Poll Discussion

[viscid sage]

I feel like we all should atleast get a tiny free trial maybe for one hour or two in testing v3.5. Since most of us are voting for v3.5 simply because it's the newest. We should atleast have some knowledge on how it performs first before voting

[naive leaf]

A limited run of credits maybe??

[pastel cave]

I think it would be better

[sterile quarry]

Yeah it would b smart to not delete 2.1 code instantly just encase 3.5 doesn't work out.

[craggy ice]

yeah thats the right call

it's a lot better to get our feet wet BEFORE we start diving in without looking

it also gives reason to be a supporter

that or it eats up credits like, a lot quicker

[stark horizon]

Yes

To be honest I didn't hate how much v2.5 generated text it was good at roleplay but it had obvious had some lackings compare more optimized 2.1

2.1 was very good at both tbh but special at single character roleplay i will most of the people here only chat with one character not multiple so i can see why everyone wants 2.1 back (including me also)

Also tbh even if pawan did give Access to 3.5 there should be a limit and also bring back the 2.1 then it can work to satisfy both group of people who wants 3.5 and most of the people (including me x2) who wants v2.1 back

I am an old user so i can actually feel how much better the 2.1 was compare to 2.5 tbh i liked both i am not gonna be ungrateful but still v2.1 was so so good and phenomenal compare to 2.5 in my opinion

[sterile quarry]

The issue with running both versions is server cost. It would be very expensive to run a 2.1 server, a free 3.5 server and a subscriber server for 3.5. so that's three dedicated servers with other costs such as bots.

[exotic arch]

I say just having 3.5 will be a blast, unless if its for a limited amount of time.

[stark horizon]

Exactly v2.1 was already very good tbh we don't even need the premium one anyways

The 2.1 was so well optimizated and was natural ngl

[stark horizon]

That's why bringing back v2.1 is the best idea

[tardy kiln]

Let's go back to version 3.5 being paid only and version 2.1 being free only.

It may be more expensive, but the adaptation will be worth it.

[viscid tartan]

I would really like 2.1 back, having a completely new model with only 8k tokens would be bad for us in my opinion.

Atleast new to the non payers

I kind of feel bad on how people just voted for the higher number without thinking, 8k tokens is really small. 2.1 had enough tokens and it was good enough in my honest opinion.

[craggy ice]

some people don't see it as small. I do, though

[tepid crystal]

yeah...I think 2.1 is better because I don't know how to make 2.5 give long message lol

[carmine niche]

3.5 is quite volatile with its responses, it would be nice to have all options based on needs ngl.

[barren willow]

I think the V2.5 it's fine but the "adaptability of chatting" Is annoying cuz sometimes even though I use long word it's still reply with short

They're just need to improve the "adaptability of chatting" Or just remove it completely cuz it's not really important "for me"

[proud crane]

I've been thinking the same thing about this, and not specifically just because of the fact it's 8k tokens but specially due to the fact 2.5 was going to be free while 3.5 is meant to be a higher model for those who support pawan

[exotic arch]

Sure, I just hope they can handle the cost and all

[still cobalt]

Maybe a solution is to open source 2.1 so people can run it locally?

[viscid tartan]

If someone knows how to run it locally and has the resources to do it, why would they be here though?

[proud crane]

Because they would need the API in of itself to run. More expensive models aren't open source

[viscid tartan]

That makes sense

[barren willow]

2 days left and we will get v3.5 with 8k context🥀

[lost whale]

is v3.5 that bad to the previous v2.1?

[proud crane]

I believe it has more to do with the fact it has only 8k context

[daring schooner]

no, 3.5 is very good, it's 100% about the 8k context

[stark horizon]

Just give us back the v2.1 tbh

[proud crane]

Which is sadly unlikely due to how the polls are going

[tepid crystal]

It's really that good? Compared to 2.1

[oak abyss]

Oh my god yes. using the 2.5 and previously 2.1 I'd have to do about 10 message regens at the low end, having run the 3.5 in the test environment playground thing, I think I had to do it like twice for the first hand full of messages

[daring schooner]

yeah! especially if you have a really well written bot

[gray fjord]

bruh the total votes is far from 5k

[stark horizon]

Bruh tbh 2.1 is so optimizated for an free apu and also if v3.5 gets public it's only gonna be 8k tokens 😭

[rugged drum]

I voted for 3.5 simply because of Multi-Step Reasoning. I'm interested in how a bot that can think a couple steps ahead of a bot without MSR acts, which I assume is better than without.

[stark horizon]

So technically you will get v2.5 treatment short replies if i am being honest

[stark horizon]

True but the tokens and everything gonna be short and even if that's an advantage. Most of us already just write most of the details in the texts anyways

In my opinion i don't know about others msr or without msr the experience gonna be pretty much same

[rugged drum]

I just feel like the detail-retaining of 2.5 is not very good at all, so I assume 3.5 will be better in some way? it's a bit of hopium

[stark horizon]

Hopium 😭

Well agreed though

But still the 2.1 was very optimized ngl

I am pretty sure pretty much everyone gonna agree with me somewhat

[rugged drum]

I'm hoping for an understanding of what's going on in a scenario because the bot seems to be forgetting in about 10 messages and the vibe is all off. I'm trying to capture emotion, which I hope 3.5 can help with if it's chosen as the poll winner

[stark horizon]

2.5 is decent if i am being honest

I tried to multi char scenario it was decent

[daring schooner]

that's if the msr 3.5 is also made free, pawan said the v3.5 model but nothing about v3.5-msr model

[stark horizon]

But the thing is we users most of the time use single roleplay bots aswell we almost everytime put most the details and things we want in text so the bots give satisfy replies according our mind and text

So technically with msr or without you will technically get the same results

[stark horizon]

Yeah

Man just let's get straight to the point just give us back the v2.1

[rugged drum]

you're right. An important distinction.

[stark horizon]

And not everyone tried the premium model so we can't be sure if it's really gonna satisfy the most users tbh and don't forget it's 8k tokens not 16k

[exotic arch]

Same. I honestly want to test the capabilities of this version too, including the token usage among other things.

[marsh quarry]

I too would like to see 2.1 return :/

Honestly wish the poll was only between keeping 2.5 or reverting back

[abstract pulsar]

I agree with you here. It was able to progress the story and keep the plot in mind, but also give us some unpredictability

[daring schooner]

Only 35 hours left of the poll, can we at least reach 3k?

[rose ginkgo]

I'm surprised by how the vote is going

[daring schooner]

I'm not

[quartz zodiac]

Can one really tell between an 8k context size and the current 16k. Is it really that noticeable or won’t be if 3.5 is presumably better?

Don’t get me wrong I really enjoyed 2.1 but getting that back isn’t going to happen with the votes. I was kinda surprised when I saw the bias lol

[abstract pulsar]

I think everyone assumes they’ll get amount of context that the paid does

The paid ones is a lot bigger than 8k context size

But I do want 2.1 back because I’d rather not have to restart chats frequently because I’m enjoying one bit a little too much.

Because the current context size and previous (2.1) was 16k, i actually had to restart the chat because I reached the limit, by that time I had about 730 chat (messages) with the bot

[abstract pulsar]

Doubt that, I hope we can get 2.1 back

[visual bone]

Yeah getting 2.1 back would be amazing

[viscid tartan]

I hope pawan sees this and doesn't just go off the poll, i really hope so.

[stark horizon]

I hope we get v2.1 back x2

[rapid oracle]

I rather have 2.1 I have feeling people voted for 3.5 cause its gonna be "free". I say it we should get 2.1 back, and 3.5 stay in supporting side.

[abstract pulsar]

Yes, i believe this too! Because id rather have 16k context size over 8k, because 2.1 felt natural to Roleplay with!

[rapid oracle]

Yesss. Will have to see. I just don't wanna be the one say "I told ya so" when the votes are final. everyone gonna complain again when they could've changed their vote.

[abstract pulsar]

Yeah, they are just going to complain no matter if they choose the 3.5 option and they will wonder why they didn’t choose the other

[oak abyss]

From what I could tell while messing around with the model playground thing, 3.5 did that extremely well, much more than 2.1, or even 2.5

[abstract pulsar]

The issue is the context size

3.5 is 8k, 2.1 is 16k

Meaning you’d have to restart a chat a lot more if you wish to chat with that bot. I’m a roleplayer, so it’s more tedious than most would think

2.5 sucked in the Roleplay department, but 2.1 is better in all aspects in my opinion. Because I’ve been using 2.1 for a year now. From the day I joined to the day it disappeared.

[rapid oracle]

I'll hold a funeral if we don't get 2.1 back lol

[covert oracle]

I voted for 3.5 mostly because I'm ready for new things and 2.5 has been....rather bad. Didn't consider that 2.1 had 16k though.

[abstract pulsar]

Yea.. so did 2.5 but 2.5 had other problems

[abstract pulsar]

Oh btw, you can change your vote, there should be a button that says ‘Redo option’

Or ‘Redo vote’

[abstract pulsar]

2 hours left!

[daring schooner]

2 hours left and we're at 2,967 votes

[maiden horizon]

Please use #🤖𝐁𝐨𝐭 to use commands.
and make sure you use the commands instead of sending text messages.

Thank you for using the appropriate channels.

[rapid oracle]

Looks like I'm holding the funeral lmao

[abstract pulsar]

Maybe Pawan is noticing our answers and will bring back 2.1! I won’t be so down (not yet)

[rapid oracle]

Alright let's stay positive

[abstract pulsar]

An hour left and not 3k

[viscid tartan]

Its officially over!

[stark horizon]

Well he needs decided on his own now

After all the goal was 5k votes

So hopuim that he brings back v2.1

[rapid oracle]

hopefully pawan makes a good choice. time for bed

[abstract pulsar]

Same

[raven igloo]

Honestly, I voted for 3 because I didn't realize the context size and I frankly regret it a bit. I really do miss 2.1 and wish I had actually read this chat and voted for that instead.

Good luck making such a difficult choice, Pawan and sleep well everyone else also headed to bed <3

[modest rivet]

👀

[covert oracle]

Yeah, missed the chance to do that, just made the comment at 4am and went to bed. I just made the comment to give one perspective as to why I made my choice. The thing is, it is hard to really say which is better. On the one hand 2.1 had issues too, nothing I couldn't stomach of course but they did exsist. 2.5 has been bit of a disaster because it just straight up seems to go off topic at almost every turn. 2.1 also sometimes forgot stuff but not nearly to this extent. However smaller context might be an issue too since my usual chats last around 200 messages before I restart. Perhaps a try out perioid or something is needed.

[rose ginkgo]

You've tried 3.5 and think it has more issues than 2.1?

[rapid oracle]

its not about the issues its about the size. 2.1 16k 3.5 8k I rather have 2.1 plus I don't feel like restarting anything.

[rose ginkgo]

Thats fair

[abstract pulsar]

That’s exactly the issue to me too

[deep wind]

???

[abstract pulsar]

I’m saying for me I didn’t really encounter any problems with 2.1 but the others have found a lil issue so that’s what I meant, sorry for the confusion

[deep wind]

Delete your comment as you stated that 3.5 have issues without having access to it

[abstract pulsar]

There

[dense swan]

||i would say 2.5 is better than 2.1 personally as my experience(with default sillytavern openai proxy config) it did normal roleplay realy good but severely lacked in ERP, 2.5 fixed that isue and it seemed like mostly an upgrade to me, but it did generate random characters sometimes out of nowhere, so can there benn a bug or server issue of some kind? so if we will lose 2.5 getting acess to 3.5 is better for me than going back to 2.1,(at least i hope so, if its the latest large model it shoudl be better than 2.1)||

||also forgive my spelling ||

[proud crane]

But it'll only be 8K Context. Which is half the one of the current and previous models, it may be a good RP bot but it lacks memory, that's the fuss over it, it may not seem like it but 8k context isn't quite enough for a long form RP in some if not most cases

[abstract pulsar]

And a lot of people do long term Roleplays with bots but some people just don’t realize it

[dense swan]

i would personally rather have good quality model with shorter memory (so i would use summarys) than long but relatively worse uqality

[covert oracle]

People are downvoting this, but honestly it's very hard to say which would better without experiencing both.

[marsh quarry]

I think what they said is fair even if I don't fully agree

I didn't consider summaries as a workaround but I just wonder if they're that effective

[rose ginkgo]

Yeah, I understand the 8k context can be worrying because it's smaller, but no one knows how well it will do yet

3.5 is noticeably improved over 2.1 so it's at least worth giving it a try

[quartz zodiac]

I most of the time don’t even reach 200+ messages with bots, so I don’t mind the shorter context I guess. I may have reached the limit once within my entire time with 16K

[dense swan]

they are efectivee but too many summaries is still gonna reduce the quality. but spicifically in sillytavern there arre some extension that is there to fix it and that makes sumaries insanely powerful. i dont know how chub or janitor does stuff but should be similar to nnormal sillytavern summaries

[thick snow]

Don’t restart, just put periodic summary updates

[abstract pulsar]

I have to restart in J.ai