feat(conf): retrieve llama-server flags from shell scripts

Author: 0uep

conf/conf.go

@@ -32,6 +32,7 @@ type (
 		ModelsDir    string                `toml:"models_dir"     yaml:"models_dir"     comment:"\nGoinfer recursively searches GGUF files in one or multiple folders separated by ':'\nList your GGUF dirs with: locate .gguf | sed -e 's,/[^/]*$,,' | uniq\nenv. var: GI_MODELS_DIR"`
 		DefaultModel string                `toml:"default_model"  yaml:"default_model"  comment:"\nThe default model name to load at startup\nCan also be set with: ./goinfer -start <model-name>"`
 		Addr         string                `toml:"addr"           yaml:"addr"           comment:"address can be 'host:port' or 'ip:por' or simply ':port' (for host = localhost)"`
+		Shells       []*ModelInfo          `toml:"-"              yaml:"-"`
 	}
 
 	// Llama holds the inference engine settings.

conf/files.go

@@ -199,6 +199,113 @@ func nameWithDir(root, truncated, name string) string {
 	return grp + "/" + name
 }
 
+// extractModelNameAndFlags search for a llama-server command line
+// and extract the model path (flag -m or --model) and the flags after -m|--model.
+func extractModelNameAndFlags(fsys fs.FS, shellPath string) (modelPath, flags []byte) {
+	//	shellPath = filepath.Clean(shellPath)
+	script, err := fs.ReadFile(fsys, shellPath)
+	if err != nil {
+		return nil, nil
+	}
+
+	script = searchLlamaServer(script)
+	if script == nil {
+		return nil, nil
+	}
+
+	script = searchModelFlag(script)
+	if script == nil {
+		return nil, nil
+	}
+
+	// trim any leading whitespace character
+	script = bytes.TrimSpace(script)
+	pos := bytes.IndexAny(script, " \t")
+	if pos < 0 {
+		modelPath = script
+	} else {
+		modelPath = script[:pos]
+		flags = oneLine(script[pos+1:])
+	}
+
+	slog.Info("Found", "from file", shellPath, "flags", flags)
+
+	return modelPath, flags
+}
+
+// searchLlamaServer searches for a llama-server command line.
+func searchLlamaServer(script []byte) []byte {
+	for {
+		var before []byte
+		var found bool
+		before, script, found = bytes.Cut(script, []byte("/llama-server"))
+		if !found {
+			return nil
+		}
+
+		// check space after "llama-server"
+		if len(script) == 0 {
+			return nil
+		}
+
+		if script[0] != ' ' && script[0] != '\t' {
+			continue
+		}
+
+		// rewind to the beginning of the line
+		pos := bytes.LastIndexByte(before, '\n')
+		if pos >= 0 {
+			before = before[pos+1:]
+		}
+
+		// skip commented lines except shebang (#!)
+		if len(before) == 0 || before[0] != '#' || (len(before) > 1 && before[1] == '!') {
+			return script[1:]
+		}
+	}
+}
+
+// searchModelFlag searches for the flag: --model or -m.
+func searchModelFlag(script []byte) []byte {
+	for {
+		before, after, found := bytes.Cut(script, []byte("--model"))
+		if !found {
+			before, after, found = bytes.Cut(script, []byte("-m"))
+			if !found {
+				return nil
+			}
+		}
+		script = after
+
+		if len(script) == 0 {
+			return nil
+		}
+
+		if script[0] != ' ' && script[0] != '\t' {
+			continue
+		}
+
+		if len(before) == 0 {
+			return script
+		}
+
+		if before[len(before)-1] != ' ' && before[len(before)-1] != '\t' {
+			continue
+		}
+
+		// rewind to the beginning of the line
+		pos := bytes.LastIndexByte(before, '\n')
+		if pos >= 0 {
+			before = before[pos+1:]
+		}
+
+		// skip commented lines
+		if before[0] != '#' {
+			return script[1:]
+		}
+	}
+}
+
 // extractFlags returns the truncated path and the llama-server flags from a file path.
 // It first checks for a companion ".sh" file; if present, its contents are used as flags.
 // Otherwise, it parses flags encoded in the filename after an '&' delimiter.
@@ -255,7 +362,7 @@ func oneLine(input []byte) []byte {
 			continue // Skip empty lines
 		}
 
-		if line[len(line)-1] == '\'' {
+		if line[len(line)-1] == '\\' {
 			line = line[:len(line)-1] // Remove trailing backslash
 		}
 
@@ -265,8 +372,10 @@ func oneLine(input []byte) []byte {
 			continue
 		}
 
+		if len(keep) > 0 {
+			keep = append(keep, ' ')
+		}
 		keep = append(keep, line...)
-		keep = append(keep, ' ')
 	}
 
 	return keep

conf/files_test.go

@@ -9,6 +9,7 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"testing/fstest"
 
 	"github.com/LM4eu/goinfer/proxy/config"
 )
@@ -242,3 +243,61 @@ func Test_nameWithGGUF(t *testing.T) {
 		})
 	}
 }
+
+const (
+	script1 = "/path/to/llama-server --host 0.0.0.0 --port 5800 --verbose-prompt --no-warmup " +
+		"	-m /path/model.gguf " +
+		`	--no-mmap --chat-template-kwargs '{"reasoning_effort": "high"}' ` +
+		"	--reasoning-format auto -c 10240 " + " #	--no-context-shift"
+
+	script2 = `#!/bin/sh
+/path/to/llama-server --host 0.0.0.0 --port 5800 --verbose-prompt \
+	--no-warmup --model /path/model.gguf --no-mmap \
+	--chat-template-kwargs '{"reasoning_effort": "high"}' \
+	--reasoning-format auto -c 10240 \
+#	--no-context-shift
+`
+	script3 = `#!/path/to/llama-server --host 0.0.0.0 --port 5800 \
+	--verbose-prompt --no-warmup -m /path/model.gguf \
+	--no-mmap --chat-template-kwargs '{"reasoning_effort": "high"}' \
+	--reasoning-format auto -c 10240 \
+#	--no-context-shift
+`
+)
+
+func Test_extractModelNameAndFlags(t *testing.T) {
+	t.Parallel()
+
+	testFS := fstest.MapFS{
+		"0.sh": &fstest.MapFile{Data: []byte("")},
+		"1.sh": &fstest.MapFile{Data: []byte(script1)},
+		"2.sh": &fstest.MapFile{Data: []byte(script2)},
+		"3.sh": &fstest.MapFile{Data: []byte(script3)},
+		"4.sh": &fstest.MapFile{Data: []byte("/path/to/llama-server --models-preset ~/bin/goinfer/models.ini")},
+		"5.sh": &fstest.MapFile{Data: []byte("llama-server --model ~/model.gguf")},
+		"6.sh": &fstest.MapFile{Data: []byte("/llama-server --model ~/model.gguf")},
+		"7.sh": &fstest.MapFile{Data: []byte("/llama-server\t-m\t~/model.gguf\t-c\t0")},
+	}
+
+	tests := []struct{ path, wantModel, wantFlags string }{
+		{"1.sh", "/path/model.gguf", `--no-mmap --chat-template-kwargs '{"reasoning_effort": "high"}' 	--reasoning-format auto -c 10240  #	--no-context-shift`},
+		{"2.sh", "/path/model.gguf", `--no-mmap --chat-template-kwargs '{"reasoning_effort": "high"}' --reasoning-format auto -c 10240`},
+		{"3.sh", "/path/model.gguf", `--no-mmap --chat-template-kwargs '{"reasoning_effort": "high"}' --reasoning-format auto -c 10240`},
+		{"4.sh", "", ""},
+		{"5.sh", "", ""},
+		{"6.sh", "~/model.gguf", ""},
+		{"7.sh", "~/model.gguf", "-c\t0"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.path, func(t *testing.T) {
+			t.Parallel()
+			gotModel, gotFlags := extractModelNameAndFlags(testFS, tt.path)
+			if string(gotModel) != tt.wantModel {
+				t.Errorf("extractModelNameAndFlags(%s) = %s, want %s", tt.path, gotModel, tt.wantModel)
+			}
+			if string(gotFlags) != tt.wantFlags {
+				t.Errorf("extractModelNameAndFlags(%s) = %s, want %s", tt.path, gotFlags, tt.wantFlags)
+			}
+		})
+	}
+}

conf/info.go

@@ -159,6 +159,7 @@ func (cfg *Cfg) updateInfo() {
 // cfg.Info. It validates each file using validateFile and warns about errors (logs).
 func (cfg *Cfg) search(params map[string]ModelParams, root string) error {
 	return filepath.WalkDir(root, func(path string, dir fs.DirEntry, err error) error {
+		fsys := os.DirFS(root)
 		switch {
 		case err != nil:
 			if dir == nil {
@@ -172,8 +173,10 @@ func (cfg *Cfg) search(params map[string]ModelParams, root string) error {
 			if err != nil {
 				slog.Warn("skip params file", "path", path, "err", err)
 			}
-		case strings.HasSuffix(path, ".gguf"):
+		case filepath.Ext(path) == ".gguf":
 			cfg.keepGUFF(root, path)
+		case filepath.Ext(path) == ".sh":
+			cfg.keepFlags(fsys, path[len(root):])
 		default:
 		}
 		return nil
@@ -231,3 +234,28 @@ func (cfg *Cfg) keepGUFF(root, path string) {
 	}
 	cfg.Info[name] = &mi
 }
+
+func (cfg *Cfg) keepFlags(fsys fs.FS, path string) {
+	modelPath, args := extractModelNameAndFlags(fsys, path)
+	if modelPath == nil {
+		return
+	}
+
+	flags := replaceDIR(path, string(args))
+
+	mi := ModelInfo{Params: nil, Flags: flags, Path: string(modelPath), Origin: path}
+	if cfg.Shells == nil {
+		slog.Debug("Found first", "shell", path)
+		cfg.Shells = []*ModelInfo{&mi}
+		return
+	}
+
+	for _, mm := range cfg.Shells {
+		if mm.Origin == mi.Origin {
+			slog.Debug("Already present", "shell", path)
+			return
+		}
+	}
+
+	cfg.Shells = append(cfg.Shells, &mi)
+}