Suggest New Models
sameedzahoor opened this issue · comments
I recently tried chat.lmsys.org. They have a lot of models. They are collecting large chat datasets for research purposes. You can try to integrate it if you can. You can even get in touch with the team as they need more users.
I checked, definitely a nice site, however the requests are sent through websockets which is a completely different protocol, but I will see what I can do.
@aandrew-me Have you heard about Horde AI? They provide quite a lots of APIs, for free. There is also this public end point which works quite nicely: https://koboldai-koboldcpp-tiefighter.hf.space/api # API docs.
https://koboldai-koboldcpp-tiefighter.hf.space # The interface.
I really like the new changes. Thank you for keeping this project up!
@aandrew-me Have you heard about Horde AI? They provide quite a lots of APIs, for free. There is also this public end point which works quite nicely: https://koboldai-koboldcpp-tiefighter.hf.space/api # API docs. https://koboldai-koboldcpp-tiefighter.hf.space # The interface. I really like the new changes. Thank you for keeping this project up!
Great, they even provided docs for their APIs, if everything works fine, I will surely add.
If you find APIs or sites to add, feel free to suggest
I had found a great site - https://codeium.com/live/general
However I couldn't reverse it, so if anyone can, would be great
@aandrew-me Have you heard about Horde AI? They provide quite a lots of APIs, for free. There is also this public end point which works quite nicely: https://koboldai-koboldcpp-tiefighter.hf.space/api # API docs. https://koboldai-koboldcpp-tiefighter.hf.space # The interface. I really like the new changes. Thank you for keeping this project up!
The responses are a bit strange, these look like text excerpts from books.
> go run *.go -provider koboldai "What are morals"
What are morals
? Whence come they, whither tend, what's their use?
Perhaps the brightest star which heaven's broad expanse surveys,Devoid of sense as it appears to mortal ken,
Yet to mankind itself affords delight and wonder.
Such to my mind seem moral truths; 'tis this that leadsAt once astray our souls, or guides them on to good.
Whence then is it these motions
Check out my links to free LLM APIs.
Check out my links to free LLM APIs.
I checked all of them
Yes, kobold is like that, but its what I like about it, it was trained on novels rather than wikipedia articles or the such, Kobold is very much experimental, but its a great addition given that its API is public, its a better alternative to scrapping other sites I think
Yes, kobold is like that, but its what I like about it, it was trained on novels rather than wikipedia articles or the such, Kobold is very much experimental, but its a great addition given that its API is public, its a better alternative to scrapping other sites I think
I see, well its definitely unique. Expect it in the next update
Okay, thank you for all the hard work!
Here I leave the link to horde, a community based project, decentralized AI but with a centralized end point, the API documentation is there too, there are a few python clients based on it too for reference, https://stablehorde.net/
They offer a lot of models, each for a different purpose, to get an API key you only need to supply an OATH account. They also provide Image Generation models, Text Generation models and others
This may not be fully on topic, but in terms of models has anyone had success using GPT-4 in tgpt? I get errors every time I try, but per my API key on OpenAI's website it is being successfully queried and the "last accessed" field updates to present day.
I am trying the following without success:
tgpt --provider openai --key "sk-xxxx" --model "gpt-4"
I either get errors about authentication, or when I get results and ask which model is being used the AI answers with some flavor of GPT 3.
@calibers16 Are you using a free api key or a paid one? Gpt4 will only work with a paid one
A paid API key (I have $5 in my account). I pay for GPT-4 as well.
A paid API key (I have $5 in my account). I pay for GPT-4 as well.
https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4
Note: this will not provide subscribers with access to the API.
A paid API key (I have $5 in my account). I pay for GPT-4 as well.
Those 5 dollars were already present, weren't they?
When I first tried I think I had a dollar the account? I added $5 yesterday and created a new API key for CLI, which I can confirm is getting pinged but I'm not getting GPT-4 responses in tgpt.
What are you getting then?
I'm using platform.openai.com.
but I'm not getting GPT-4 responses in tgpt.
Why do you think so? It just does not know it's name.
Try ask complex questions. E.g.:
There are two horses in a cart. Each horse is running at a speed of 45 km/h. What is the speed of the wagon?
Answer from 3.5:
The speed of the wagon is the combined speed of the two horses. Since each horse is running at a speed of 45 km/h, the total speed of the wagon would be 45 km/h + 45 km/h = 90 km/h.
From 4:
The speed of the wagon is also 45 km/h. The speed of the horses doesn't add up because they're pulling the same wagon. They work together to maintain the speed, not to increase it.
Yes, GPT-4 always says its using GPT-3
It also makes sense to set not just generic gpt-4 (which resolves to gpt-4-0613), but more recent gpt-4-1106-preview, aka GPT-4 Turbo.
BTW, gpt-3.5-turbo-1106 is also newer than gpt-3.5-turbo (which resolves to gpt-3.5-turbo-0613 atm)
Ahhhhhh, you guys are the best. I was assuming from my previous experience with ChatGPT's website that it would correctly identify itself. Thanks for the tip @johnd0e to utilize GPT-4 Turbo!
I don't have much knowledge about this project, but I'm curious if it's possible to run models locally. I came across an interesting article on uncensored local models that discusses the ethics and necessity of having uncensored models. However, I believe it could also be a nice to have the possibility to run local models that are somewhat censored.
Edit: The article also includes 4 models that are uncensored.
I don't have much knowledge about this project, but I'm curious if it's possible to run models locally. I came across an interesting article on uncensored local models that discusses the ethics and necessity of having uncensored models. However, I believe it could also be a nice to have the possibility to run local models that are somewhat censored.
Edit: The article also includes 4 models that are uncensored.
This is what you're looking for - https://gpt4all.io
There are many models available for use
I don't have much knowledge about this project, but I'm curious if it's possible to run models locally. I came across an interesting article on uncensored local models that discusses the ethics and necessity of having uncensored models. However, I believe it could also be a nice to have the possibility to run local models that are somewhat censored.
Edit: The article also includes 4 models that are uncensored.This is what you're looking for - https://gpt4all.io There are many models available for use
What are the main differences between a project like gpt4all and this one? Thank you by the way for sending me that link, it looks quite interesting, I might give it a try.
@Myrdincx This project doesn't use local models
Fair enough, thank you for the quick responses. Ill make sure to check both projects out :)
Hi @aandrew-me, I have seen 2 interesting api to generate images.
Pollinators
Does not require a key but shows watermarks on the images.
https://image.pollinations.ai/prompt/{prompt}
Prodia
Requires key and does not have watermarks.
1000 API calls free
GeminiProChat
https://gemini-chatbot-sigma.vercel.app
Based on Gemini Pro API, It's Free
Why'd we want a useless AI?
what?
what?
Haven't you seen the Gemini drama? Also, Gemini refuses to produce code, so for developing its not very useful. Neither for writing or generating images since its TOO biased.
Haven't you seen the Gemini drama?
i guess no!?
Gemini refuses to produce code
afaik, GeminiPro does. maybe not the best quality but good to have alternatives i guess..
Weird, I wanted it to help me with Regex and it told me it would be "unethical" because of copyright, etc etc etc.
what?
Haven't you seen the Gemini drama? Also, Gemini refuses to produce code, so for developing its not very useful. Neither for writing or generating images since its TOO biased.
I'm not a fan of Gemini, but making code does it. Now for 3 days you can edit and run code snippets directly in the Gemini UI.
Did you think about providing support for Perplexity AI - perplexity.ai as a option ? It's free on the browser, uses ChatGPT 3.5 and might be a backup option but I don't know if it works via terminal.
Did you think about providing support for Perplexity AI - perplexity.ai as a option ? It's free on the browser, uses ChatGPT 3.5 and might be a backup option but I don't know if it works via terminal.
I had tried, but I couldn't. Probably because it uses websockets or because of cloudflare, cant exactly remember
GeminiProChat
https://gemini-chatbot-sigma.vercel.app
Based on Gemini Pro API, It's Free
well, it doesn't use gemini-pro. very likely gpt-3.5
well, it doesn't use gemini-pro. very likely gpt-3.5
What makes you think so?
well, it doesn't use gemini-pro. very likely gpt-3.5
What makes you think so?
hmmm... looks like google made gemini-pro API free for everyone (upto 60q/min & no privacy). we can make direct api call.
curl -X POST "https://generativelanguage.googleapis.com/v1beta/models/gemini-1.0-pro:generateContent?key=${API_KEY}" -H 'Content-Type: application/json' -d '{"contents":[{"parts":[{"text":"what is the meaning of life?\n\n\n"}]}],"generationConfig":{"temperature":0.9,"topK":1,"topP":1,"maxOutputTokens":2048,"stopSequences":[]},"safetySettings":[{"category":"HARM_CATEGORY_HARASSMENT","threshold":"BLOCK_ONLY_HIGH"},{"category":"HARM_CATEGORY_HATE_SPEECH","threshold":"BLOCK_ONLY_HIGH"},{"category":"HARM_CATEGORY_SEXUALLY_EXPLICIT","threshold":"BLOCK_ONLY_HIGH"},{"category":"HARM_CATEGORY_DANGEROUS_CONTENT","threshold":"BLOCK_ONLY_HIGH"}]}'
@aandrew-me Sorry to bother, but could you add HuggingChat? It seems simple, and it works excellently, they provide lots of models and users can even make custom assitants, it'd be great if you provided support for it, since it can even do math, image generation and etc using the c4ai model. Also, support for adding "personalities"/"assistants" like these? https://hf.co/chat/assistant/65bead914b6dff9202988a88
I found this API implementation you could use as reference: https://github.com/Soulter/hugging-chat-api
@aandrew-me Sorry to bother, but could you add HuggingChat? It seems simple, and it works excellently, they provide lots of models and users can even make custom assitants, it'd be great if you provided support for it, since it can even do math, image generation and etc using the c4ai model. Also, support for adding "personalities"/"assistants" like these? https://hf.co/chat/assistant/65bead914b6dff9202988a88
I found this API implementation you could use as reference: https://github.com/Soulter/hugging-chat-api
No need to apologize, the whole purpose - of this issue is to propose new providers. I will try to add
Here's a reversed API for HugginChat, sadly, its JS. But you can implement this in Go: https://github.com/erucix/Huggingface-API
@xplshn I don't like the process of needing cookies. How long do they last? Also the site keeps throwing errors on random occasions
That's weird, I use huggingchat regularly and it doesn't give me errors. You can also use HuggingChat without logging in, so I guess that could be exploited too? Their default model changes constantly (for the better), for now it is c4a, which is pretty good, its the first model to rival GPT-4o
There are other APIs in GO that use the Email and Password too, if you dislike using cookies. And there's this library too for using the inference endpoints:
https://github.com/hupe1980/go-huggingface
https://github.com/Kardbord/hfapigo
I found that duckduckgo launched their "Chat" - there are gpt3, claude, llama and mixtral: