baalajimaestro/ollama
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Merge pull request #1212 from jmorganca/mxyng/metal

Browse source 
enable metal for fp32, q5_0, q5_1
This commit is contained in:
Michael Yang 2023-11-20 13:56:39 -08:00 committed by GitHub
commit 0b7ade0d4c
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 1 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
llm/llm.go
View file

@ -41,20 +41,13 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
if runtime.GOOS == "darwin" { if runtime.GOOS == "darwin" {
switch ggml.FileType() { switch ggml.FileType() {
case "Q8_0": case "F32", "Q5_0", "Q5_1", "Q8_0":
if ggml.Name() != "gguf" && opts.NumGPU != 0 { if ggml.Name() != "gguf" && opts.NumGPU != 0 {
// GGML Q8_0 do not support Metal API and will // GGML Q8_0 do not support Metal API and will
// cause the runner to segmentation fault so disable GPU // cause the runner to segmentation fault so disable GPU
log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0") log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
opts.NumGPU = 0 opts.NumGPU = 0
} }
case "F32", "Q5_0", "Q5_1":
if opts.NumGPU != 0 {
// F32, Q5_0, Q5_1, and Q8_0 do not support Metal API and will
// cause the runner to segmentation fault so disable GPU
log.Printf("WARNING: GPU disabled for F32, Q5_0, Q5_1, and Q8_0")
opts.NumGPU = 0
}
} }
var requiredMemory int64 var requiredMemory int64