@@ -1,5 +1,5 @@
/**
- * llama.cpp - git c574bddb368424b5996cbee2ec45ec050967d404
+ * llama.cpp - git 8183159cf3def112f6d1fe94815fce70e1bffa12
*
* MIT License
@@ -1,7 +1,7 @@
//go:build darwin
//go:build mpi
//go:build opencl
@@ -128,11 +128,6 @@ func New(model string, opts api.Options) (*LLM, error) {
C.llama_backend_init(C.bool(llm.UseNUMA))
- // TODO: GQA == 8 suggests 70B model which doesn't support metal
- if llm.NumGQA == 8 {
- llm.NumGPU = 0
- }
-
params := C.llama_context_default_params()
params.seed = C.uint(llm.Seed)
params.n_ctx = C.int(llm.NumCtx)