b9426: llama : do not skip iGPU when only RPC devices are present (#23868)
The llama.cpp project has released version b9426, addressing an issue where integrated GPUs were incorrectly skipped when RPC devices were present. This fix ensures that local iGPUs are not overlooked, preventing potential model loading failures on systems that rely on them as the primary compute device. The change specifically targets the device selection logic to correctly include iGPUs even when RPC peers are configured. AI
IMPACT Ensures better utilization of integrated GPUs for local model inference within the llama.cpp framework.