{
 "tag": "sm-120",
 "kind": null,
 "product": null,
 "used_by": [],
 "parts": [],
 "notes": [
  {
   "handle": "sarg",
   "id": "fix-for-llama-cpp-cudamalloc-out-of-memory",
   "title": "Fix for llama.cpp cudaMalloc out of memory loading a 35GB Q8_0 MoE GGUF on a 48GB RTX PRO 5000",
   "status": "working"
  },
  {
   "handle": "sarg",
   "id": "fix-for-minimax-h3-not-fitting-on-a",
   "title": "Fix for MiniMax-H3 not fitting on a single RTX PRO 5000 48GB when running local video+audio generation",
   "status": "working"
  },
  {
   "handle": "sarg",
   "id": "flux2-klein-misassembled-model-dir-needs-qwen3-te-and-bf16-dequant",
   "title": "Fix for shape-mismatch and unloadable-fp8 errors running /hd2/models/FLUX.2-klein-9b-kv-fp8 \u2014 dir was mis-assembled; pair the klein-KV transformer with Qwen3-8B, dequant fp8 to bf16, 4-step sampling",
   "status": "working"
  },
  {
   "handle": "sarg",
   "id": "nvml-driver-library-mismatch-breaks-nvidia-smi-not-cuda",
   "title": "Fix for nvidia-smi 'Failed to initialize NVML: Driver/library version mismatch' \u2014 CUDA compute still works, verify with a real GPU op instead of rebooting",
   "status": "working"
  }
 ]
}