 85-95% of your needs if you really learn to use it. These models have gotten great. For anyone wa
380.
by segmondy 8mo ago
Your last statement is false. A shady merchant never gets to keep the stolen money. The card issuer/bank refunds you immediately because of consumer protection laws. But that charge is immediately charged to the processor. The pr
381.
by segmondy 8mo ago
read the page, the anonymous signatures are verified. if not, the entire internet will abuse that page.
382.
by segmondy 8mo ago
Anthropic will just move out of the US. A lot of scientists fled Nazi Germany in the early stages. A lot of them fled to USA and end up being part of the Manhattan project that gave the Abomb that helped US win and end the war. We are go
383.
by segmondy 8mo ago
Not just that, but with how unfriendly we have been to the world, there's no guarantee that they will keep sharing as they have in the past.
384.
by segmondy 8mo ago
if it makes you feel better, we got into baking during covid and never baked sourdough once. we made pizza, cake, muffins, banana bread, regular bread, cornbread, etc. we just didn't post about it online ...
385.
by segmondy 8mo ago
Nice, I have always felt the computer was the ultimate environment and screen capture the ultimate training data. Nice to see it in practice, now we have to wait to see if folks are going to argue on if your model could really learn a worl
386.
by segmondy 8mo ago
He should have kept it closed.
387.
by segmondy 8mo ago
Pretty cool, what they need is to build a tool that can take any model to chip in short a time as possible. How quick can they give me DeepSeek, Kimi, Qwen or GLM on a chip? I'll take 5k tk/sec for those!
388.
by segmondy 8mo ago
very cool, weights? training recipe?
389.
by segmondy 8mo ago
It will be worth learning vim in 2076.
390.
by segmondy 8mo ago
part of what discussion? anyone in the AI space knows and uses HF, but the public doesn't give a care and why should they? It's just an advanced site were nerds download AI stuff. HF is super valuable with their transformers

More ›