DPSCoach

# Model Path Cleanup Guide ## Problem Fixed The app was confused about model locations and had an incorrect SHA hash for the downloaded model. This has been fixed by: 1. **Single runtime directory**: All models now live in `%APPDATA%\DPSCoach\models\` 2. **Unique filenames**: No more generic `model.gguf` - each model has a unique name 3. **Atomic downloads with SHA verification**: Downloads verify hash before finalizing 4. **Clear error messages**: Validation errors now show exact file paths and SHA mismatches ## What Changed - **Default model path**: `%APPDATA%\DPSCoach\models\qwen2.5-7b-instruct-q4_k_m.gguf` - **Fallback model path**: `%APPDATA%\DPSCoach\models\qwen2.5-0.5b-instruct-q4_k_m.gguf` - **Expected SHA for default model**: `8A45E4E923F03F4106BCC375CB83FC383C2C570529ACB2085C6468DF10B50AD5` - **Expected SHA for fallback model**: `9FECC3B3CD76BBA89D504F29B616EEDF7DA85B96540E490CA5824D3F7D2776A0` ## Cleanup Steps ### 1. Delete the incorrectly downloaded model ```powershell # This file has wrong SHA (65B8FCD9... instead of expected 8A45E4E9...) Remove-Item "$env:APPDATA\DPSCoach\models\Qwen2.5-7B-Instruct-Q4_K_M.gguf" -Force ``` ### 2. Verify cleanup ```powershell Get-ChildItem "$env:APPDATA\DPSCoach\models\" -Filter "*.gguf" ``` Should show nothing (or only the fallback model if you want to keep it). ### 3. Download the correct model **Option A: Use the app UI** - Launch the app: `python -m app.main` - Click "Download Model (~4.7 GB)" button - Wait for download + SHA verification - App will now validate the correct file **Option B: Manual download and verify** ```powershell # Create directory New-Item -ItemType Directory -Force -Path "$env:APPDATA\DPSCoach\models" # Download (this will take a while - ~4.7 GB) $url = "https://huggingface.co/bartowski/Qwen2.5-7B-Instruct-GGUF/resolve/main/Qwen2.5-7B-Instruct-Q4_K_M.gguf" $dest = "$env:APPDATA\DPSCoach\models\qwen2.5-7b-instruct-q4_k_m.gguf" Invoke-WebRequest -Uri $url -OutFile $dest # Verify SHA256 $hash = (Get-FileHash -Path $dest -Algorithm SHA256).Hash Write-Host "Downloaded SHA: $hash" Write-Host "Expected SHA: 8A45E4E923F03F4106BCC375CB83FC383C2C570529ACB2085C6468DF10B50AD5" if ($hash -eq "8A45E4E923F03F4106BCC375CB83FC383C2C570529ACB2085C6468DF10B50AD5") { Write-Host "✓ SHA matches! Model is valid." -ForegroundColor Green } else { Write-Host "✗ SHA mismatch! Delete and re-download." -ForegroundColor Red } ``` ### 4. Optional: Keep fallback model in repo If you want to use the smaller fallback model (0.5B) from the repo: ```powershell # Copy to runtime location with correct name Copy-Item "models\model.gguf" "$env:APPDATA\DPSCoach\models\qwen2.5-0.5b-instruct-q4_k_m.gguf" # Verify it Get-FileHash "$env:APPDATA\DPSCoach\models\qwen2.5-0.5b-instruct-q4_k_m.gguf" -Algorithm SHA256 # Should show: 9FECC3B3CD76BBA89D504F29B616EEDF7DA85B96540E490CA5824D3F7D2776A0 # To use fallback model instead of default: $env:DPSCOACH_USE_FALLBACK_MODEL = "1" python -m app.main ``` ## Verification ### Check model locations ```powershell powershell scripts\find_models.ps1 ``` ### Expected output after cleanup: ``` === GGUF Model Inventory === Location: APPDATA\DPSCoach Path: C:\Users\Allen\AppData\Roaming\DPSCoach - qwen2.5-7b-instruct-q4_k_m.gguf Full path: C:\Users\Allen\AppData\Roaming\DPSCoach\models\qwen2.5-7b-instruct-q4_k_m.gguf Size: 4466.13 MB SHA256: 8A45E4E923F03F4106BCC375CB83FC383C2C570529ACB2085C6468DF10B50AD5 Location: LOCALAPPDATA\DPSCoach Path: C:\Users\Allen\AppData\Local\DPSCoach (directory does not exist) Location: Repo models Path: C:\Users\Allen\Desktop\Data Analyst Projects\tl-dps-mcp\models - model.gguf Full path: C:\Users\Allen\Desktop\Data Analyst Projects\tl-dps-mcp\models\model.gguf Size: 637.81 MB SHA256: 9FECC3B3CD76BBA89D504F29B616EEDF7DA85B96540E490CA5824D3F7D2776A0 ``` ## Environment Variables ### Model selection - **Default (Qwen 7B)**: No env var needed - **Fallback (Qwen 0.5B)**: `$env:DPSCOACH_USE_FALLBACK_MODEL = "1"` ### Coach routing - **Default (analysis packet)**: No env var needed - **Legacy SQL mode**: `$env:DPSCOACH_USE_LEGACY_SQL = "1"` ## Troubleshooting ### App shows "Model validation failed: SHA256 mismatch" - The file is corrupted or incomplete - Delete it and re-download using the app or manual steps above - Error message will show expected vs found SHA ### "Model file not found" - File is not in the expected location - Run `powershell scripts\find_models.ps1` to see where files are - Use app download button or manual download steps ### Want to use a different model Use "Choose Local Model File..." in the app UI - it will validate: 1. GGUF magic header 2. Minimum file size 3. SHA256 (if you configure expected hash in code) ## Summary - ✓ Runtime models directory: `%APPDATA%\DPSCoach\models\` - ✓ Default model filename: `qwen2.5-7b-instruct-q4_k_m.gguf` - ✓ Fallback model filename: `qwen2.5-0.5b-instruct-q4_k_m.gguf` - ✓ Downloads verify SHA before finalizing (.part → final) - ✓ Validation errors show exact paths and SHA mismatches - ✓ No more generic `model.gguf` confusion