matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 08:17:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #2056 from dhiltgen/slog

Browse Source 
Mechanical switch from log to slog
This commit is contained in:
Daniel Hiltgen
2024-01-18 14:27:24 -08:00
committed by GitHub
parent e5da190bac ecbfc0182f
commit abec7f06e5
14 changed files with 96 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
llm/dyn_ext_server.go
View File

@@ -25,7 +25,7 @@ import (
"context"
"encoding/json"
"fmt"
"log"
"log/slog"
"os"
"path/filepath"
"runtime"
@@ -69,7 +69,7 @@ var llm *dynExtServer
func newDynExtServer(library, model string, adapters, projectors []string, opts api.Options) (LLM, error) {
if !mutex.TryLock() {
log.Printf("concurrent llm servers not yet supported, waiting for prior server to complete")
slog.Info("concurrent llm servers not yet supported, waiting for prior server to complete")
mutex.Lock()
}
updatePath(filepath.Dir(library))
@@ -87,7 +87,7 @@ func newDynExtServer(library, model string, adapters, projectors []string, opts
s: srv,
options: opts,
}
log.Printf("Loading Dynamic llm server: %s", library)
slog.Info(fmt.Sprintf("Loading Dynamic llm server: %s", library))
var sparams C.ext_server_params_t
sparams.model = C.CString(model)
@@ -136,7 +136,7 @@ func newDynExtServer(library, model string, adapters, projectors []string, opts
sparams.n_threads = C.uint(opts.NumThread)
log.Printf("Initializing llama server")
slog.Info("Initializing llama server")
initResp := newExtServerResp(128)
defer freeExtServerResp(initResp)
C.dyn_llama_server_init(llm.s, &sparams, &initResp)
@@ -144,7 +144,7 @@ func newDynExtServer(library, model string, adapters, projectors []string, opts
return nil, extServerResponseToErr(initResp)
}
log.Printf("Starting llama main loop")
slog.Info("Starting llama main loop")
C.dyn_llama_server_start(llm.s)
return llm, nil
}
@@ -158,7 +158,7 @@ func (llm *dynExtServer) Predict(ctx context.Context, predict PredictOpts, fn fu
imageData = append(imageData, ImageData{Data: i, ID: cnt})
}
}
log.Printf("loaded %d images", len(imageData))
slog.Info(fmt.Sprintf("loaded %d images", len(imageData)))
request := map[string]any{
"prompt": predict.Prompt,
@@ -370,7 +370,7 @@ func updatePath(dir string) {
}
}
newPath := strings.Join(append([]string{dir}, pathComponents...), ";")
log.Printf("Updating PATH to %s", newPath)
slog.Info(fmt.Sprintf("Updating PATH to %s", newPath))
os.Setenv("PATH", newPath)
}
// linux and darwin rely on rpath

16
llm/llm.go
View File

@@ -3,7 +3,7 @@ package llm
import (
"context"
"fmt"
"log"
"log/slog"
"os"
"runtime"
@@ -36,7 +36,7 @@ func New(workDir, model string, adapters, projectors []string, opts api.Options)
}
if opts.NumCtx > int(ggml.NumCtx()) {
log.Printf("WARNING: requested context length is greater than model's max context length (%d > %d), using %d instead", opts.NumCtx, ggml.NumCtx(), ggml.NumCtx())
slog.Warn(fmt.Sprintf("requested context length is greater than model's max context length (%d > %d), using %d instead", opts.NumCtx, ggml.NumCtx(), ggml.NumCtx()))
opts.NumCtx = int(ggml.NumCtx())
}
@@ -63,7 +63,7 @@ func New(workDir, model string, adapters, projectors []string, opts api.Options)
}
if size+kv+graph > vram {
log.Println("not enough vram available, falling back to CPU only")
slog.Info("not enough vram available, falling back to CPU only")
info.Library = "cpu"
info.Variant = gpu.GetCPUVariant()
opts.NumGPU = 0
@@ -73,7 +73,7 @@ func New(workDir, model string, adapters, projectors []string, opts api.Options)
opts.NumGPU = 1
default:
if info.Library == "cpu" {
log.Println("GPU not available, falling back to CPU")
slog.Info("GPU not available, falling back to CPU")
opts.NumGPU = 0
break
}
@@ -107,7 +107,7 @@ func New(workDir, model string, adapters, projectors []string, opts api.Options)
// 1 + 2 must fit on the main gpu
min := graph + kv*layers/maxlayers
if layers <= 0 || min > avg {
log.Printf("not enough vram available, falling back to CPU only")
slog.Info("not enough vram available, falling back to CPU only")
info.Library = "cpu"
info.Variant = gpu.GetCPUVariant()
opts.NumGPU = 0
@@ -135,9 +135,9 @@ func newLlmServer(gpuInfo gpu.GpuInfo, model string, adapters, projectors []stri
if demandLib != "" {
libPath := availableDynLibs[demandLib]
if libPath == "" {
log.Printf("Invalid OLLAMA_LLM_LIBRARY %s - not found", demandLib)
slog.Info(fmt.Sprintf("Invalid OLLAMA_LLM_LIBRARY %s - not found", demandLib))
} else {
log.Printf("Loading OLLAMA_LLM_LIBRARY=%s", demandLib)
slog.Info(fmt.Sprintf("Loading OLLAMA_LLM_LIBRARY=%s", demandLib))
dynLibs = []string{libPath}
}
}
@@ -148,7 +148,7 @@ func newLlmServer(gpuInfo gpu.GpuInfo, model string, adapters, projectors []stri
if err == nil {
return srv, nil
}
log.Printf("Failed to load dynamic library %s %s", dynLib, err)
slog.Warn(fmt.Sprintf("Failed to load dynamic library %s %s", dynLib, err))
err2 = err
}

14
llm/payload_common.go
View File

@@ -6,7 +6,7 @@ import (
"fmt"
"io"
"io/fs"
"log"
"log/slog"
"os"
"path/filepath"
"runtime"
@@ -103,13 +103,13 @@ func rocmDynLibPresent() bool {
}
func nativeInit(workdir string) error {
log.Printf("Extracting dynamic libraries...")
slog.Info("Extracting dynamic libraries...")
if runtime.GOOS == "darwin" {
err := extractPayloadFiles(workdir, "llama.cpp/ggml-metal.metal")
if err != nil {
if err == payloadMissing {
// TODO perhaps consider this a hard failure on arm macs?
log.Printf("ggml-meta.metal payload missing")
slog.Info("ggml-meta.metal payload missing")
return nil
}
return err
@@ -120,7 +120,7 @@ func nativeInit(workdir string) error {
libs, err := extractDynamicLibs(workdir, "llama.cpp/build/*/*/*/lib/*")
if err != nil {
if err == payloadMissing {
log.Printf("%s", payloadMissing)
slog.Info(fmt.Sprintf("%s", payloadMissing))
return nil
}
return err
@@ -142,8 +142,8 @@ func nativeInit(workdir string) error {
variants[i] = variant
i++
}
log.Printf("Dynamic LLM libraries %v", variants)
log.Printf("Override detection logic by setting OLLAMA_LLM_LIBRARY")
slog.Info(fmt.Sprintf("Dynamic LLM libraries %v", variants))
slog.Debug("Override detection logic by setting OLLAMA_LLM_LIBRARY")
return nil
}
@@ -163,7 +163,7 @@ func extractDynamicLibs(workDir, glob string) ([]string, error) {
for _, file := range files {
pathComps := strings.Split(file, "/")
if len(pathComps) != pathComponentCount {
log.Printf("unexpected payload components: %v", pathComps)
slog.Error(fmt.Sprintf("unexpected payload components: %v", pathComps))
continue
}