Load model dry-run #3

Merged

nicoboss merged 4 commits intomradermacherfrom

load-model-dry-run

Mar 9, 2025

Commits on Mar 1, 2025

Implemented a way for llama.cpp to go through all the steps of loading a model without loading a model in order to validate the model and compute the memory required to load the model
nicoboss
committed

Commits on Mar 9, 2025

Introduced DRYRUN environment variable to toggle the dry-run functionality
nicoboss
committed
Implemented dry-run support for ggml_backend_cuda_host_buffer_type_alloc_buffer to allow testing memory requirements of situations where some or all layers are in RAM but are processed using the GP…
nicoboss
committed
Fixed indention consistency by always using spaces instead of taps to follow the llama.cpp code styling guidelines
nicoboss
committed