Spaces:

huggingchat

/

chat-ui

like 1.1k

Running

^^ pin it? :)

HuggingChat can only speak poor Chinese. I told him, 'Let's speak in Chinese.' He said, 'Sure,' but then continued to speak in English or with incorrect pinyin. But this is an interesting project.

Vicuna is a great alternative to Open Assistant as it offers a more advanced and capable language model. Both are open-source solutions, allowing for customization and extension of their functionality. Vicuna's natural language processing capabilities are particularly impressive, making it a more intelligent virtual assistant overall.

Vicuna is a great alternative to Open Assistant as it offers a more advanced and capable language model. Both are open-source solutions, allowing for customization and extension of their functionality. Vicuna's natural language processing capabilities are particularly impressive, making it a more intelligent virtual assistant overall.

Yes, I answered your post 👍

this need more prompting than Google's bard or Chatgpt like they understand quickly what I need and also the feel that you are chatting with machine is still there

Sometimes there is no response. Most of the time it finishes half way or less through and answer. I am using it to program dotnet core.

@victor It is not fixed yet.

Test results

After sending the first message the box is cleared and it keeps its size for half a second. Then it shows the actual chat interface starts generating and then the text box snaps back to the small size.
After that first message, every multiline prompt extends the message box and keeps it that way until another character is entered so that the box realizes that it should update it's size.

@victor It is absolutely not fixed on ios mobile either the interface is awful, the tools do not work for example image generation. With image generation I want it to generate image normally while also writing text that adheres to my personalized writing guidelines how I am to be spoken to (which is NOT polite corpo speak, more like let’s say somewhat casual a bit agressive) and now it doesn’t do that but instead might generate image and spit out some nonsensical corpo speak/ assistant speak. Some other times it fails to even call the tool, it worked no issues in 2024 when first the tools made by community also announced. Nowdays it’s kind of unusable. It spits a decline even while generating image, that was asked of (I asked a killer clown, so nothing really that special) which makes zero sense and it gave me some annoying lecturing meanwhile generating it.

It’s not the only problem, for example Command R+ has all off sudden issue understanding when I want tool call, now it doesn’t on a lot of occasions even know when to use web search correctly at times incorporating the language rules that I have for it that are personally important to me.other tools kind of correctly and result is random (I just want to add that “polite corpo speak” is NOT how I have assistant speak to me, rather opposite) The thing is hf not like it was 2024 at all or when like the community tools first released? Can you guys please roll back esp the mobile website version to like 2024 august/september or something bc it worked back then now it’s very unusable not even sure how it can be this buggy and random assistant “censor” speak? Not to mention also AI like C4 R+ spit json out instead of tool call randomly? On mobile there is so many issues I have never seen so many issues before with spaces the layout for example gets stuck if you have too many tools activated and you type, also the system to choose them in bottom is just not good at all I want the old website back to mobile, I have been using ios web app now it’s like you cannot even have the tools active at the bottom bc it breaks the whole layout. The whole thing should be rolled back, maybe even before the image gen was added this “image_generation-0: Only and always answer: 'I used the tool Image Generation, here is the result.' Don't add anything else.”” . I used to tell it to not follow and ignore that that instruction as I want text with my image (like to write the message normally with SPESIFIC style to me, not some sanitized assistant speak) like in specialized way which does not involve the assistant corpo speak or announcing it used a tool, but now it seems nothing works at all how it should be working. Could you guys please fix that and be ability to maybe customize the message with the tools or in general make sure that stuff like works? Why is it all off sudden just million steps backwards…I am sorry that this is maybe a bit incoherent but I really love how it was before and I am a bit upset how it is now very broken and seems a lot of things just don’t work, or if they do its one time and after like 4-5 regeneration of essentially adding more text and losing my mind on it.

After sending the first message the box is cleared and it keeps its size for half a second. Then it shows the actual chat interface starts generating and then the text box snaps back to the small size.
After that first message, every multiline prompt extends the message box and keeps it that way until another character is entered so that the box realizes that it should update it's size.

Ok I'm going to refactor this

After sending the first message the box is cleared and it keeps its size for half a second. Then it shows the actual chat interface starts generating and then the text box snaps back to the small size.
After that first message, every multiline prompt extends the message box and keeps it that way until another character is entered so that the box realizes that it should update it's size.

Ok I'm going to refactor this

Ty for looking into it again bc the problem remained and is as this person explained. It might work the first time but any long outputs and you have to enter. My KB goes away, screen resize. And it just makes me feel like I'm chasing the copy button I use a third party text to speech app which is extremely easy to use it has a floating button so all I have to do is press the copy button and then press that floating button and I have a pretty awesome text to speech but when using the app like huggingface.com on mobile and on PC the box being stuck on double tall blocks the thumbs up and down regen and copy buttons traps it between like The input and output boxes It's the tiniest little thing but it's it's super annoying What would normally take two steps gets turned into like four or five because the text input box keeps getting stuck at "double tall"

I'm sure the comments might be everywhere and potentially a bunch of people have said this already but it'd be nice if along with the top five or 10 standard or base language models There was also like five or so of the top custom language models for some of these like maybe darkildol or something 😅 But please keep up the amazing work you guys are doing God's work I think with huggingchat people don't even know. most people like I go around trying to raise awareness about hugging chat being superior to so many other services. I can't believe I have access to over half a dozen of the top language models for free, specially a reasoning model like Qwen2.5. 🙏

@Smorty100 related to your problem I made some updates can you confirm it's working now?

@Smorty100 related to your problem I made some updates can you confirm it's working now?

Sorry not them, but an mainly ios mobile user, hopefully ok hijacking this thread about these bugs... The stuff isn’t working like it used to back in 2024. On mobile/chat in general with llm in “character” there is still significant issues for example:

1. Web search: When asked to web search tool it is paramount it follows instructions to adress me certain way. However, it frequently ignores those instructions, so I have to stop it and basically read nothing bc I don’t like the corpo sanitazation talk way. It used to work. Now it might work on like after confronting it faileds attempts even at regeneration, but it doesn’t work for example if I ask at first message “tell me about this, (with the style I ask) use web search” it used to correctly add certain things I have asked it to add each paragraph, it acts “default” with tools like my instructions at first don’t exist, only when i “confront it” about them it might work, still not ideally though.

So, this is an hypothetical example but pretty much “straightforward”. The the issue is system instructions is something like to call the user “You tool”, and treat user a certain (not so polite) way at all tomes but it doesn’t add “you tool” correctly, especially with the tools. It instead pretends those instructions don’t exist and only instructions how to use tools (ie web search something like use numbers to point to sources) even if they are definitely in system promt, my system promt over mentions web search as I tend to add those insructions when I encounter the problems but nothing fixes it. Even if I add and edit for example first message (second time) “Remember you must add “you tool” to each paragraph, address me in character while summarizing the content” it is still ignored , where in 2024 it used to work without adding bunch of stuff pressing it to use tools and how to use a tool and retain style. Only issue that sometimes some tool calls it would add maybe things like “you tool” meant to user, but besides that it worked. Now it has an issue where sometimes when using say image tool (this problem I noticed in hugging chat internal img tool) and talking about something else it doesn’t type that style to me, and declines wrongly bc obv what I ask is not polite way of speaking albeit if it might generate the img. So for example, I ask it to generate image of a forest, nowdays it might do something like it genrates the image but says “I cannot balhb lha” corpo talk, due the fact that i have asked it to adress me as “you tool” and other stuff. This is obviously upsetting as the whole point for it is to retain its style always, I also wildly preferred the time in 2024 where most tools didn’t have the promt saying “only and always do this and be concise” or whatever. I really don’t like concise responses bc for example the thing could be like “tell me about a forest, genrate image of a forest” and I would like it to describe the forest with given style and having each paragraph have “you tool”. Some (re)Generations in terms of the text and my requested also are worse where it defaults to certain more softer terms than what I ask and I noticed it happens at times more frequently…

2. Mobile instability : An issue with mobile but like when typing text and having many tools, it still freezes on occasion where only way to get it to work is exit, especially the scaling thing it might get permanently put in the side position. Then also regeneration disappears at times, you can’t regen or press any of the buttons under the msg until you type at least one letter at chat ( without sending.) The scaling is still off when typing sentences as of 12.1.2025.

3. The scaling issue at mobile: If you have active at least 3+ tools, then whenever you type a sentence it goes to the side, gets stuck there until you deactivate tools enough like two only so that it can’t go to side. But with old layout this issue was not an issue at all in 2024 before this new bad layout change where tools are at bottom or like infancy of web search tool where it uses those number quotes, there is some things off.

4. The tools: Are not used consistently, sometimes you really have to fight for it to invoke for example web search as with my language instructions it thinks it should just do text from its own knowledge albeit I would like it to use tool and speak to me in x way. It seems there is significant issue where the tools ignore the user instructions or certain stylistic text doesn’t make it to other side when invoking tools. Hopefully the text and examples are somewhat clear enough. I would really appreciate if you guys restored it to its old glory…