update llama.cpp submodule to `ceca1ae` (ollama#3064) 

update llama.cpp submodule to `c29af7e` (ollama#2868) 

fix `build_windows.ps1` script to run `go build` with the correct flags 

update llama.cpp submodule to `c14f72d` 

handle race condition while setting raw mode in windows (ollama#2509) 

Merge pull request ollama#2403 from dhiltgen/handle_tmp_cleanup Ensure the libraries are present

fix error on `ollama run` with a non-existent model 

Merge pull request ollama#2197 from dhiltgen/remove_rocm_image Add back ROCm container support

update submodule to `cd4fddb29f81d6a1f6d51a0c016bc6b486d68def` 

revisit memory allocation to account for full kv cache on main gpu