Benchmark tests won't run after warming up!

Hi @cjpais,

Thanks for creating such a great tool for benchmarking and a website to share the results.
I was trying to benchmark my RTX 5090 on Windows 11 and CUDA 12.8.1. But it was not successful and the tool stopped at the warming up step!

Here are some logs:

PS C:\Users\Amir\Downloads> .\localscore-0.9.2.exe -m .\Llama-3.2-1B-Instruct-Q4_K_M.gguf
import_cuda_impl: initializing gpu module...
get_rocm_bin_path: note: amdclang++.exe not found on $PATH
get_rocm_bin_path: note: $HIP_PATH/bin/amdclang++.exe does not exist
get_rocm_bin_path: note: /opt/rocm/bin/amdclang++.exe does not exist
get_rocm_bin_path: note: clang++.exe not found on $PATH
get_rocm_bin_path: note: $HIP_PATH/bin/clang++.exe does not exist
get_rocm_bin_path: note: /opt/rocm/bin/clang++.exe does not exist
link_cuda_dso: note: dynamically linking /C/Users/Amir/.llamafile/v/0.9.2/ggml-rocm.dll
link_cuda_dso: warning: library not found: failed to load library
link_cuda_dso: note: dynamically linking /C/Users/Amir/.llamafile/v/0.9.2/ggml-rocm.dll
link_cuda_dso: warning: library not found: failed to load library
link_cuda_dso: note: dynamically linking /C/Users/Amir/.llamafile/v/0.9.2/ggml-cuda.dll
link_cuda_dso: GPU support loaded

=================== LocalScore Runtime Information ===================

llamafile version:   0.9.2
llama.cpp commit:    a30b324

======================================================================

========================= System Information =========================

Kernel Type:         Windows
Kernel Release:      10.0
Version:             Cosmopolitan 3.9.7 MODE=x86_64
System Architecture: x86_64
CPU:                 13th Gen Intel Core i9-13900KS (alderlake)
RAM:                 95.8 GiB

======================================================================

=================== Active GPU (GPU 0) Information ===================

GPU Name:                  NVIDIA GeForce RTX 5090
VRAM:                      32.0 GiB
Streaming Multiprocessors: 170
CUDA Capability:           12.0

======================================================================

Loading model... Model loaded.
Warming up......
PS C:\Users\Amir\Downloads>

The GPU and the CUDA toolkit seem to be detected correctly but the benchmark fails to run!

Also, the output of the nvidia-smi command:

PS C:\Users\Amir> nvidia-smi
Tue Apr  8 23:50:17 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 572.83                 Driver Version: 572.83         CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                  Driver-Model | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 5090      WDDM  |   00000000:01:00.0  On |                  N/A |
|  0%   44C    P0             50W /  600W |    1067MiB /  32607MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

Let me know if you need more info.

Reproduce the reported command on Windows 11 with the RTX 5090, CUDA 12.8.1, and the Llama-3.2-1B-Instruct-Q4_K_M.gguf model, then compare the logs around GPU loading and warming up. Done means the benchmark continues past warming up and produces benchmark results instead of exiting.