highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-03-29 11:11:47 +01:00
Code Issues Packages Projects Releases Wiki Activity

recent llama.cpp update added kernels for fp32, q5_0, and q5_1

Browse Source
This commit is contained in:
Michael Yang 2023-11-20 13:44:12 -08:00
parent 8c4022b06b
commit 19b7a4d715
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" {
switch ggml.FileType() {
case "Q8_0":
case "F32", "Q5_0", "Q5_1", "Q8_0":
if ggml.Name() != "gguf" && opts.NumGPU != 0 {
// GGML 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
}
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