Tutorial CHATGPT Alternative (mixtral-8x7b) and other LLMs

[reonjy]

Good alternative for GPT-3.5 model..
Free
Unfiltered
Good for roleplay

Models:

• mixtral-8x7b (default)
• nous-mixtral-8x7b
• mistral-7b
• openchat-3.5
• gemma-7b
• gpt-3.5-turbo (new)

Credits:
You do not have permission to view the full content of this post. Log in or register now.
Feedback is highly appreciated.

You do not have permission to view the full content of this post. Log in or register now.

 

[YMHAR]

Salamaaat Papsi, bookmarked.

 

[Park Jaechan]

Thank you. Bookmarked na ts

 

[reonjy]

Here to add/customize your own custom instructions!

 

[Techie BeLLe]

Thanks for sharing TS

 

[reonjy]

Salamaaat Papsi, bookmarked.

Thank you. Bookmarked na ts

Thanks for sharing TS

welcome mga paps, tnx sa feedback..

 

[alist1986]

He he. Nabuhay ka ulit kapusa!
Ayos yan at tinuloy mo ulit si Librechat.

Doon sa mga bagong gagamit, kung ano lang yung makita ninyo sa first post ni Reonjy, yan lang yung gumagana.
(No plugins, Bing and ChatGPT). Sa tingin ko, WIP pa ito kaya abang na lang kayo.
It only function as normal chatbots.

 

[reonjy]

He he. Nabuhay ka ulit kapusa!
Ayos yan at tinuloy mo ulit si Librechat.

Doon sa mga bagong gagamit, kung ano lang yung makita ninyo sa first post ni Reonjy, yan lang yung gumagana.
(No plugins, Bing and ChatGPT). Sa tingin ko, WIP pa ito kaya abang na lang kayo.
It only function as normal chatbots.

musta na lods? long time.. yup, yun lang muna ang gumagana..

 

[alist1986]

musta na lods? long time.. yup, yun lang muna ang gumagana..

Ganyan talaga pag umpisa. Parang alam ko na yung balak mo pagtingin lang sa setup ng UI he he. May balak kang i-support yung plugins sa open-LLMs? Challenge yan.

Pahinga muna ako sa threads bukod sa malapit na yang Lenten season. Mga 2 weeks para masulit.
Wala naman mahilig mag-setup ng open-source LLM/GPT UIs, atbp. for personal use dito. Puro ganitong ready-made or free links yung mas preferred ng visitors simula pa last year. But I don't mind dahil di madali para sa lahat. Sa mga advanced UI technology ako naglalagi ngayon (embeddings, training LLMs, adding knowledgebases, etc.). Nagsawa na ako sa kaka-test ng mga minor projects na halos pare-parehas ang features sa github. Mas mainam yung mapag-aralan yung inner workings ng LLM para maintindihan natin yung processing architecture ng NLP ng mga yan - yung best methods for efficient use. Pansin ko kasi, hindi accurate yung results pag direkta ka lang makipag-usap sa GPT or with documents dahil di naman lahat ay mababasa nyan kung limited pa yung token (input/output) capabilities niya o kahit pa umabot ng 128k. Kung malalaki na yung files at marami, kulang-kulang yung sagot he he. Mas effective pa yung may web search kahit papaano. Using GPT as it is is not enough in all cases. Limited lang yung maibibigay na tulong nyan sa tamad magbasa. magsulat o mag-search manually he he. Yung grammar lang ang nakita kong advantage at pag-condense ng collected contents.

Laganap na rin kasi yung mga NO-CODE-AI playgrounds na karamihan ay free or freemium. Yan yung platform na mas attractive sa common users. Marami dyan ay GTM (Go-To-Market) projects, bago gradually maging commercial. Kaya mas OK kung mauna ka (habang free pa) he he.

Yung mga past threads mo naman yung nag-umpisa ng interest ko sa LLMs at GPTs, atbp.. Isang "keyword" mo lang, sapat na sa'kin. Kaya self-sufficient na ako ngayon, thanks to you. Doble-kayod nga lang bago ko makuha yung process he he. Natuto pa akong mag-code ng kaunti sa python to make any AI GUI I want with a reliable GPT tutor. Hindi kasi nakukulitan sa'kin ang AI sa dami ng tanong ko at hindi pwedeng di siya sasagot! Nakuha sa tiyaga.

 

[reonjy]

Ganyan talaga pag umpisa. Parang alam ko na yung balak mo pagtingin lang sa setup ng UI he he. May balak kang i-support yung plugins sa open-LLMs? Challenge yan.

Pahinga muna ako sa threads bukod sa malapit na yang Lenten season. Mga 2 weeks para masulit.
Wala naman mahilig mag-setup ng open-source LLM/GPT UIs, atbp. for personal use dito. Puro ganitong ready-made or free links yung mas preferred ng visitors simula pa last year. But I don't mind dahil di madali para sa lahat. Sa mga advanced UI technology ako naglalagi ngayon (embeddings, training LLMs, adding knowledgebases, etc.). Nagsawa na ako sa kaka-test ng mga minor projects na halos pare-parehas ang features sa github. Mas mainam yung mapag-aralan yung inner workings ng LLM para maintindihan natin yung processing architecture ng NLP ng mga yan - yung best methods for efficient use. Pansin ko kasi, hindi accurate yung results pag direkta ka lang makipag-usap sa GPT or with documents dahil di naman lahat ay mababasa nyan kung limited pa yung token (input/output) capabilities niya o kahit pa umabot ng 128k. Kung malalaki na yung files at marami, kulang-kulang yung sagot he he. Mas effective pa yung may web search kahit papaano. Using GPT as it is is not enough in all cases. Limited lang yung maibibigay na tulong nyan sa tamad magbasa. magsulat o mag-search manually he he. Yung grammar lang ang nakita kong advantage at pag-condense ng collected contents.

Laganap na rin kasi yung mga NO-CODE-AI playgrounds na karamihan ay free or freemium. Yan yung platform na mas attractive sa common users. Marami dyan ay GTM (Go-To-Market) projects, bago gradually maging commercial. Kaya mas OK kung mauna ka (habang free pa) he he.

Yung mga past threads mo naman yung nag-umpisa ng interest ko sa LLMs at GPTs, atbp.. Isang "keyword" mo lang, sapat na sa'kin. Kaya self-sufficient na ako ngayon, thanks to you. Doble-kayod nga lang bago ko makuha yung process he he. Natuto pa akong mag-code ng kaunti sa python to make any AI GUI I want with a reliable GPT tutor. Hindi kasi nakukulitan sa'kin ang AI sa dami ng tanong ko at hindi pwedeng di siya sasagot! Nakuha sa tiyaga.

agree lods, kadalasan dito eh, gusto yung ready-made na.. Para sa'kin, mas ok talaga yung kaw mismo may direct control sa llm mo, at if knowledgeable ka, mas bet ko yung kaw mismo magtrain ng llm mo ng pre-set knowledge or expertise na gusto mo xang may "lamang".

anyway, thanks din kasi nagiging napaka informative ng thread pag nandiyan ka pakner!

 

[alist1986]

agree lods, kadalasan dito eh, gusto yung ready-made na.. Para sa'kin, mas ok talaga yung kaw mismo may direct control sa llm mo, at if knowledgeable ka, mas bet ko yung kaw mismo magtrain ng llm mo ng pre-set knowledge or expertise na gusto mo xang may "lamang".

anyway, thanks din kasi nagiging napaka informative ng thread pag nandiyan ka pakner!

He he, nasanay kasi akong nagtuturo ng professionals; di ako spoonfeeder. Dapt kasi mag-train ka from scratch then move up the ladder sa iyong learning path. Ang requirement naman is keeping attention to details to understand how things really work up to the results - getting the complete overview as much as possible. Sharing of ideas with others keeps you in shape thru interactions. Natututo ka rin sa ibang tao. Lahat naman ng may lisensya, ganyan yung process, continuous learning...Requirement yan para mag-renew. Kaya this applies to any branch of knowledge as well, di ba?

Pero dyan sa Librechat, maraming leeway kayong mga programmers na manipulahin yang plugins to be used by othe LLMs aside from OpenAI (by default). Binanggit naman yan sa documentation, pero di pa ginagawa sa releases.

Ako naman ay mostly dependent sa readymade codes, kaya sa forks ako nagbaka-sakali para makuha yung logic to modify some parts na similar sa gusto kong mangyari.

Sa parada ng Perplexity Labs bots mo, itong You do not have permission to view the full content of this post. Log in or register now. ang gamit ko or just the You do not have permission to view the full content of this post. Log in or register now..

Or yung You do not have permission to view the full content of this post. Log in or register now. ng author basta may login access ka lang sa lahat ng providers nya (kahit free). May direct installers din yan kung binasa ng reader.

Ang simple cheatmode naman sa Librechat is either run a working app in huggingface using docker or clone/duplicate/run them there. Mabagal yung free CPU usage, pero good for personal use. Bahala na lang sila sa mga api requirements by searching the link sources for the apis. Normally, 1 or 2 lang sapat na. No need for permission from authors dahil normal yung app sharing sa HF. Yung project ni You do not have permission to view the full content of this post. Log in or register now. yung best para sa akin. Malikot yan tulad mo he he.

Sa beginners, this Librechat installer is the easiest to use for windows from the same author. You just need to get the requirements for the .env file: You do not have permission to view the full content of this post. Log in or register now. . Pag nasanay na kayo, yon namang You do not have permission to view the full content of this post. Log in or register now.. Yung ginagawa ng Reonjy is his own version, so use your imagination how you can duplicate his work by reading the You do not have permission to view the full content of this post. Log in or register now..

Pakner, di ko pala nabanggit, hanapin mo itong You do not have permission to view the full content of this post. Log in or register now. sa github and use it as your OpenAI API + add the endpoint. Ok yan, gagana yung plugins sa Librechat malamang since OpenAI protocol yung makikita niya. Sa mga GUI ko lang at browser plugins na-test ito dahil key lang yung requirement sa Librechat GUI.. Hangga't buhay yung proxy server OKs na yan. Pag OK sa server mo, just advice the users to use their free Gemini Pro api keys. May limitations lang siyempre. 'Hope this helps.

 

[New Username]

Maraming salamat po sir , pa try ako

 

[alist1986]

Pakner, kung sakaling mag-test ka ng gemini-openai-proxy, test mo na lang kung ok yung proxied endpoint at di nasusunod yung nakalagay sa curl command sa github channel. Ayaw sa You do not have permission to view the full content of this post. Log in or register now., pero dati naman ay ok. Ang working endpoint sa Chatbox UI is You do not have permission to view the full content of this post. Log in or register now..

{"message":"Welcome to the OpenAI API! Documentation is available at https://platform.openai.com/docs/api-reference"}

Ang recommended is go1.21.1 at docker 3 pero nag-edit na lang ako na go1.22.1 at docker 4 ang gamitin. Sa docker siya lang mapapatakbo, pero naiba yung working proxied endpoint ngayon he he.

Yung default docker commands lang, ok na siyang gamitin.

Mapili siya ng GUI para kumagat yung proxied endpoint. Pero pwedeng gamitin yung "gpt-3.5-turbo" at "gpt-4-vision-preview" as stated.
Doon sa interisado, this is just a work-around to fool a UI like Librechat that it is using OpenAI to integrate with ChatGPT plugins.

 

[KrazierThanEver]

thanks idle

 

[alist1986]

...Pasensya pakner sa last posts ko. Di ko namalayan, marami na palang ai proxies na ganyan para sa lahat (free and ρáíd) calling LLM APIs using the OpenAI format. Gemini-openai-proxy is just one of many available. Para yang PAI model, na pwedeng i-select yung GPT3.5-turbo sa UI, pero you're actually using the PAI LLM model. Kakalat na yung mga fake OpenAIs if somebody wants to, he he.

Anyway, back to Librechat mo;

Here to add/customize your own custom instructions!

View attachment 2909391

Doon sa di familiar, better read this : You do not have permission to view the full content of this post. Log in or register now.

The "presets" feature in our app is a powerful tool that allows users to save and load predefined settings for their conversations. Users can import and export these presets as JSON files, set a default preset.....

Pwede kayong mag-ipon ng presets by saving and importing later. yung nasa kanan ng arror yung import button. Para yang LLMs na may iba't ibang personality or task-defined feature, similar to using system prompts as AI assistant, academic, storywriter, etc. Pero dito pati LLM parameters ay kasama sa presets.
Basahin nyo rinYou do not have permission to view the full content of this post. Log in or register now.para malaman ninyo ang ibig sabihin ng mga paramters na yan - temperature, top_p, etc.
You do not have permission to view the full content of this post. Log in or register now. naman yung site para makapili o makaisip ng inyong gagamiting system prompt or custom instructions sa presets ninyo.
(Click nyo lang yung may pulang txt for the links.)
Regardless kung sa Perplexity labs yan galing or from other LLM API source, they generally use 16k contextual length, and their ELO (performance rating) levels is at par or above OpenAI's GPT3.5-turbo model (Luma man o bago) - good as the old GPT-4s. Reliable sila since tuned ang mga yan sa OpenAI datasets din. Since walang special tags yan sa hulihan, I'll assume they're good for chat completions. completions, coding, etc...Best is search for each model and find out where they are good at in comparison with the rest.
Ex. You do not have permission to view the full content of this post. Log in or register now., You do not have permission to view the full content of this post. Log in or register now.....Hanapin nyo dyan sa HF yung iba.

Intayin nyo na lang yung ibang added features dyan like attachments, plugins, etc....di pa yan working.

 

[ATMOZF3ARx]

thankssss

 

[reonjy]

...Pasensya pakner sa last posts ko. Di ko namalayan, marami na palang ai proxies na ganyan para sa lahat (free and ρáíd) calling LLM APIs using the OpenAI format. Gemini-openai-proxy is just one of many available. Para yang PAI model, na pwedeng i-select yung GPT3.5-turbo sa UI, pero you're actually using the PAI LLM model. Kakalat na yung mga fake OpenAIs if somebody wants to, he he.

Anyway, back to Librechat mo;

Doon sa di familiar, better read this : You do not have permission to view the full content of this post. Log in or register now.

Pwede kayong mag-ipon ng presets by saving and importing later. yung nasa kanan ng arror yung import button. Para yang LLMs na may iba't ibang personality or task-defined feature, similar to using system prompts as AI assistant, academic, storywriter, etc. Pero dito pati LLM parameters ay kasama sa presets.
Basahin nyo rinYou do not have permission to view the full content of this post. Log in or register now.para malaman ninyo ang ibig sabihin ng mga paramters na yan - temperature, top_p, etc.
You do not have permission to view the full content of this post. Log in or register now. naman yung site para makapili o makaisip ng inyong gagamiting system prompt or custom instructions sa presets ninyo.
(Click nyo lang yung may pulang txt for the links.)
Regardless kung sa Perplexity labs yan galing or from other LLM API source, they generally use 16k contextual length, and their ELO (performance rating) levels is at par or above OpenAI's GPT3.5-turbo model (Luma man o bago) - good as the old GPT-4s. Reliable sila since tuned ang mga yan sa OpenAI datasets din. Since walang special tags yan sa hulihan, I'll assume they're good for chat completions. completions, coding, etc...Best is search for each model and find out where they are good at in comparison with the rest.
Ex. You do not have permission to view the full content of this post. Log in or register now., You do not have permission to view the full content of this post. Log in or register now.....Hanapin nyo dyan sa HF yung iba.

Intayin nyo na lang yung ibang added features dyan like attachments, plugins, etc....di pa yan working.

sorry, late repz, medyo nabusy, uu meron na other LLMs na merong "function calling" support. Etong mga models na gamit ko, this is from HF.

I will update once supported na yung plugins using these models, tsaka yung file upload kasi napaka useful.

Tsaka sa mahilig mag roleplay, pagsawaan niyo kasi no limits, hanap lang kayo ng prompts..

 

[alist1986]

sorry, late repz, medyo nabusy, uu meron na other LLMs na merong "function calling" support. Etong mga models na gamit ko, this is from HF.

I will update once supported na yung plugins using these models, tsaka yung file upload kasi napaka useful.

Tsaka sa mahilig mag roleplay, pagsawaan niyo kasi no limits, hanap lang kayo ng prompts..

Ako nga sa unang tingin, akala ko Perplexity Labs pero maraming nag-provide din ng APIs nyan since open-source sila. Pero sa comparison, mas tuned ang PL models lalo na yung "instruct" versions nila compared sa ibang base models HF.
Outside of Lbrechat, pwedeng magamit yung Chatgpt plugin using openai's chatgpt-retrieval plugin or langchain. gptindex, atbp. Ginagawa siyang web service api na may openai specs, plugin manifest at https endpoint. But most LLMs are not trained for openai specs kaya even if they work, their performance will be worse. Yan yung limitation sa ngayon sa nabasa ko. Mga ilang buwan pa siguro, makakahabol na rin yung LLMs creating their own plugin index for their models. Ang nakita ko lang plugin na stable sa kanila is web search and chatting with documents but not using chatgpt plugin but a custom one like langchain. Nalalaman ko lang ito sa kakabasa sa mga ai forums nila. Sa mga commercial LLMs baka nakuha na nila yung integration sa plugins with their own development. Sa open source, yung LllamaIndex lang yung alam kong meron.
Nag-eksperimento ako sa LLM+plugin using LLM python module pero sa Librechat ay hindi pa. Ang alam ko sa ngayon is using plugins as a service locally or online, saka ko siya idikit sa LLM manually paisa-isa. Ang pakay ko talaga is preparing a large custom vector database either to tune or to add the knowledgebase sa mapusuan kong LLM in the future - pag na-build ko na yung AI PC ko he he. Kaya sa embedding muna ako nakatuon. LLMs are freely available everywhere. Collect ka lang ng maraming apis, links or accounts, para makarami ka.
At tama ka, sa mga curious newbies na gusto ng roleplay bots, pwede rin ang mga yan basta alam nila yung pag-input ng presets. Best is kopyahin nila yung parameter settings ng mga "roleplay ai" bots online at dyan ipasok. Di ko pa na-verify kung may un-filtered models dyan sa list, but they need to try if they want to he he. Kalimitan naman sa open source LLMs ay may certain degree of tolerance sa roleplay. Basta makuha mo yung correct prompts, it will function well. But forcing them to do îllégâl tasks is not advisable, though some do.
Ang pansin ko rin dito sa Librechat mo, di gumagana yung AIs pag ginamit ko yung "/chat/new". Sa "/c/new/" lang siya working. Sa official demo link, ok naman siya kung mag-switch ako. Pero di man importante ito.
Balikan ko ulit itong Librechat configuration guide para makapag-test ng custom LC config files pag nakarating na ako sa aking next destination he he. Bakasyon muna para makapag-fishing ang hunting, nang maiba naman yung atmosphere he he.
Bisitahin mo pakner yung Coze.com (kung sakaling di mo pa alam). Free and ρrémíùm yung AI playground. Magsasawa yung user sa plugins and create their own bots.

 

[reonjy]

Ako nga sa unang tingin, akala ko Perplexity Labs pero maraming nag-provide din ng APIs nyan since open-source sila. Pero sa comparison, mas tuned ang PL models lalo na yung "instruct" versions nila compared sa ibang base models HF.
Outside of Lbrechat, pwedeng magamit yung Chatgpt plugin using openai's chatgpt-retrieval plugin or langchain. gptindex, atbp. Ginagawa siyang web service api na may openai specs, plugin manifest at https endpoint. But most LLMs are not trained for openai specs kaya even if they work, their performance will be worse. Yan yung limitation sa ngayon sa nabasa ko. Mga ilang buwan pa siguro, makakahabol na rin yung LLMs creating their own plugin index for their models. Ang nakita ko lang plugin na stable sa kanila is web search and chatting with documents but not using chatgpt plugin but a custom one like langchain. Nalalaman ko lang ito sa kakabasa sa mga ai forums nila. Sa mga commercial LLMs baka nakuha na nila yung integration sa plugins with their own development. Sa open source, yung LllamaIndex lang yung alam kong meron.
Nag-eksperimento ako sa LLM+plugin using LLM python module pero sa Librechat ay hindi pa. Ang alam ko sa ngayon is using plugins as a service locally or online, saka ko siya idikit sa LLM manually paisa-isa. Ang pakay ko talaga is preparing a large custom vector database either to tune or to add the knowledgebase sa mapusuan kong LLM in the future - pag na-build ko na yung AI PC ko he he. Kaya sa embedding muna ako nakatuon. LLMs are freely available everywhere. Collect ka lang ng maraming apis, links or accounts, para makarami ka.
At tama ka, sa mga curious newbies na gusto ng roleplay bots, pwede rin ang mga yan basta alam nila yung pag-input ng presets. Best is kopyahin nila yung parameter settings ng mga "roleplay ai" bots online at dyan ipasok. Di ko pa na-verify kung may un-filtered models dyan sa list, but they need to try if they want to he he. Kalimitan naman sa open source LLMs ay may certain degree of tolerance sa roleplay. Basta makuha mo yung correct prompts, it will function well. But forcing them to do îllégâl tasks is not advisable, though some do.
Ang pansin ko rin dito sa Librechat mo, di gumagana yung AIs pag ginamit ko yung "/chat/new". Sa "/c/new/" lang siya working. Sa official demo link, ok naman siya kung mag-switch ako. Pero di man importante ito.
Balikan ko ulit itong Librechat configuration guide para makapag-test ng custom LC config files pag nakarating na ako sa aking next destination he he. Bakasyon muna para makapag-fishing ang hunting, nang maiba naman yung atmosphere he he.
Bisitahin mo pakner yung Coze.com (kung sakaling di mo pa alam). Free and ρrémíùm yung AI playground. Magsasawa yung user sa plugins and create their own bots.

hehe, vacation ka muna pakner! chineck ko yung Coze, ayos din ah.. goods na goods din! tingnan ko ano pwede magawa dito tnx for this.

 

[kilatjr69]

salamat boss