Merge pull request #1 from jmorganca/main

ok

cmd/cmd.go

@@ -11,6 +11,7 @@ import (
 	"io"
 	"log"
 	"net"
+	"net/http"
 	"os"
 	"os/exec"
 	"os/signal"
@@ -98,21 +99,20 @@ func RunHandler(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
-	models, err := client.List(context.Background())
+	name := args[0]
+	// check if the model exists on the server
+	_, err = client.Show(context.Background(), &api.ShowRequest{Name: name})
 	if err != nil {
-		return err
-	}
-
-	canonicalModelPath := server.ParseModelPath(args[0])
-	for _, model := range models.Models {
-		if model.Name == canonicalModelPath.GetShortTagname() {
-			return RunGenerate(cmd, args)
-		}
-	}
-
+		var statusError api.StatusError
+		switch {
+		case errors.As(err, &statusError) && statusError.StatusCode == http.StatusNotFound:
 			if err := PullHandler(cmd, args); err != nil {
 				return err
 			}
+		case err != nil:
+			return err
+		}
+	}
 
 	return RunGenerate(cmd, args)
 }
@@ -731,21 +731,6 @@ func RunServer(cmd *cobra.Command, _ []string) error {
 		origins = strings.Split(o, ",")
 	}
 
-	if noprune := os.Getenv("OLLAMA_NOPRUNE"); noprune == "" {
-		if err := server.PruneLayers(); err != nil {
-			return err
-		}
-
-		manifestsPath, err := server.GetManifestPath()
-		if err != nil {
-			return err
-		}
-
-		if err := server.PruneDirectory(manifestsPath); err != nil {
-			return err
-		}
-	}
-
 	return server.Serve(ln, origins)
 }
 

docs/import.md

@@ -185,7 +185,7 @@ python convert.py <path to model directory>
 python convert-falcon-hf-to-gguf.py <path to model directory>
 
 # GPTNeoXForCausalLM
-python convert-falcon-hf-to-gguf.py <path to model directory>
+python convert-gptneox-hf-to-gguf.py <path to model directory>
 
 # GPTBigCodeForCausalLM
 python convert-starcoder-hf-to-gguf.py <path to model directory>

readline/buffer.go

@@ -2,6 +2,7 @@ package readline
 
 import (
 	"fmt"
+	"os"
 
 	"github.com/emirpasic/gods/lists/arraylist"
 	"golang.org/x/term"
@@ -17,7 +18,8 @@ type Buffer struct {
 }
 
 func NewBuffer(prompt *Prompt) (*Buffer, error) {
-	width, height, err := term.GetSize(0)
+	fd := int(os.Stdout.Fd())
+	width, height, err := term.GetSize(fd)
 	if err != nil {
 		fmt.Println("Error getting size:", err)
 		return nil, err

readline/readline.go

@@ -51,11 +51,12 @@ func (i *Instance) Readline() (string, error) {
 	}
 	fmt.Print(prompt)
 
-	termios, err := SetRawMode(syscall.Stdin)
+	fd := int(syscall.Stdin)
+	termios, err := SetRawMode(fd)
 	if err != nil {
 		return "", err
 	}
-	defer UnsetRawMode(syscall.Stdin, termios)
+	defer UnsetRawMode(fd, termios)
 
 	buf, _ := NewBuffer(i.Prompt)
 

readline/term_windows.go

@@ -0,0 +1,62 @@
+package readline
+
+import (
+	"syscall"
+	"unsafe"
+)
+
+const (
+	enableLineInput       = 2
+	enableWindowInput     = 8
+	enableMouseInput      = 16
+	enableInsertMode      = 32
+	enableQuickEditMode   = 64
+	enableExtendedFlags   = 128
+	enableProcessedOutput = 1
+	enableWrapAtEolOutput = 2
+	enableAutoPosition    = 256 // Cursor position is not affected by writing data to the console.
+	enableEchoInput       = 4   // Characters are written to the console as they're read.
+	enableProcessedInput  = 1   // Enables input processing (like recognizing Ctrl+C).
+)
+
+var kernel32 = syscall.NewLazyDLL("kernel32.dll")
+
+var (
+	procGetConsoleMode = kernel32.NewProc("GetConsoleMode")
+	procSetConsoleMode = kernel32.NewProc("SetConsoleMode")
+)
+
+type State struct {
+	mode uint32
+}
+
+// IsTerminal checks if the given file descriptor is associated with a terminal
+func IsTerminal(fd int) bool {
+	var st uint32
+	r, _, e := syscall.SyscallN(procGetConsoleMode.Addr(), uintptr(fd), uintptr(unsafe.Pointer(&st)), 0)
+	// if the call succeeds and doesn't produce an error, it's a terminal
+	return r != 0 && e == 0
+}
+
+func SetRawMode(fd int) (*State, error) {
+	var st uint32
+	// retrieve the current mode of the terminal
+	_, _, e := syscall.SyscallN(procGetConsoleMode.Addr(), uintptr(fd), uintptr(unsafe.Pointer(&st)), 0)
+	if e != 0 {
+		return nil, error(e)
+	}
+	// modify the mode to set it to raw
+	raw := st &^ (enableEchoInput | enableProcessedInput | enableLineInput | enableProcessedOutput)
+	// apply the new mode to the terminal
+	_, _, e = syscall.SyscallN(procSetConsoleMode.Addr(), uintptr(fd), uintptr(raw), 0)
+	if e != 0 {
+		return nil, error(e)
+	}
+	// return the original state so that it can be restored later
+	return &State{st}, nil
+}
+
+func UnsetRawMode(fd int, state *State) error {
+	_, _, err := syscall.SyscallN(procSetConsoleMode.Addr(), uintptr(fd), uintptr(state.mode), 0)
+	return err
+}

server/routes.go

@@ -614,6 +614,22 @@ var defaultAllowOrigins = []string{
 }
 
 func Serve(ln net.Listener, allowOrigins []string) error {
+	if noprune := os.Getenv("OLLAMA_NOPRUNE"); noprune == "" {
+		// clean up unused layers and manifests
+		if err := PruneLayers(); err != nil {
+			return err
+		}
+
+		manifestsPath, err := GetManifestPath()
+		if err != nil {
+			return err
+		}
+
+		if err := PruneDirectory(manifestsPath); err != nil {
+			return err
+		}
+	}
+
 	config := cors.DefaultConfig()
 	config.AllowWildcard = true