172 lines
4.6 KiB
PowerShell
172 lines
4.6 KiB
PowerShell
param(
|
|
[string]$SshTarget = "docker-gpu",
|
|
[string]$DockerHost = "ssh://docker-gpu",
|
|
[switch]$PlanOnly,
|
|
[switch]$SkipDownload,
|
|
[switch]$SkipPreflight,
|
|
[switch]$SkipTrain,
|
|
[switch]$SkipConvert,
|
|
[switch]$SkipRestart,
|
|
[switch]$SkipSmoke,
|
|
[switch]$DetachedDownload,
|
|
[switch]$DetachedConvert,
|
|
[string]$BaseModelCardId = "qwen3-coder-30b-a3b-instruct",
|
|
[string]$BaseModelDir = "/models/base/qwen3-coder-30b-a3b-instruct",
|
|
[string]$AdapterDir = "/models/adapters/1c/qwen3-coder-30b-a3b-1c-lora-v1",
|
|
[string]$AdapterGgufPath = "/models/adapters/1c/qwen3-coder-30b-a3b-1c-lora-v1.gguf",
|
|
[string]$InferenceBaseUrl = "http://docker-gpu.cin.su:8081",
|
|
[string]$InferenceModel = "qwen3-coder-1c-q6"
|
|
)
|
|
|
|
$ErrorActionPreference = "Stop"
|
|
|
|
function Invoke-Step {
|
|
param(
|
|
[Parameter(Mandatory = $true)]
|
|
[string]$Name,
|
|
[Parameter(Mandatory = $true)]
|
|
[string[]]$Command
|
|
)
|
|
|
|
Write-Host ""
|
|
Write-Host "== $Name"
|
|
Write-Host ($Command -join " ")
|
|
if ($PlanOnly) {
|
|
return
|
|
}
|
|
& $Command[0] @($Command | Select-Object -Skip 1)
|
|
if ($LASTEXITCODE -ne 0) {
|
|
throw "Step failed with exit code ${LASTEXITCODE}: ${Name}"
|
|
}
|
|
}
|
|
|
|
$steps = @()
|
|
|
|
$steps += @{
|
|
Name = "GPU host preflight"
|
|
Command = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/check_gpu_host.ps1",
|
|
"-SshTarget", $SshTarget
|
|
)
|
|
}
|
|
|
|
if (-not $SkipDownload) {
|
|
$downloadCommand = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/download_hf_range_gpu.ps1",
|
|
"-CardId", $BaseModelCardId,
|
|
"-LocalDir", $BaseModelDir,
|
|
"-ContainerName", "llm-hf-range-qwen3-coder-base"
|
|
)
|
|
if ($DetachedDownload) {
|
|
$downloadCommand += "-Detached"
|
|
}
|
|
$steps += @{
|
|
Name = "Download base HF model"
|
|
Command = $downloadCommand
|
|
}
|
|
}
|
|
|
|
if (-not $SkipPreflight) {
|
|
$steps += @{
|
|
Name = "Training preflight"
|
|
Command = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/run_1c_lora_training_gpu.ps1",
|
|
"-PreflightOnly"
|
|
)
|
|
}
|
|
}
|
|
|
|
if (-not $SkipTrain) {
|
|
$steps += @{
|
|
Name = "Train LoRA adapter"
|
|
Command = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/run_1c_lora_training_gpu.ps1"
|
|
)
|
|
}
|
|
}
|
|
|
|
if (-not $SkipConvert) {
|
|
$convertCommand = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/convert_1c_lora_to_gguf_gpu.ps1",
|
|
"-AdapterDir", $AdapterDir,
|
|
"-BaseModelDir", $BaseModelDir,
|
|
"-OutputFile", $AdapterGgufPath
|
|
)
|
|
if ($DetachedConvert) {
|
|
$convertCommand += "-Detached"
|
|
}
|
|
$steps += @{
|
|
Name = "Convert adapter to GGUF"
|
|
Command = $convertCommand
|
|
}
|
|
}
|
|
|
|
if (-not $SkipRestart) {
|
|
$steps += @{
|
|
Name = "Stop Q6 service"
|
|
Command = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/manage_gpu_q6_service.ps1",
|
|
"-Action", "stop"
|
|
)
|
|
}
|
|
$steps += @{
|
|
Name = "Start Q6 service with LoRA"
|
|
Command = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/manage_gpu_q6_service.ps1",
|
|
"-Action", "start",
|
|
"-LoraPath", $AdapterGgufPath
|
|
)
|
|
}
|
|
$steps += @{
|
|
Name = "Check Q6 service status"
|
|
Command = @(
|
|
"powershell", "-NoProfile", "-ExecutionPolicy", "Bypass",
|
|
"-File", "scripts/manage_gpu_q6_service.ps1",
|
|
"-Action", "status",
|
|
"-LoraPath", $AdapterGgufPath
|
|
)
|
|
}
|
|
}
|
|
|
|
if (-not $SkipSmoke) {
|
|
$steps += @{
|
|
Name = "Check inference endpoint"
|
|
Command = @(
|
|
"python", "scripts/check_inference_endpoint.py",
|
|
"--base-url", $InferenceBaseUrl,
|
|
"--expected-model", $InferenceModel,
|
|
"--print"
|
|
)
|
|
}
|
|
$steps += @{
|
|
Name = "Run 1C smoke eval"
|
|
Command = @(
|
|
"python", "scripts/run_1c_smoke_eval.py",
|
|
"--base-url", $InferenceBaseUrl,
|
|
"--model", $InferenceModel
|
|
)
|
|
}
|
|
}
|
|
|
|
foreach ($step in $steps) {
|
|
if ($SkipPreflight -and $step.Name -eq "GPU host preflight") {
|
|
continue
|
|
}
|
|
Invoke-Step -Name $step.Name -Command $step.Command
|
|
}
|
|
|
|
if ($PlanOnly) {
|
|
Write-Host ""
|
|
Write-Host "Plan complete. Re-run without -PlanOnly to execute."
|
|
} else {
|
|
Write-Host ""
|
|
Write-Host "Q6 LoRA training/publish flow completed."
|
|
}
|