51 lines
1.5 KiB
PowerShell
51 lines
1.5 KiB
PowerShell
param(
|
|
[Parameter(Mandatory = $true)]
|
|
[string]$SshTarget,
|
|
|
|
[int]$SshPort = 22,
|
|
[int]$GpuCount = 4,
|
|
[int]$Runs = 2,
|
|
[string]$TopologyClass = "mixed",
|
|
[string]$SshKey = (Join-Path $HOME ".ssh\inceptal-dev-envs")
|
|
)
|
|
|
|
$ErrorActionPreference = "Stop"
|
|
|
|
$repo = "/workspace/h3-runtime"
|
|
$python = "/runpod-volume/h3-runtime/h3-venv/bin/python"
|
|
$benchmark = "benchmarks/t2va-dialogue-quoted-1344x768-243f-base12-sdpa-seed440420.json"
|
|
$model = "/runpod-volume/ComfyUI/models/diffusion_models/minimax_h3_fl2va_pruned_nvfp4.safetensors"
|
|
$textEncoder = "/runpod-volume/ComfyUI/models/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors"
|
|
$reportRoot = "/runpod-volume/h3-benchmarks"
|
|
$reportStem = "rtxpro6000-server-${GpuCount}gpu-${TopologyClass}-direct-ulysses-sdpa-1344x768-243f-base12-seed440420"
|
|
|
|
$common = @(
|
|
"PYTHONPATH=$repo/src"
|
|
$python
|
|
"-m torch.distributed.run"
|
|
"--standalone"
|
|
"--nnodes=1"
|
|
"--nproc-per-node=$GpuCount"
|
|
"tools/distributed_t2va.py"
|
|
"--attention sdpa"
|
|
"--mode ulysses"
|
|
"--model $model"
|
|
"--text-encoder $textEncoder"
|
|
) -join " "
|
|
|
|
$commands = @(
|
|
"stty -echo"
|
|
"cd $repo"
|
|
"unset H3_DISABLE_MMAP H3_FAST_SAFETENSORS"
|
|
)
|
|
|
|
for ($run = 1; $run -le $Runs; $run++) {
|
|
$report = "$reportRoot/$reportStem-run$run.json"
|
|
$commands += "$common --benchmark $benchmark --report $report"
|
|
}
|
|
|
|
$commands += "exit"
|
|
$commandsText = ($commands -join "`n") + "`n"
|
|
|
|
$commandsText | & ssh -tt -p $SshPort -i $SshKey $SshTarget
|
|
exit $LASTEXITCODE
|