Skip to content

GitLab

Unverified Commit d06bc0cb authored by Bruce MacDonald's avatar Bruce MacDonald Committed by GitHub
Browse files

enable q8, q5, 5_1, and f32 for linux gpu (#699)

parent d104b7e9
Show whitespace changes
Inline Side-by-side
Showing with 17 additions and 14 deletions
llm/llm.go
View file @ d06bc0cb
...@@ -5,6 +5,7 @@ import ( ...@@ -5,6 +5,7 @@ import (
"fmt" "fmt"
"log" "log"
"os" "os"
"runtime"
"github.com/pbnjay/memory" "github.com/pbnjay/memory"
...@@ -37,6 +38,7 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error ...@@ -37,6 +38,7 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
return nil, err return nil, err
} }
if runtime.GOOS == "darwin" {
switch ggml.FileType() { switch ggml.FileType() {
case "Q8_0": case "Q8_0":
if ggml.Name() != "gguf" && opts.NumGPU != 0 { if ggml.Name() != "gguf" && opts.NumGPU != 0 {
...@@ -53,6 +55,7 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error ...@@ -53,6 +55,7 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
opts.NumGPU = 0 opts.NumGPU = 0
} }
} }
}
totalResidentMemory := memory.TotalMemory() totalResidentMemory := memory.TotalMemory()
switch ggml.ModelType() { switch ggml.ModelType() {
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment