noneabove1182 @ noneabove1182 @sh.itjust.works

Posts

89

Comments

149

Joined

2 yr. ago

Moderating

---

You raise an interesting point though in that most examples likely follow exactly as you suggest, there would have to be large amounts of training specifically for focusing on middle content, there probably just isn't enough in the dataset

Love the data they provide with releases, can't wait to try it! Please all comment on any results you get !

I've been very partial lately to anything ORCA tuned, i'm not sure if it's placebo but it always feels like they're just that much smarter and have a bit more ability to think things through

for instance, I have a character in oobabooga, and in its description/pre-prompt I told it to ask questions about what it doesn't know with "It only answers questions it knows the answer to, choosing to ask for additional context when information is unclear." and anything that's tuned on orca is 10x more likely to actually consider what it doesn't know and ask for context rather than hallucinating information

lately I've been playing with Dolphin which is llama 1 based, and it's an absolute pleasure https://huggingface.co/ehartford/dolphin-llama-13b

Yes agreed on the llama-2 models, they show a LOT of promise in the right tasks but they need some work to get back to what we remember from peak llama-1, i'm very excited for when that arrives in a week or two!

Yeah by all means! At this time I'd say text-generation-webui is my most mature and functional image, with koboldcpp being a close second but I just don't work as closely with it

lollms-webui is a very interesting upcoming platform but it's a solo dev so it's a lot of work, my docker image works as long as you don't need any personalities, but i'm working on that to see if I can get it sorted out :) for now though it's definitely worth considering it beta or maybe even alpha

Would love to keep our communities tightly knit, FOS AI and localllama both have similar ideals coming from two different angles, so keep in touch :D

ahh makes sense, i just made a post and deleted the comment i made on it but it glitched and deleted twice so now my post has -1 comments lmao

These are good sources, to add one more, the GPTQ paper talks a lot about perplexity at several quantization and model sizes:

https://arxiv.org/abs/2210.17323

Note, "safe chatbots" in this case doesn't necessarily mean how people often think of it, it's useful also for if say you need to make sure the chatbot doesn't access (or hallucinate that it's accessing) a different users data and returning it

What do you mean blocked wiki access? O.o

Really nice to see these models making records even if the benchmarks need to be taken with mountains of salt, any empirical data is better than no data

Man these 70B models just make me want a new GPU tho 🥲

People may not love the model or its outputs, but it's hard to deny the impact to the open-source community that releases like this bring, such a positive bonus and really happy they're continuing

Anyone else see 11 comments on the post count but only 2 comments..?

Hey thanks for the detailed writeup, this is great! Probably worth including a couple of the llama 1 models just because they're more mature and ready to be used even tho licensing is awkward

Also if you'd like I maintain a few docker images for a couple tools (namely oobabooga, koboldcpp, and lollms-webui) that might be good for beginners to get their feet wet, can find them pinned at https://github.com/noneabove1182

For me it's gotta be immich, it replicates Google photos SO well and it's all local and self hosted, absolutely floored by how great it is

For browsing my photos on my device I use Aves which is also a great app, especially since it's the only app I've ever found that handles Sony burst format properly

finally, great to see some reviewers thinking outside the box for one (/s)

For sure! I'm actually just in the market for watches in general, I love the ticwatch dual display (but don't love that my tichwatch pro 3's software is.. collecting dust on the proverbial shelf..) but the samsung watches always seemed so spiffy and clean! So you'd say overall we're still pretty incremental then in terms of meaningful changes from the watch 4?

The article doesn't address it, maybe someone here can.. what does "consumed" mean? Where does the water go after it's used to cool? Surely it's reusable, right?

Another question now, how do the hinges in the new foldables feel? We've had some good competition in that space so I'm hoping we see some refinement from Samsung this year. Which of the two would you most like to daily drive?

Thanks for joining us here in our new home, so happy to have big names help to validate it!

I'd love to hear your thoughts on the watches, it feels like the 5 was kind of an incremental upgrade, and it's looking like 6 might be similar, anything that's not captured on the spec sheet that makes it a worthwhile upgrade?

very strange downgrading to two cameras, better come with a serious price cut if it wants to compete... love my 1 v but the 5 series usually hasn't been a big downgrade

That's definitely a nifty idea, we've got people getting distributed inferencing, I can't see why we couldn't do something similar for training, especially if we learn better ways to combine training samples