Using free AI power from our Harbour apps !!!

[Antonio Linares]

Dear friends,

This is the fastest free AI available at the moment (GPT4All):

https://wormhole.app/p4LxB#Do1u8TMntAycXyxfg2X5vA

I do appreciate if you download it and check if the EXE is running for you.

Its over 3 gigs. Be patient, it will take time :-)

You can also download the same included files from github directly:

https://github.com/nomic-ai/gpt4all/blob/main/chat/gpt4all-lora-quantized-win64.exe

https://the-eye.eu/public/AI/models/nomic-ai/gpt4all/gpt4all-lora-quantized.bin

I wanted to share mines just to asure that we all test the same EXE :-)

This morning meanwhile I was testing and helping on some python code of GPT4All dev team I realized (I saw and debugged the code) that they just were creating a process with the EXE and routing stdin and stdout, so I thought it is a perfect ocassion to use the geat processes functions developed by Prezmek! 

So here you have a very early gpt4all.prg to do it. IT DOES NOT WORK YET but I guess

you may want to participate in this amazing adventure.

go.bat

@setlocal
@call "%ProgramFiles%\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvarsall.bat" amd64
c:\harbour\bin\win\msvc64\hbmk2 gpt4all.prg -comp=msvc64
gpt4all.exe
@endlocal

gpt4all.prg (we only reach to show the prompt ">" routed from GPT4All EXE)

function Main()

   local hProcess, hStdIn, hStdOut, hStdErr, cBuffer, nSize := 1024, cMsg := ""

   hProcess = hb_processOpen( "gpt4all-lora-quantized-win64.exe", @hStdIn, @hStdOut, @hStdErr, .t. )
   ? "GPT4All running..."

   cBuffer = Space( 1024 )
   while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
      InKey( 0.1 )
   end
   ? nReaded, AllTrim( cBuffer ), "Hi my friend, how are you ?"

   FWrite( hStdIn, "Hi my friend, how are you ?" + Chr( 13 ) )  

   while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
      InKey( 0.1 )
   end
   ? nReaded, AllTrim( cBuffer )

   while .t.
      cBuffer = Space( nSize )
      nReaded = hb_PRead( hStdOut, @cBuffer, nSize, 500 )

      if nReaded > 0
         cMsg += Left( cBuffer, nReaded )
      endif

      if nReaded < 0  
         exit  
      endif  
   end

   ? cMsg

   hb_processClose( hProcess, .t. )
   FClose( hStdIn )
   FClose( hStdOut )
   FClose( hStdErr )
   
return nil

You may take the data of your DBFs, SQL values, external PDFs,

your email account, github repos, etc... and connect that data with your local

and free GPT4All EXE and make any query you or your user may need. This concept

is named "langChain" as it is the foundation of the chatGPT plugins.

Think "langChain" as "senses" for the "artificial brain" :-) This is being developed now

for GPT4All too. So we may have all that power quite soon. In the meantime we should

be able to send large queries to GPT4All. If you want to submit another line, end your input in '\' (explained by GPT4All)

I assume you may want to participate in this amazing Harbour new adventure and I do appreciate your time and cooperation.

best regards

[Antonio Linares]

Enhanced gpt4all.prg version (still buggy)

function Main()

   local hProcess, hStdIn, hStdOut, hStdErr, cBuffer, nSize := 1024

   hProcess = hb_processOpen( "gpt4all-lora-quantized-win64.exe", @hStdIn, @hStdOut, @hStdErr, .T. )

   ? "GPT4All running..."

   cBuffer = Space( 1024 )
   while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
      InKey( 0.1 )
   end

   ? RemoveANSIEscapeCodes( AllTrim( cBuffer ) ), "Hi my friend, how are you ?"
   ?

   FWrite( hStdIn, "Hi my friend, how are you ?" + hb_Eol() )

   while ! Empty( cBuffer )  

      while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
         InKey( 0.1 )
      end

      ?? RemoveANSIEscapeCodes( AllTrim( cBuffer ) )      
   end  

   hb_processClose( hProcess, .t. )
   FClose( hStdIn )
   FClose( hStdOut )
   FClose( hStdErr )

   InKey( 0 )

return nil

function RemoveANSIEscapeCodes( cString )
    local cCleanString := ""
    local lInEscapeSeq := .F.
    local cEscapeChars := Chr( 27 ) + "["
    local cEndChars := "mK", n

    for n := 1 to Len( cString )
        cChar := Substr( cString, n, 1 )
        if lInEscapeSeq
            if At( cChar, cEndChars ) > 0 .or. ( cChar == "0" .and. Right( cCleanString, 1 ) == Chr( 27 ) )
                lInEscapeSeq := .F.
            endif
        elseif At( cChar, cEscapeChars ) > 0
            lInEscapeSeq := .T.
        else
            cCleanString += cChar
        endif
    next

return cCleanString

best regards

[Antonio Linares]

Enhanced version (still buggy)

function Main()

   local hProcess, hStdIn, hStdOut, hStdErr, cBuffer, nSize := 1024

   hProcess = hb_processOpen( "gpt4all-lora-quantized-win64.exe", @hStdIn, @hStdOut, @hStdErr, .T. )
   ? "GPT4All running..."

   cBuffer = Space( 1024 )
   while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
      InKey( 0.1 )
   end

   ?? RemoveANSIEscapeCodes( AllTrim( cBuffer ) ), "Hi my friend, how are you ?"

   ?

   FWrite( hStdIn, "Hi my friend, how are you ?" + hb_Eol() )

   while ! Empty( cBuffer )  
      while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
         InKey( 0.1 )
      end

      ?? RemoveANSIEscapeCodes( AllTrim( cBuffer ) )

      Alert( RemoveANSIEscapeCodes( AllTrim( cBuffer ) ) )

   end  

   hb_processClose( hProcess, .t. )
   FClose( hStdIn )
   FClose( hStdOut )
   FClose( hStdErr )

   InKey( 0 )

return nil

function RemoveANSIEscapeCodes( cString )

   local cCleanString := "", n

   for n := 1 TO Len( cString )
      cChar := SubStr( cString, n, 1 )
      if cChar == Chr( 27 )
         while cChar != "m"
            n++
            cChar = SubStr( cString, n, 1 )
         end      
         cChar = ""
      endif  

      cCleanString := cCleanString + cChar
   next

return cCleanString

[Antonio Linares]

I just read this in the discord chat for GPT4All:

gpt4all is much better than llama 7b (facebook data) because it's been fine tuned. That's the whole point

Regardless of all this, be aware that the llama models are released under a non-commercial license, so you are not meant to ever use them (or any models derived from them) in any commercial production setting

best regards

[Antonio Linares]

random sample from the GPT-J model (that you can use commercially!) we will be releasing tonight tomorrow morning

just readed in discord chat

[Antonio Linares]

Working version!

I don't understand why it can't close/kill the process. Any hints are apreciated.

GPT4All running...

> Hi my friend, how are you ?

I ' m doing great thanks for asking ! How about yourself ?
>

gpt4all.prg

function Main()

   local hProcess, hStdIn, hStdOut, hStdErr, cBuffer, nSize := 1024

   hProcess = hb_processOpen( "gpt4all-lora-quantized-win64.exe", @hStdIn, @hStdOut, @hStdErr, .F. )
   ? "GPT4All running..."

   cBuffer = Space( nSize )

   while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0
      InKey( 0.1 )
   end

   ?? RemoveANSIEscapeCodes( AllTrim( cBuffer ) ), "Hi my friend, how are you ?"
   ?

   FWrite( hStdIn, "Hi my friend, how are you ?" + hb_Eol() )

   while ! Empty( cBuffer )  

      cBuffer = Space( nSize )

      while ( nReaded := hb_PRead( hStdOut, @cBuffer, nSize, 500 ) ) == 0

      end

      ?? RemoveANSIEscapeCodes( AllTrim( cBuffer ) )

      ?? " "
   end  

   if ! hb_processClose( hProcess, .T. )
      ? "can't close GPT4All process..."
   end

   FClose( hStdIn )
   FClose( hStdOut )
   FClose( hStdErr )

return nil

function RemoveANSIEscapeCodes( cString )

   local cCleanString := "", n

   for n := 1 TO Len( cString )
      cChar := SubStr( cString, n, 1 )
      if cChar == Chr( 27 )
         while cChar != "m"
            n++
            cChar = SubStr( cString, n, 1 )
         end      
         cChar = ""
      endif  

      cCleanString := cCleanString + cChar
   next

return cCleanString

[Antonio Linares]

Enhanced version:

https://fivetechsupport.com/forums/viewtopic.php?p=258426#p258426

[Antonio Linares]

Converting it into a Class. Not working yet (previous version works fine)

#include "hbclass.ch"
#include "FileIO.ch"

function Main()

   local dummy := QOut( "GPT4All running..." )
   local oAI := GPT4All():New()

   oAI:Read()
   oAI:Write( "Which is the largest city in the world ?" )
   // oAI:Read()

   oAI:End()

return nil  

#define BUFFER_SIZE  1024
#define LOGFILE_NAME "GPT4All.log"

CLASS GPT4All

   DATA   hProcess
   DATA   hStdIn, hStdOut, hStdErr

   METHOD New()
   METHOD Write( cText ) INLINE ::Log( "write: " + cText ), FWrite( ::hStdIn, cText + hb_eol() )
   METHOD Read()
   METHOD End()
   METHOD Log( cMsg )

ENDCLASS  

METHOD New( cPrompt ) CLASS GPT4All

   local hStdIn, hStdOut

   ::hProcess = hb_processOpen( "gpt4all-lora-quantized-win64.exe", @hStdIn, @hStdOut, @::hStdErr )

   ::hStdIn = hStdIn
   ::hStdOut = hStdOut

   FErase( LOGFILE_NAME )
   ::Log( "process: " + AllTrim( Str( ::hProcess ) ) )
   ::Log( "stdin: " + AllTrim( Str( ::hStdIn ) ) )
   ::Log( "stdout: " + AllTrim( Str( ::hStdOut ) ) )

return Self

METHOD Read() CLASS GPT4All

   local nReaded, cBuffer, cResult

   cBuffer = "x"

   while ! Empty( cBuffer )  

      cBuffer = Space( BUFFER_SIZE )
      nReaded = 0
      while ( nReaded += hb_PRead( ::hStdOut, @cBuffer, BUFFER_SIZE, 1000 ) ) == 0
      end

      ?? RemoveANSIEscapeCodes( SubStr( cBuffer, 1, nReaded ) )
   end

   ::Log( "after read" )

return

METHOD End() CLASS GPT4All

   local nResult := hb_processClose( ::hProcess )

   FClose( ::hStdIn )
   FClose( ::hStdOut )
   FClose( ::hStdErr )

   ::Log( "End" )

return nResult  

METHOD Log( cMsg ) CLASS GPT4All

   local hLogFile

   if ! File( LOGFILE_NAME )
      FClose( FCreate( LOGFILE_NAME ) )
   endif      

   if( ( hLogFile := FOpen( LOGFILE_NAME, FO_WRITE ) ) != -1 )
      FSeek( hLogFile, 0, FS_END )
      FWrite( hLogFile, cMsg + hb_eol(), Len( cMsg ) + Len( hb_eol() ) )
      FClose( hLogFile )
   endif

return nil  

function RemoveANSIEscapeCodes( cString )

   local cCleanString := "", n

   for n := 1 TO Len( cString )
      cChar := SubStr( cString, n, 1 )
      if cChar == Chr( 27 )
         while cChar != "m"
            n++
            cChar = SubStr( cString, n, 1 )
         end      
         cChar = ""
      endif  

      cCleanString := cCleanString + cChar
   next

return cCleanString

[Antonio Linares]

Already working fine as a Class. There is a delay after the prompt ">" is shown that is a bug in my code. Hopefully we solve it soon :-)

The response speed is really good :-)

https://fivetechsupport.com/forums/viewtopic.php?p=258431#p258431

[Antonio Linares]

Official repo:

https://github.com/FiveTechSoft/Harbour_GPT4All

Explanation of the project:

https://github.com/FiveTechSoft/Harbour_GPT4All/blob/main/README.md

best regards

[José M. C. Quintas]

What about it?

José M. C. Quintas

--
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
Unsubscribe: harbour-user...@googlegroups.com
Web: https://groups.google.com/group/harbour-users
---
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to harbour-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/harbour-users/c3d5bdd2-0162-4b9a-b79d-ac2ba346800dn%40googlegroups.com.

[Antonio Linares]

Dear Jose,

Try cloning the repo to download it:

git clone https://github.com/nomic-ai/gpt4all

best regards,

[Antonio Linares]

Modified demo code to work in a loop where the user can write what he wants. The Harbour app can talk to the AI without user intervention too.

#include "hbclass.ch"
#include "FileIO.ch"

function Main()

   local dummy := QOut( "Loading GPT4All... cpu speed: " + AllTrim( CPUSpeed() ),;
                        If( ! CpuHasAvx2(), "not", "" ) + " AVX2 support. Type exit to finish" )
   local oAI := GPT4All():New(), cMsg

   oAI:Read()
   while ! Empty( cMsg := oAI:Input() )
      if cMsg != "exit"
         oAI:Write( cMsg )
         oAI:Read()
      else
         exit
      end      
   end  

   oAI:End()

return nil  

repo updated

best regards

[Antonio Linares]

You need at least 6 gigs (better 8) of RAM and a CPU with AVX2 support, so the IA engine works fine

[alkresin]

Antonio,

  thanks for sharing this stuff.

 Personally I do not believe in the fundamental possibility of creating a so-called strong ai, at least, with the help of existing technologies and computational methods based on a Turing machine, but, IMO, trying to create such intelligence is the only task truly deserving of a programmer-researcher.

And, of course, for a lot of tasks, imitation of  the intelligence gives a satisfactory result.

Regards, Alexander.

вторник, 4 апреля 2023 г. в 12:47:37 UTC+3, Antonio Linares: