Fix for windows not installing models correctly

[okretsej]

i had a problem with the models not being loaded in. coming back as a null even tho it gave a success marker. now as why i am here should be obvious for an LLM portable. as i use LLMs daily. i had this issue fixed for you, i dont know where or if you had one already done. so here is the snippet from perplexity about the windows install since i am not using the other OS's atm, along with the code fixes, i ran it it worked out things load now.

Root Cause Analysis — install-core.ps1

Three bugs existed on a single line (line 472):

$null = & "$USB_Drive\Shared\bin\ollama-windows.exe" create $m.Local -f "Modelfile-$($m.Local)" 2>&1

Bug 1 — Silent failures:
$null = discards all output including errors. When ollama create failed,
the script printed "imported successfully!" anyway and moved on invisibly.

Bug 2 — Broken ZIP extraction (line 418):
The installer downloaded ollama-windows-amd64.zip but only moved ollama.exe
out of it, discarding llama-server.exe, llama-quantize.exe, and all runtime
DLLs bundled in the same ZIP. The installer was breaking its own engine on
every fresh install.

Bug 3 — Premature create call (line 467):
Start-Sleep -Seconds 5 is a blind wait. On a USB drive Ollama often isn't
ready in 5 seconds, so ollama create was hitting an unresponsive server.

════════════════════════════════════════════════════════════
FIX 1 — Verbose ollama create output
════════════════════════════════════════════════════════════

Replace:
try {
$null = & "$USB_Drive\Shared\bin\ollama-windows.exe" create $m.Local -f "Modelfile-$($m.Local)" 2>&1
Write-Host " $($m.Name) imported successfully!" -ForegroundColor Green
} catch {
Write-Host " ERROR: Failed to import $($m.Name)" -ForegroundColor Red
}

With:
Write-Host " Running: ollama-windows.exe create $($m.Local) -f Modelfile-$($m.Local)" -ForegroundColor DarkGray
$createOutput = & "$USB_Drive\Shared\bin\ollama-windows.exe" create $m.Local -f "Modelfile-$($m.Local)" 2>&1
Write-Host " Output: $createOutput" -ForegroundColor DarkGray
if ($LASTEXITCODE -eq 0) {
Write-Host " $($m.Name) imported successfully!" -ForegroundColor Green
} else {
Write-Host " ERROR: Failed to import $($m.Name) (exit $LASTEXITCODE)" -ForegroundColor Red
Write-Host " $createOutput" -ForegroundColor Red
}

════════════════════════════════════════════════════════════
FIX 2 — Extract ALL binaries from the Ollama ZIP
════════════════════════════════════════════════════════════

Replace the extraction block with:
New-Item -ItemType Directory -Force -Path $TempOllamaDir | Out-Null
Expand-Archive -Path $OllamaDest -DestinationPath $TempOllamaDir -Force

Get-ChildItem -Path $TempOllamaDir -Recurse -File | ForEach-Object {
    $dest = "$USB_Drive\Shared\bin\$($_.Name)"
    Move-Item -Path $_.FullName -Destination $dest -Force
    Write-Host "      Extracted: $($_.Name)" -ForegroundColor DarkGray
}

if (Test-Path "$USB_Drive\Shared\bin\ollama.exe") {
    Rename-Item "$USB_Drive\Shared\bin\ollama.exe" "ollama-windows.exe" -Force
}

Remove-Item $TempOllamaDir -Force -Recurse -ErrorAction SilentlyContinue
Remove-Item $OllamaDest -Force -ErrorAction SilentlyContinue
Write-Host "      Ollama Setup Complete!" -ForegroundColor Green

════════════════════════════════════════════════════════════
FIX 3 — Wait until Ollama is actually ready
════════════════════════════════════════════════════════════

Replace:
Start-Sleep -Seconds 5

With:
Write-Host " Waiting for Ollama to be ready..." -ForegroundColor DarkGray
$waited = 0
do {
Start-Sleep -Seconds 1
$waited++
$response = & curl.exe -s http://127.0.0.1:11434/api/tags 2>&1
} while ($response -notmatch '"models"' -and $waited -lt 30)
if ($waited -ge 30) {
Write-Host " WARNING: Ollama did not respond in 30 seconds!" -ForegroundColor Red
} else {
Write-Host " Ollama is ready (took ${waited}s)." -ForegroundColor Green
}

════════════════════════════════════════════════════════════
REMAINING WATCH POINTS (priority order)
════════════════════════════════════════════════════════════

1. chat_server.py ~line 722
   Empty model field proxied directly to Ollama → 400
   Fix: validate model exists before proxying

2. FastChatUI.html fetchModels()
   No retry if /api/tags returns [] on first load
   Fix: add retry loop with backoff

3. start-fast-chat.bat
   start /b hides Ollama window, crashes invisible
   Fix: use visible window during debugging

4. chat_server.py _proxy_ollama()
   No request body validation at all
   Fix: add schema check before forwarding

[okretsej]

install-core.txt
updated script

[okretsej]

sorry if this wasn't a good way to do this, if you need anything from me let me know.

[jmyrosotomayor1-debug]

yea, windows installation not working for me either

[okretsej]

Jmy, grab the install core.txt open the the script in notepad++(or similar) copy paste that txt into that script
Its a full drop in replacement that fixes the issue on the windows install. You can also just change.txt to ps1 and replace that file. It will load in everything correctly.

[okretsej]

I had perplexity deep research the code in this hub and it found the problem and gave me a fix. Thats what the first post is about the second post is the full drop in replacement for install_core.py. dont really use github often didnt know how to contact the authors so i put everything i did here.