Tostino

27 ms·

721.
by Tostino 3y ago
For a society that has utterly failed them in the first place...not a chance in my book.
722.
by Tostino 3y ago
That's not how that works. At all.
723.
by Tostino 3y ago
Right, they are only good because the original GE has nothing to do with them now.
724.
by Tostino 3y ago
For a small business, floating the difference for that extra time period can be painful.
725.
by Tostino 3y ago
Doesn't mean it's not illegal, just that they won't punish themselves.
726.
by Tostino 3y ago
Not OP, but Librechat has been great for me. Single interface for local models, GPT4, Claude, etc. It just added support for configuring an arbitrary number of custom endpoints for your locally hosted models, and is improving a ton. https:
727.
by Tostino 3y ago
Also intimidates people into signing NDAs for things like sexual assault.
728.
by Tostino 3y ago
I hope OP fails fantastically. I don't want more companies putting up bullshit filters that have a ton of false positives.
729.
by Tostino 3y ago
They said collectively, meaning across the entire fleet of phones deployed.
730.
by Tostino 3y ago
Hmm, it's not showing up for me to try in my account... Here is the site talking about the features though: https://blog.youtube/news-and-events/new-youtube-premium-fea...
731.
by Tostino 3y ago
This can sometimes be fixed with a few-shot example for in-context-learning. But you are right that the model can go off the rails if it is being forced too far from where its 'happy place' is, especially for smaller models.
732.
by Tostino 3y ago
No, you will have much larger index bloat as you add rows as page splits happen to accommodate the new rows.
733.
by Tostino 3y ago
Yes... You had plenty of people saying "this is going to be your own version of Afghanistan/Iraq if you decide to do what you want to do..." Israel wanted revenge, not a solution. Their strategy has guaranteed that there wil
734.
by Tostino 3y ago
So use a non human algorithm to optimize that problem by maybe collecting some huristics during a benchmark phase, and then the quantization step can take that into account based on specific pathways that are never hit because another exper
735.
by Tostino 3y ago
That second point is personally the sticking point for me. No reason the less-well-off should subsidize those who have the ability to purchase a new vehicle.
736.
by Tostino 3y ago
I mean, we lost all our manufacturing in that time period. We have far more service jobs (e.g. not inherently dangerous) in the economy now compared to 1970.
737.
by Tostino 3y ago
I'm sorry, what are you talking about? Care to justify your position?
738.
by Tostino 3y ago
The additional features they added were quite nice though...until it got all the extra telemetry and bloat was added.
739.
by Tostino 3y ago
Them being very tailored to the main model is true, and it takes a lot of work for the lora author to target multiple base models, so they generally just target one. I plan on releasing my next iteration of my model (trained on generating k
740.
by Tostino 3y ago
Yes there is...HF chat templates is something that is being standardized on, slowly. It's just a jinja template embedded in the tokenizer that the model creator can include.
741.
by Tostino 3y ago
As a counter point, I cannot be sure that the words you are speaking mean the same thing to you in your mind as they do to me when I hear you.
742.
by Tostino 3y ago
Completely agree, even guilting it into doing more of the coding isn't working well any more. I have absolutely noticed the difference. It was right around gpt4 turbo introduction.
743.
by Tostino 3y ago
I don't know how many tokens are required to get good results, because I simply didn't mark mine as "special_tokens" due to the issues that I had read about. I got great results, whereas others who have tried special tok
744.
by Tostino 3y ago
Try doing the same thing in your dataset, but don't actually add them as "special tokens", and just let them just be multiple tokens. Adding new tokens needs a ton of data to train what the token means. Reusing existing token
745.
by Tostino 3y ago
IBM is first, and foremost to service's company. They just care about the government contracts to implement whatever solution is leading.
746.
by Tostino 3y ago
Not yet. Your faith in people not being led astray by the LLM is too high right now IMO. I've seen way too many people who I consider generally intelligent, completely falling for the convincing bullshit. Mostly in my online communitie
747.
by Tostino 3y ago
Heh, so it totally depends on the use case. I use GPT4 constantly to chat through issues I am working on and get different perspectives. I cannot do that with local models. On the other hand, I have been processing a ton of text transcripts
748.
by Tostino 3y ago
Yeah, i'll be training a Mistral variant soon (and some of the larger, newer models as well). I had a few dataset issues I know about that I wanted to fix first.
749.
by Tostino 3y ago
Honestly, I am excited for people to re-implement the API for this using open models because of this. Maybe let the API stabilize a little, and i'm sure some implementations will pop-up.
750.
by Tostino 3y ago
Yeah, I will be soon. I was busy adding `chat template` support to vLLM recently, so the model (and any others that implement it properly) will work seamlessly with a clone of the OpenAI chat/completions endpoint. https://gi

More ›