fix: guard reconcile call against None numeric_ids; add torchvision lower bounds

This commit is contained in:
LeoBorcherding 2026-05-11 00:27:10 -05:00
commit 265a09a341
2 changed files with 5 additions and 6 deletions

View file

@ -662,11 +662,10 @@ def get_visible_gpu_utilization() -> Dict[str, Any]:
)
if result is not None:
result["backend"] = _backend_label(device)
if IS_ROCM:
numeric_ids = parent_visible_spec.get("numeric_ids")
if IS_ROCM and numeric_ids is not None:
# Fix unified-memory VRAM on AMD iGPUs (Strix Halo etc.)
_reconcile_rocm_unified_memory(
result, parent_visible_spec["numeric_ids"]
)
_reconcile_rocm_unified_memory(result, numeric_ids)
return result
# Torch-based fallback for CUDA (nvidia-smi unavailable, AMD ROCm) and XPU (Intel)

View file

@ -61,8 +61,8 @@ _ROCM_TORCH_PKG_SPECS: dict[str, tuple[str, str, str]] = {
# Default for rocm7.1 and earlier: torch 2.x below 2.11
"_default": (
"torch>=2.4,<2.11.0",
"torchvision<0.26.0",
"torchaudio<2.11.0",
"torchvision>=0.19,<0.26.0",
"torchaudio>=2.4,<2.11.0",
),
}
_PYTORCH_WHL_BASE = (