update llama.cpp

This commit is contained in:
Michael Yang
2023-08-03 11:49:58 -07:00
parent f0b365a478
commit c5bcf32823
19 changed files with 623 additions and 298 deletions

View File

@@ -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)