Skip to content

Commit 9a55f28

Browse files
Minh141120Minh141120
committed
fix: enhaance windows binary verification and server start tests for vulkan compatibility
1 parent 95d25ce commit 9a55f28

File tree

1 file changed

+209
-119
lines changed

1 file changed

+209
-119
lines changed

‎.github/workflows/test-binaries.yml‎

Lines changed: 209 additions & 119 deletions
Original file line numberDiff line numberDiff line change
@@ -310,58 +310,73 @@ jobs:
310310
echo "Available arguments:"
311311
./llama/build/bin/${{ matrix.binary-name }} --help || echo "Help check completed"
312312
313-
- name: Verify binary (Windows)
314-
if: runner.os == 'Windows'
315-
shell: pwsh
316-
run: |
317-
Write-Host "Testing binary basic functionality..."
318-
319-
# Test version command
320-
try {
321-
$versionOutput = & ".\llama\build\bin\${{ matrix.binary-name }}" --version 2>&1
322-
Write-Host "Version output:"
323-
Write-Host $versionOutput
324-
} catch {
325-
Write-Host "Version check failed: $($_.Exception.Message)"
326-
}
327-
328-
Write-Host "Available arguments:"
329-
try {
330-
# Use Start-Process to capture output properly and avoid exit code issues
331-
$helpProcess = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
332-
-ArgumentList "--help" `
333-
-RedirectStandardOutput "help_output.txt" `
334-
-RedirectStandardError "help_error.txt" `
335-
-Wait -PassThru -WindowStyle Hidden
336-
337-
if (Test-Path "help_output.txt") {
338-
$helpContent = Get-Content "help_output.txt" -Raw
339-
if ($helpContent) {
340-
Write-Host $helpContent
341-
}
342-
}
343-
344-
if (Test-Path "help_error.txt") {
345-
$errorContent = Get-Content "help_error.txt" -Raw
346-
if ($errorContent) {
347-
Write-Host "Help stderr:"
348-
Write-Host $errorContent
349-
}
350-
}
351-
352-
Write-Host "Help command exit code: $($helpProcess.ExitCode)"
353-
354-
} catch {
355-
Write-Host "Help check failed: $($_.Exception.Message)"
356-
Write-Host "This might be normal for some binary versions"
357-
}
358-
359-
# Clean up temp files
360-
Remove-Item -Path "help_output.txt" -ErrorAction SilentlyContinue
361-
Remove-Item -Path "help_error.txt" -ErrorAction SilentlyContinue
362-
363-
# Don't fail the step - verification is informational
364-
Write-Host "Binary verification completed"
313+
- name: Verify binary (Windows)
314+
if: runner.os == 'Windows'
315+
shell: pwsh
316+
run: |
317+
Write-Host "Testing binary basic functionality..."
318+
319+
# Test if binary exists and is accessible
320+
if (-not (Test-Path ".\llama\build\bin\${{ matrix.binary-name }}")) {
321+
Write-Host "ERROR: Binary not found at expected location"
322+
exit 1
323+
}
324+
325+
Write-Host "Binary found, testing basic functionality..."
326+
327+
# Test version command with better error handling
328+
try {
329+
$versionOutput = & ".\llama\build\bin\${{ matrix.binary-name }}" --version 2>&1
330+
Write-Host "Version output:"
331+
Write-Host $versionOutput
332+
} catch {
333+
Write-Host "Version check failed: $($_.Exception.Message)"
334+
Write-Host "This might be normal for some binary versions"
335+
}
336+
337+
Write-Host "Available arguments:"
338+
try {
339+
# Use Start-Process to capture output properly and avoid exit code issues
340+
$helpProcess = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
341+
-ArgumentList "--help" `
342+
-RedirectStandardOutput "help_output.txt" `
343+
-RedirectStandardError "help_error.txt" `
344+
-Wait -PassThru -WindowStyle Hidden
345+
346+
if (Test-Path "help_output.txt") {
347+
$helpContent = Get-Content "help_output.txt" -Raw
348+
if ($helpContent) {
349+
Write-Host $helpContent
350+
}
351+
}
352+
353+
if (Test-Path "help_error.txt") {
354+
$errorContent = Get-Content "help_error.txt" -Raw
355+
if ($errorContent) {
356+
Write-Host "Help stderr:"
357+
Write-Host $errorContent
358+
}
359+
}
360+
361+
Write-Host "Help command exit code: $($helpProcess.ExitCode)"
362+
363+
# If help command crashes, that's okay - we'll test actual functionality
364+
if ($helpProcess.ExitCode -eq -1073741515) {
365+
Write-Host "Binary crashed during help command (access violation) - this is expected for some builds"
366+
Write-Host "Will proceed to test actual model loading and inference"
367+
}
368+
369+
} catch {
370+
Write-Host "Help check failed: $($_.Exception.Message)"
371+
Write-Host "This might be normal for some binary versions"
372+
}
373+
374+
# Clean up temp files
375+
Remove-Item -Path "help_output.txt" -ErrorAction SilentlyContinue
376+
Remove-Item -Path "help_error.txt" -ErrorAction SilentlyContinue
377+
378+
# Don't fail the step - verification is informational
379+
Write-Host "Binary verification completed"
365380
366381
- name: Test server startup (Linux/macOS)
367382
if: runner.os != 'Windows'
@@ -379,13 +394,24 @@ jobs:
379394
# Try to start server without --server argument (which doesn't exist in this version)
380395
echo "Attempting to start server..."
381396
382-
# Method 1: Try modern server startup (no --server flag)
383-
./llama/build/bin/${{ matrix.binary-name }} \
384-
--model models/Lucy-Q4_0.gguf \
385-
--port 8080 --host 127.0.0.1 \
386-
--ctx-size 512 \
387-
--n-gpu-layers 0 &
388-
SERVER_PID=$!
397+
# Method 1: Try modern server startup (no --server flag)
398+
# For Vulkan builds, force CPU mode to avoid driver issues
399+
if [[ "${{ matrix.name }}" == *"vulkan"* ]]; then
400+
echo "Vulkan build detected, forcing CPU mode to avoid driver issues..."
401+
./llama/build/bin/${{ matrix.binary-name }} \
402+
--model models/Lucy-Q4_0.gguf \
403+
--port 8080 --host 127.0.0.1 \
404+
--ctx-size 512 \
405+
--n-gpu-layers 0 \
406+
--n-gpu-layers-tok 0 &
407+
else
408+
./llama/build/bin/${{ matrix.binary-name }} \
409+
--model models/Lucy-Q4_0.gguf \
410+
--port 8080 --host 127.0.0.1 \
411+
--ctx-size 512 \
412+
--n-gpu-layers 0 &
413+
fi
414+
SERVER_PID=$!
389415
390416
echo "Server PID: $SERVER_PID"
391417
sleep 5
@@ -394,13 +420,24 @@ jobs:
394420
if ! kill -0 $SERVER_PID 2>/dev/null; then
395421
echo "Modern format failed, trying legacy format..."
396422
397-
# Method 2: Try legacy short arguments
398-
./llama/build/bin/${{ matrix.binary-name }} \
399-
-m models/Lucy-Q4_0.gguf \
400-
-p 8080 \
401-
-c 512 \
402-
--n-gpu-layers 0 &
403-
SERVER_PID=$!
423+
# Method 2: Try legacy short arguments
424+
# For Vulkan builds, force CPU mode to avoid driver issues
425+
if [[ "${{ matrix.name }}" == *"vulkan"* ]]; then
426+
echo "Vulkan build detected, forcing CPU mode to avoid driver issues..."
427+
./llama/build/bin/${{ matrix.binary-name }} \
428+
-m models/Lucy-Q4_0.gguf \
429+
-p 8080 \
430+
-c 512 \
431+
--n-gpu-layers 0 \
432+
--n-gpu-layers-tok 0 &
433+
else
434+
./llama/build/bin/${{ matrix.binary-name }} \
435+
-m models/Lucy-Q4_0.gguf \
436+
-p 8080 \
437+
-c 512 \
438+
--n-gpu-layers 0 &
439+
fi
440+
SERVER_PID=$!
404441
405442
sleep 5
406443
@@ -579,68 +616,104 @@ jobs:
579616
# Fallback: Direct completion test
580617
echo "Testing direct completion mode..."
581618
582-
# Try different completion argument formats
583-
echo "Trying modern completion format..."
584-
./llama/build/bin/${{ matrix.binary-name }} \
585-
--model models/Lucy-Q4_0.gguf \
586-
--prompt "Hello" \
587-
--n-predict 5 \
588-
--ctx-size 512 \
589-
--n-gpu-layers 0 \
590-
--temp 0.1 > completion_output.txt 2>&1
619+
# Try different completion argument formats
620+
echo "Trying modern completion format..."
621+
# For Vulkan builds, force CPU mode to avoid driver issues
622+
if [[ "${{ matrix.name }}" == *"vulkan"* ]]; then
623+
echo "Vulkan build detected, forcing CPU mode to avoid driver issues..."
624+
./llama/build/bin/${{ matrix.binary-name }} \
625+
--model models/Lucy-Q4_0.gguf \
626+
--prompt "Hello" \
627+
--n-predict 5 \
628+
--ctx-size 512 \
629+
--n-gpu-layers 0 \
630+
--n-gpu-layers-tok 0 \
631+
--temp 0.1 > completion_output.txt 2>&1
632+
else
633+
./llama/build/bin/${{ matrix.binary-name }} \
634+
--model models/Lucy-Q4_0.gguf \
635+
--prompt "Hello" \
636+
--n-predict 5 \
637+
--ctx-size 512 \
638+
--n-gpu-layers 0 \
639+
--temp 0.1 > completion_output.txt 2>&1
640+
fi
591641
COMPLETION_EXIT_CODE=$?
592642
593643
echo "Modern completion exit code: $COMPLETION_EXIT_CODE"
594644
echo "Modern completion output:"
595645
cat completion_output.txt || echo "No output"
596646
597-
# Check if we got any meaningful output (even with exit code 1)
598-
if [ -s completion_output.txt ] && ! grep -q "error:" completion_output.txt && grep -q "Hello" completion_output.txt; then
599-
echo "[PASSED] Modern completion test passed (got meaningful output)"
600-
echo "Completion output:"
601-
cat completion_output.txt
602-
exit 0
603-
fi
604-
605-
# Try legacy format
606-
echo "Trying legacy completion format..."
607-
./llama/build/bin/${{ matrix.binary-name }} \
608-
-m models/Lucy-Q4_0.gguf \
609-
-p "Hello" \
610-
-n 5 \
611-
-c 512 \
612-
--n-gpu-layers 0 > completion_output2.txt 2>&1
647+
# Check if we got any meaningful output (even with exit code 1)
648+
if [ -s completion_output.txt ] && ! grep -q "error:" completion_output.txt && (grep -q "Hello" completion_output.txt || grep -q "llama_print_timings" completion_output.txt); then
649+
echo "[PASSED] Modern completion test passed (got meaningful output)"
650+
echo "Completion output:"
651+
cat completion_output.txt
652+
exit 0
653+
fi
654+
655+
# Try legacy format
656+
echo "Trying legacy completion format..."
657+
# For Vulkan builds, force CPU mode to avoid driver issues
658+
if [[ "${{ matrix.name }}" == *"vulkan"* ]]; then
659+
echo "Vulkan build detected, forcing CPU mode to avoid driver issues..."
660+
./llama/build/bin/${{ matrix.binary-name }} \
661+
-m models/Lucy-Q4_0.gguf \
662+
-p "Hello" \
663+
-n 5 \
664+
-c 512 \
665+
--n-gpu-layers 0 \
666+
--n-gpu-layers-tok 0 > completion_output2.txt 2>&1
667+
else
668+
./llama/build/bin/${{ matrix.binary-name }} \
669+
-m models/Lucy-Q4_0.gguf \
670+
-p "Hello" \
671+
-n 5 \
672+
-c 512 \
673+
--n-gpu-layers 0 > completion_output2.txt 2>&1
674+
fi
613675
COMPLETION_EXIT_CODE=$?
614676
615677
echo "Legacy completion exit code: $COMPLETION_EXIT_CODE"
616678
echo "Legacy completion output:"
617679
cat completion_output2.txt || echo "No output"
618680
619-
if [ -s completion_output2.txt ] && ! grep -q "error:" completion_output2.txt && grep -q "Hello" completion_output2.txt; then
620-
echo "[PASSED] Legacy completion test passed (got meaningful output)"
621-
echo "Completion output:"
622-
cat completion_output2.txt
623-
exit 0
624-
fi
625-
626-
# Try simplest format
627-
echo "Trying simplest completion format..."
628-
./llama/build/bin/${{ matrix.binary-name }} \
629-
-m models/Lucy-Q4_0.gguf \
630-
-p "Hello" \
631-
-n 5 > completion_output3.txt 2>&1
681+
if [ -s completion_output2.txt ] && ! grep -q "error:" completion_output2.txt && (grep -q "Hello" completion_output2.txt || grep -q "llama_print_timings" completion_output2.txt); then
682+
echo "[PASSED] Legacy completion test passed (got meaningful output)"
683+
echo "Completion output:"
684+
cat completion_output2.txt
685+
exit 0
686+
fi
687+
688+
# Try simplest format
689+
echo "Trying simplest completion format..."
690+
# For Vulkan builds, force CPU mode to avoid driver issues
691+
if [[ "${{ matrix.name }}" == *"vulkan"* ]]; then
692+
echo "Vulkan build detected, forcing CPU mode to avoid driver issues..."
693+
./llama/build/bin/${{ matrix.binary-name }} \
694+
-m models/Lucy-Q4_0.gguf \
695+
-p "Hello" \
696+
-n 5 \
697+
--n-gpu-layers 0 \
698+
--n-gpu-layers-tok 0 > completion_output3.txt 2>&1
699+
else
700+
./llama/build/bin/${{ matrix.binary-name }} \
701+
-m models/Lucy-Q4_0.gguf \
702+
-p "Hello" \
703+
-n 5 > completion_output3.txt 2>&1
704+
fi
632705
COMPLETION_EXIT_CODE=$?
633706
634707
echo "Simple completion exit code: $COMPLETION_EXIT_CODE"
635708
echo "Simple completion output:"
636709
cat completion_output3.txt || echo "No output"
637710
638-
if [ -s completion_output3.txt ] && ! grep -q "error:" completion_output3.txt && grep -q "Hello" completion_output3.txt; then
639-
echo "[PASSED] Simple completion test passed (got meaningful output)"
640-
echo "Completion output:"
641-
cat completion_output3.txt
642-
exit 0
643-
fi
711+
if [ -s completion_output3.txt ] && ! grep -q "error:" completion_output3.txt && (grep -q "Hello" completion_output3.txt || grep -q "llama_print_timings" completion_output3.txt); then
712+
echo "[PASSED] Simple completion test passed (got meaningful output)"
713+
echo "Completion output:"
714+
cat completion_output3.txt
715+
exit 0
716+
fi
644717
645718
echo "[FAILED] All completion formats failed"
646719
echo "Modern format output:"
@@ -660,11 +733,20 @@ jobs:
660733
run: |
661734
Write-Host "Testing ${{ matrix.binary-name }} server startup..."
662735
663-
# Start server with CPU mode and capture output
664-
$logFile = "server_output.log"
665-
$process = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
666-
-ArgumentList "--model", "models\Lucy-Q4_0.gguf", "--port", "8080", "--host", "127.0.0.1", "--n-gpu-layers", "0", "--ctx-size", "512" `
667-
-WindowStyle Hidden -PassThru -RedirectStandardOutput $logFile -RedirectStandardError "server_error.log"
736+
# Start server with CPU mode and capture output
737+
$logFile = "server_output.log"
738+
739+
# For Vulkan builds, force CPU mode to avoid driver issues
740+
if ("${{ matrix.name }}" -like "*vulkan*") {
741+
Write-Host "Vulkan build detected, forcing CPU mode to avoid driver issues..."
742+
$process = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
743+
-ArgumentList "--model", "models\Lucy-Q4_0.gguf", "--port", "8080", "--host", "127.0.0.1", "--n-gpu-layers", "0", "--n-gpu-layers-tok", "0", "--ctx-size", "512" `
744+
-WindowStyle Hidden -PassThru -RedirectStandardOutput $logFile -RedirectStandardError "server_error.log"
745+
} else {
746+
$process = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
747+
-ArgumentList "--model", "models\Lucy-Q4_0.gguf", "--port", "8080", "--host", "127.0.0.1", "--n-gpu-layers", "0", "--ctx-size", "512" `
748+
-WindowStyle Hidden -PassThru -RedirectStandardOutput $logFile -RedirectStandardError "server_error.log"
749+
}
668750
669751
Write-Host "Server PID: $($process.Id)"
670752
@@ -755,10 +837,18 @@ jobs:
755837
run: |
756838
Write-Host "Testing inference with ${{ matrix.binary-name }}..."
757839
758-
# Start server
759-
$process = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
760-
-ArgumentList "--model", "models\Lucy-Q4_0.gguf", "--port", "8080", "--host", "127.0.0.1", "--n-gpu-layers", "0", "--ctx-size", "512" `
761-
-WindowStyle Hidden -PassThru
840+
# Start server
841+
# For Vulkan builds, force CPU mode to avoid driver issues
842+
if ("${{ matrix.name }}" -like "*vulkan*") {
843+
Write-Host "Vulkan build detected, forcing CPU mode to avoid driver issues..."
844+
$process = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
845+
-ArgumentList "--model", "models\Lucy-Q4_0.gguf", "--port", "8080", "--host", "127.0.0.1", "--n-gpu-layers", "0", "--n-gpu-layers-tok", "0", "--ctx-size", "512" `
846+
-WindowStyle Hidden -PassThru
847+
} else {
848+
$process = Start-Process -FilePath ".\llama\build\bin\${{ matrix.binary-name }}" `
849+
-ArgumentList "--model", "models\Lucy-Q4_0.gguf", "--port", "8080", "--host", "127.0.0.1", "--n-gpu-layers", "0", "--ctx-size", "512" `
850+
-WindowStyle Hidden -PassThru
851+
}
762852
763853
# Wait for server to start
764854
for ($i = 1; $i -le 30; $i++) {

0 commit comments

Comments
 (0)