redox99

35 ms·

721.
by redox99 3y ago
That's definitely not true for private pilots.
722.
by redox99 3y ago
But that API isn't OpenAI compatible AFAIK
723.
by redox99 3y ago
Is there a way to make it work with ooba+exllama? (much faster than llamacpp)
724.
by redox99 3y ago
Your typical, old school airplane autopilot will happily crash into a mountain, not unlike how basic cruise control will happily crash into a wall.
725.
by redox99 3y ago
What's wrong with Autopilot? It's a very good analogy to autopilot in airplanes.
726.
by redox99 3y ago
It is fair in the context of another driver being worried about being hurt by a Tesla.
727.
by redox99 3y ago
> Something I’m not seeing yet in these comments is a discussion of the other automakers’ implementation of the same features. Because these threads always become Tesla hate instead of an objective analysis. All of these lane centering A
728.
by redox99 3y ago
Kind of hilarious what it would go for Python, of all languages.
729.
by redox99 3y ago
Interesting. That seems a pretty odd choice. Without a prompt format (like almost all instruction finetunes have) the model might continue your instruction instead of answering the instruction, since it does not have a delimiter for where t
730.
by redox99 3y ago
Yeah but I assume it has a prompt format, for example CodeLlama34B Instruct the prompt format is [INST] Write a function in Python that prints HelloWorld [/INST]
731.
by redox99 3y ago
Could you please add the prompt format to the model card, or at least here?
732.
by redox99 3y ago
I actually tried the 4 bit quants (Q4_K_M) and was a bit unimpressed. Switching to Q6_K made a huge difference, but it doesn't fit on my 3090 so it was very slow. And testing on perplexity's website which I presume is fp16 seemed
733.
by redox99 3y ago
34B is basically an exact fit for a 3090, when quantized to 4 bits.
734.
by redox99 3y ago
If all you care about is convenience, then anything cloud or SaaS will be better for you.
735.
by redox99 3y ago
I think it's kind of like Walter White being embarrassed to work as a high school teacher instead of as a professor at some fancy university.
736.
by redox99 3y ago
I assume he/she is editing it as we speak?
737.
by redox99 3y ago
On runpod there is a TheBloke template with everything set up for you. An A6000 is good enough to run 70b 4bit.
738.
by redox99 3y ago
Chassis intrusion is almost always just a small switch pushing against the side panel of the case. And yes there's a million ways you could bypass that.
739.
by redox99 3y ago
If you want to run them fast, a 12GB GPU (e.g 3060) for the 13B and a 24GB GPU for the 34B (e.g 3090). Otherwise llama.cpp CPU inference would work on most machines.
740.
by redox99 3y ago
The good thing is that if it's only finetuned on 15k instructions, we should see a community made model like that very soon.
741.
by redox99 3y ago
Probably the researchers at meta are only interested in research, and productionizing this would be up to other teams.
742.
by redox99 3y ago
Meh. If you're using it commercially you're probably deploying it on a server where you're not limited by the 24GB and you can just run llama 2 70b. The majority of people who want to run it locally on 24GB either want rolepl
743.
by redox99 3y ago
I can't imagine it being better than Llama1 33B, after all this code finetuning.
744.
by redox99 3y ago
Also Tesla a lot more room to grow in term of revenue (they're still a small % of cars solds) than nvidia (already dominates gaming and datacenter). I admit I'm biased in favor of Tesla however.
745.
by redox99 3y ago
The highlight IMO > The Code Llama models provide stable generations with up to 100,000 tokens of context. All models are trained on sequences of 16,000 tokens and show improvements on inputs with up to 100,000 tokens. Edit: Reading the
746.
by redox99 3y ago
This comment will age like milk, when in a few years (nvidia 50 series I'd say) path tracing becomes the norm for high end gaming.
747.
by redox99 3y ago
They are not similar at all. Tesla P/E is 67, Nvidia's P/E is 244.
748.
by redox99 3y ago
> However not only is that not happening, nobody is even trying. A lot of companies are trying. It's just that nvidia is really good and significantly ahead.
749.
by redox99 3y ago
It's interesting that the image has a "mirrored along the X axis" kind of look.
750.
by redox99 3y ago
Sensitivity and specificity?

More ›