ollama/cmd/cmd.go

1354 lines
31 KiB
Go
Raw Permalink Normal View History

package cmd
import (
"bufio"
"context"
"crypto/ed25519"
"crypto/rand"
"encoding/json"
"encoding/pem"
2023-07-06 15:43:04 -07:00
"errors"
2023-07-06 12:24:49 -04:00
"fmt"
2023-07-18 14:01:19 -07:00
"io"
"log"
"math"
"net"
"net/http"
"os"
"os/signal"
"path/filepath"
"runtime"
"strconv"
2023-07-07 10:12:58 -07:00
"strings"
"sync/atomic"
"syscall"
2023-07-06 15:43:04 -07:00
"time"
"github.com/containerd/console"
"github.com/mattn/go-runewidth"
2023-07-18 09:09:45 -07:00
"github.com/olekukonko/tablewriter"
2023-07-06 13:49:31 -07:00
"github.com/spf13/cobra"
"golang.org/x/crypto/ssh"
"golang.org/x/term"
2023-07-06 13:49:31 -07:00
"github.com/ollama/ollama/api"
"github.com/ollama/ollama/envconfig"
"github.com/ollama/ollama/format"
"github.com/ollama/ollama/parser"
"github.com/ollama/ollama/progress"
"github.com/ollama/ollama/runner"
"github.com/ollama/ollama/server"
"github.com/ollama/ollama/types/model"
"github.com/ollama/ollama/version"
)
var errModelfileNotFound = errors.New("specified Modelfile wasn't found")
func getModelfileName(cmd *cobra.Command) (string, error) {
filename, _ := cmd.Flags().GetString("file")
if filename == "" {
filename = "Modelfile"
}
absName, err := filepath.Abs(filename)
if err != nil {
return "", err
}
_, err = os.Stat(absName)
if err != nil {
return "", err
}
return absName, nil
}
func CreateHandler(cmd *cobra.Command, args []string) error {
2023-11-14 16:33:24 -08:00
p := progress.NewProgress(os.Stderr)
defer p.Stop()
var reader io.Reader
filename, err := getModelfileName(cmd)
if os.IsNotExist(err) {
if filename == "" {
reader = strings.NewReader("FROM .\n")
} else {
return errModelfileNotFound
}
} else if err != nil {
2023-11-14 14:07:40 -08:00
return err
} else {
f, err := os.Open(filename)
if err != nil {
return err
}
reader = f
defer f.Close()
2023-11-14 14:07:40 -08:00
}
modelfile, err := parser.ParseFile(reader)
2023-11-14 14:07:40 -08:00
if err != nil {
return err
}
status := "gathering model components"
spinner := progress.NewSpinner(status)
p.Add(status, spinner)
req, err := modelfile.CreateRequest(filepath.Dir(filename))
2023-11-14 14:07:40 -08:00
if err != nil {
return err
}
spinner.Stop()
2023-11-14 14:07:40 -08:00
req.Name = args[0]
quantize, _ := cmd.Flags().GetString("quantize")
if quantize != "" {
req.Quantize = quantize
}
2023-11-15 16:59:49 -08:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
if len(req.Files) > 0 {
fileMap := map[string]string{}
for f, digest := range req.Files {
if _, err := createBlob(cmd, client, f, digest, p); err != nil {
2023-11-14 14:07:40 -08:00
return err
}
fileMap[filepath.Base(f)] = digest
}
req.Files = fileMap
}
2023-11-14 14:07:40 -08:00
if len(req.Adapters) > 0 {
fileMap := map[string]string{}
for f, digest := range req.Adapters {
if _, err := createBlob(cmd, client, f, digest, p); err != nil {
2023-11-14 14:07:40 -08:00
return err
}
fileMap[filepath.Base(f)] = digest
2023-11-14 14:07:40 -08:00
}
req.Adapters = fileMap
2023-11-14 14:07:40 -08:00
}
2023-07-17 14:14:41 -07:00
2024-04-22 11:02:25 -07:00
bars := make(map[string]*progress.Bar)
fn := func(resp api.ProgressResponse) error {
2023-11-14 16:33:24 -08:00
if resp.Digest != "" {
bar, ok := bars[resp.Digest]
if !ok {
bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
2023-11-14 16:33:24 -08:00
bars[resp.Digest] = bar
p.Add(resp.Digest, bar)
}
bar.Set(resp.Completed)
} else if status != resp.Status {
spinner.Stop()
status = resp.Status
spinner = progress.NewSpinner(status)
p.Add(status, spinner)
}
return nil
}
if err := client.Create(cmd.Context(), req, fn); err != nil {
if strings.Contains(err.Error(), "path or Modelfile are required") {
return fmt.Errorf("the ollama server must be updated to use `ollama create` with this client")
}
return err
}
return nil
}
func createBlob(cmd *cobra.Command, client *api.Client, path string, digest string, p *progress.Progress) (string, error) {
realPath, err := filepath.EvalSymlinks(path)
2024-04-25 14:41:30 -07:00
if err != nil {
return "", err
}
bin, err := os.Open(realPath)
if err != nil {
return "", err
}
defer bin.Close()
// Get file info to retrieve the size
fileInfo, err := bin.Stat()
if err != nil {
return "", err
}
fileSize := fileInfo.Size()
var pw progressWriter
status := fmt.Sprintf("copying file %s 0%%", digest)
spinner := progress.NewSpinner(status)
p.Add(status, spinner)
defer spinner.Stop()
done := make(chan struct{})
defer close(done)
go func() {
ticker := time.NewTicker(60 * time.Millisecond)
defer ticker.Stop()
for {
select {
case <-ticker.C:
spinner.SetMessage(fmt.Sprintf("copying file %s %d%%", digest, int(100*pw.n.Load()/fileSize)))
case <-done:
spinner.SetMessage(fmt.Sprintf("copying file %s 100%%", digest))
return
}
}
}()
if err = client.CreateBlob(cmd.Context(), digest, io.TeeReader(bin, &pw)); err != nil {
return "", err
}
return digest, nil
}
type progressWriter struct {
n atomic.Int64
}
func (w *progressWriter) Write(p []byte) (n int, err error) {
w.n.Add(int64(len(p)))
return len(p), nil
}
2024-09-11 16:36:21 -07:00
func loadOrUnloadModel(cmd *cobra.Command, opts *runOptions) error {
p := progress.NewProgress(os.Stderr)
defer p.StopAndClear()
spinner := progress.NewSpinner("")
p.Add("", spinner)
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
req := &api.GenerateRequest{
Model: opts.Model,
KeepAlive: opts.KeepAlive,
}
return client.Generate(cmd.Context(), req, func(api.GenerateResponse) error { return nil })
}
func StopHandler(cmd *cobra.Command, args []string) error {
opts := &runOptions{
Model: args[0],
KeepAlive: &api.Duration{Duration: 0},
}
if err := loadOrUnloadModel(cmd, opts); err != nil {
if strings.Contains(err.Error(), "not found") {
return fmt.Errorf("couldn't find model \"%s\" to stop", args[0])
}
return err
2024-09-11 16:36:21 -07:00
}
return nil
}
func RunHandler(cmd *cobra.Command, args []string) error {
interactive := true
opts := runOptions{
cmd: defer stating model info until necessary (#5248) This commit changes the 'ollama run' command to defer fetching model information until it really needs it. That is, when in interactive mode. It also removes one such case where the model information is fetch in duplicate, just before calling generateInteractive and then again, first thing, in generateInteractive. This positively impacts the performance of the command: ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.168 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.220 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.217 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 4% cpu 0.652 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 5% cpu 0.498 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 3% cpu 0.479 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total
2024-06-24 20:14:03 -07:00
Model: args[0],
WordWrap: os.Getenv("TERM") == "xterm-256color",
Options: map[string]interface{}{},
}
format, err := cmd.Flags().GetString("format")
if err != nil {
return err
}
opts.Format = format
keepAlive, err := cmd.Flags().GetString("keepalive")
if err != nil {
return err
}
if keepAlive != "" {
d, err := time.ParseDuration(keepAlive)
if err != nil {
return err
}
opts.KeepAlive = &api.Duration{Duration: d}
}
prompts := args[1:]
// prepend stdin to the prompt if provided
if !term.IsTerminal(int(os.Stdin.Fd())) {
in, err := io.ReadAll(os.Stdin)
if err != nil {
return err
}
prompts = append([]string{string(in)}, prompts...)
opts.WordWrap = false
interactive = false
}
opts.Prompt = strings.Join(prompts, " ")
if len(prompts) > 0 {
interactive = false
}
// Be quiet if we're redirecting to a pipe or file
if !term.IsTerminal(int(os.Stdout.Fd())) {
interactive = false
}
nowrap, err := cmd.Flags().GetBool("nowordwrap")
if err != nil {
return err
}
opts.WordWrap = !nowrap
cmd: defer stating model info until necessary (#5248) This commit changes the 'ollama run' command to defer fetching model information until it really needs it. That is, when in interactive mode. It also removes one such case where the model information is fetch in duplicate, just before calling generateInteractive and then again, first thing, in generateInteractive. This positively impacts the performance of the command: ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.168 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.220 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.217 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 4% cpu 0.652 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 5% cpu 0.498 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 3% cpu 0.479 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total
2024-06-24 20:14:03 -07:00
// Fill out the rest of the options based on information about the
// model.
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
name := args[0]
info, err := func() (*api.ShowResponse, error) {
showReq := &api.ShowRequest{Name: name}
info, err := client.Show(cmd.Context(), showReq)
var se api.StatusError
if errors.As(err, &se) && se.StatusCode == http.StatusNotFound {
if err := PullHandler(cmd, []string{name}); err != nil {
return nil, err
}
return client.Show(cmd.Context(), &api.ShowRequest{Name: name})
}
return info, err
}()
if err != nil {
return err
}
if len(info.ProjectorInfo) != 0 {
opts.MultiModal = true
}
for k := range info.ModelInfo {
if strings.Contains(k, ".vision.") {
opts.MultiModal = true
break
}
}
cmd: defer stating model info until necessary (#5248) This commit changes the 'ollama run' command to defer fetching model information until it really needs it. That is, when in interactive mode. It also removes one such case where the model information is fetch in duplicate, just before calling generateInteractive and then again, first thing, in generateInteractive. This positively impacts the performance of the command: ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.168 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.220 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.217 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 4% cpu 0.652 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 5% cpu 0.498 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 3% cpu 0.479 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total
2024-06-24 20:14:03 -07:00
opts.ParentModel = info.Details.ParentModel
if interactive {
2024-09-11 16:36:21 -07:00
if err := loadOrUnloadModel(cmd, &opts); err != nil {
2024-07-26 13:39:38 -07:00
return err
}
for _, msg := range info.Messages {
switch msg.Role {
case "user":
fmt.Printf(">>> %s\n", msg.Content)
case "assistant":
state := &displayResponseState{}
displayResponse(msg.Content, opts.WordWrap, state)
fmt.Println()
fmt.Println()
}
}
cmd: defer stating model info until necessary (#5248) This commit changes the 'ollama run' command to defer fetching model information until it really needs it. That is, when in interactive mode. It also removes one such case where the model information is fetch in duplicate, just before calling generateInteractive and then again, first thing, in generateInteractive. This positively impacts the performance of the command: ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.168 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.220 total ; time ./before run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./before run llama3 'hi' 0.02s user 0.01s system 2% cpu 1.217 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 4% cpu 0.652 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 5% cpu 0.498 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with or would you like to chat? ./after run llama3 'hi' 0.01s user 0.01s system 3% cpu 0.479 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total ; time ./after run llama3 'hi' Hi! It's nice to meet you. Is there something I can help you with, or would you like to chat? ./after run llama3 'hi' 0.02s user 0.01s system 5% cpu 0.507 total
2024-06-24 20:14:03 -07:00
return generateInteractive(cmd, opts)
}
return generate(cmd, opts)
2023-07-06 14:18:40 -04:00
}
2023-07-20 16:09:23 -07:00
func PushHandler(cmd *cobra.Command, args []string) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
insecure, err := cmd.Flags().GetBool("insecure")
if err != nil {
return err
}
2023-11-14 16:33:24 -08:00
p := progress.NewProgress(os.Stderr)
defer p.Stop()
bars := make(map[string]*progress.Bar)
var status string
var spinner *progress.Spinner
2023-11-14 16:33:24 -08:00
2023-07-18 18:51:30 -07:00
fn := func(resp api.ProgressResponse) error {
2023-11-14 16:33:24 -08:00
if resp.Digest != "" {
if spinner != nil {
spinner.Stop()
}
2023-11-14 16:33:24 -08:00
bar, ok := bars[resp.Digest]
if !ok {
bar = progress.NewBar(fmt.Sprintf("pushing %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
2023-11-14 16:33:24 -08:00
bars[resp.Digest] = bar
p.Add(resp.Digest, bar)
}
bar.Set(resp.Completed)
} else if status != resp.Status {
if spinner != nil {
spinner.Stop()
}
2023-11-14 16:33:24 -08:00
status = resp.Status
spinner = progress.NewSpinner(status)
p.Add(status, spinner)
}
return nil
}
2023-11-14 16:33:24 -08:00
request := api.PushRequest{Name: args[0], Insecure: insecure}
n := model.ParseName(args[0])
2023-10-12 15:56:40 -07:00
if err := client.Push(cmd.Context(), &request, fn); err != nil {
if spinner != nil {
spinner.Stop()
}
if strings.Contains(err.Error(), "access denied") {
return errors.New("you are not authorized to push to this namespace, create the model under a namespace you own")
}
2023-11-14 16:33:24 -08:00
return err
}
p.Stop()
2023-11-15 16:59:49 -08:00
spinner.Stop()
destination := n.String()
if strings.HasSuffix(n.Host, ".ollama.ai") || strings.HasSuffix(n.Host, ".ollama.com") {
destination = "https://ollama.com/" + strings.TrimSuffix(n.DisplayShortest(), ":latest")
}
fmt.Printf("\nYou can find your model at:\n\n")
fmt.Printf("\t%s\n", destination)
2023-11-14 16:33:24 -08:00
return nil
}
2023-07-20 16:09:23 -07:00
func ListHandler(cmd *cobra.Command, args []string) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
2023-07-18 09:09:45 -07:00
2023-10-12 15:56:40 -07:00
models, err := client.List(cmd.Context())
2023-07-18 09:09:45 -07:00
if err != nil {
return err
}
var data [][]string
for _, m := range models.Models {
if len(args) == 0 || strings.HasPrefix(strings.ToLower(m.Name), strings.ToLower(args[0])) {
data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), format.HumanTime(m.ModifiedAt, "Never")})
2023-07-18 14:01:19 -07:00
}
2023-07-18 09:09:45 -07:00
}
table := tablewriter.NewWriter(os.Stdout)
2023-08-28 20:50:24 -07:00
table.SetHeader([]string{"NAME", "ID", "SIZE", "MODIFIED"})
2023-07-18 09:09:45 -07:00
table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
table.SetAlignment(tablewriter.ALIGN_LEFT)
table.SetHeaderLine(false)
table.SetBorder(false)
table.SetNoWhiteSpace(true)
table.SetTablePadding(" ")
2023-07-18 09:09:45 -07:00
table.AppendBulk(data)
table.Render()
return nil
}
func ListRunningHandler(cmd *cobra.Command, args []string) error {
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
models, err := client.ListRunning(cmd.Context())
if err != nil {
return err
}
var data [][]string
for _, m := range models.Models {
if len(args) == 0 || strings.HasPrefix(m.Name, args[0]) {
var procStr string
switch {
case m.SizeVRAM == 0:
procStr = "100% CPU"
case m.SizeVRAM == m.Size:
procStr = "100% GPU"
case m.SizeVRAM > m.Size || m.Size == 0:
procStr = "Unknown"
default:
sizeCPU := m.Size - m.SizeVRAM
cpuPercent := math.Round(float64(sizeCPU) / float64(m.Size) * 100)
procStr = fmt.Sprintf("%d%%/%d%% CPU/GPU", int(cpuPercent), int(100-cpuPercent))
}
2024-09-11 16:36:21 -07:00
var until string
delta := time.Since(m.ExpiresAt)
if delta > 0 {
until = "Stopping..."
} else {
until = format.HumanTime(m.ExpiresAt, "Never")
}
data = append(data, []string{m.Name, m.Digest[:12], format.HumanBytes(m.Size), procStr, until})
}
}
table := tablewriter.NewWriter(os.Stdout)
table.SetHeader([]string{"NAME", "ID", "SIZE", "PROCESSOR", "UNTIL"})
table.SetHeaderAlignment(tablewriter.ALIGN_LEFT)
table.SetAlignment(tablewriter.ALIGN_LEFT)
table.SetHeaderLine(false)
table.SetBorder(false)
table.SetNoWhiteSpace(true)
table.SetTablePadding(" ")
table.AppendBulk(data)
table.Render()
return nil
}
2023-07-20 16:09:23 -07:00
func DeleteHandler(cmd *cobra.Command, args []string) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
2023-07-20 16:09:23 -07:00
// Unload the model if it's running before deletion
opts := &runOptions{
Model: args[0],
KeepAlive: &api.Duration{Duration: 0},
}
if err := loadOrUnloadModel(cmd, opts); err != nil {
if !strings.Contains(err.Error(), "not found") {
return fmt.Errorf("unable to stop existing running model \"%s\": %s", args[0], err)
}
}
for _, name := range args {
req := api.DeleteRequest{Name: name}
2023-10-12 15:56:40 -07:00
if err := client.Delete(cmd.Context(), &req); err != nil {
return err
}
fmt.Printf("deleted '%s'\n", name)
2023-07-20 16:09:23 -07:00
}
return nil
}
2023-09-06 11:04:17 -07:00
func ShowHandler(cmd *cobra.Command, args []string) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
2023-09-06 11:04:17 -07:00
if err != nil {
return err
}
license, errLicense := cmd.Flags().GetBool("license")
modelfile, errModelfile := cmd.Flags().GetBool("modelfile")
parameters, errParams := cmd.Flags().GetBool("parameters")
system, errSystem := cmd.Flags().GetBool("system")
template, errTemplate := cmd.Flags().GetBool("template")
for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate} {
if boolErr != nil {
return errors.New("error retrieving flags")
}
}
flagsSet := 0
showType := ""
if license {
flagsSet++
showType = "license"
}
if modelfile {
flagsSet++
showType = "modelfile"
}
if parameters {
flagsSet++
showType = "parameters"
}
if system {
flagsSet++
showType = "system"
}
if template {
flagsSet++
showType = "template"
}
if flagsSet > 1 {
return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
}
req := api.ShowRequest{Name: args[0]}
resp, err := client.Show(cmd.Context(), &req)
if err != nil {
return err
}
if flagsSet == 1 {
switch showType {
case "license":
fmt.Println(resp.License)
case "modelfile":
fmt.Println(resp.Modelfile)
case "parameters":
fmt.Println(resp.Parameters)
case "system":
fmt.Print(resp.System)
case "template":
fmt.Print(resp.Template)
}
return nil
2023-09-06 11:04:17 -07:00
}
return showInfo(resp, os.Stdout)
}
func showInfo(resp *api.ShowResponse, w io.Writer) error {
tableRender := func(header string, rows func() [][]string) {
fmt.Fprintln(w, " ", header)
table := tablewriter.NewWriter(w)
table.SetAlignment(tablewriter.ALIGN_LEFT)
table.SetBorder(false)
table.SetNoWhiteSpace(true)
table.SetTablePadding(" ")
switch header {
case "Template", "System", "License":
table.SetColWidth(100)
}
table.AppendBulk(rows())
table.Render()
fmt.Fprintln(w)
2023-09-06 11:04:17 -07:00
}
tableRender("Model", func() (rows [][]string) {
if resp.ModelInfo != nil {
arch := resp.ModelInfo["general.architecture"].(string)
rows = append(rows, []string{"", "architecture", arch})
rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ModelInfo["general.parameter_count"].(float64)))})
rows = append(rows, []string{"", "context length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)].(float64), 'f', -1, 64)})
rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ModelInfo[fmt.Sprintf("%s.embedding_length", arch)].(float64), 'f', -1, 64)})
} else {
rows = append(rows, []string{"", "architecture", resp.Details.Family})
rows = append(rows, []string{"", "parameters", resp.Details.ParameterSize})
}
rows = append(rows, []string{"", "quantization", resp.Details.QuantizationLevel})
return
})
if resp.ProjectorInfo != nil {
tableRender("Projector", func() (rows [][]string) {
arch := resp.ProjectorInfo["general.architecture"].(string)
rows = append(rows, []string{"", "architecture", arch})
rows = append(rows, []string{"", "parameters", format.HumanNumber(uint64(resp.ProjectorInfo["general.parameter_count"].(float64)))})
rows = append(rows, []string{"", "embedding length", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.embedding_length", arch)].(float64), 'f', -1, 64)})
rows = append(rows, []string{"", "dimensions", strconv.FormatFloat(resp.ProjectorInfo[fmt.Sprintf("%s.vision.projection_dim", arch)].(float64), 'f', -1, 64)})
return
})
}
if resp.Parameters != "" {
tableRender("Parameters", func() (rows [][]string) {
scanner := bufio.NewScanner(strings.NewReader(resp.Parameters))
for scanner.Scan() {
if text := scanner.Text(); text != "" {
rows = append(rows, append([]string{""}, strings.Fields(text)...))
}
}
return
})
}
head := func(s string, n int) (rows [][]string) {
scanner := bufio.NewScanner(strings.NewReader(s))
for scanner.Scan() && (len(rows) < n || n < 0) {
if text := scanner.Text(); text != "" {
rows = append(rows, []string{"", strings.TrimSpace(text)})
}
}
return
}
if resp.System != "" {
tableRender("System", func() [][]string {
return head(resp.System, 2)
})
}
if resp.License != "" {
tableRender("License", func() [][]string {
return head(resp.License, 2)
})
}
return nil
}
2023-07-24 11:27:28 -04:00
func CopyHandler(cmd *cobra.Command, args []string) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
2023-07-24 11:27:28 -04:00
req := api.CopyRequest{Source: args[0], Destination: args[1]}
2023-10-12 15:56:40 -07:00
if err := client.Copy(cmd.Context(), &req); err != nil {
2023-07-24 11:27:28 -04:00
return err
}
fmt.Printf("copied '%s' to '%s'\n", args[0], args[1])
return nil
}
2023-07-20 16:09:23 -07:00
func PullHandler(cmd *cobra.Command, args []string) error {
insecure, err := cmd.Flags().GetBool("insecure")
if err != nil {
return err
}
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
2023-11-14 16:33:24 -08:00
p := progress.NewProgress(os.Stderr)
defer p.Stop()
bars := make(map[string]*progress.Bar)
var status string
var spinner *progress.Spinner
2023-11-14 16:33:24 -08:00
2023-07-18 18:51:30 -07:00
fn := func(resp api.ProgressResponse) error {
2023-11-14 16:33:24 -08:00
if resp.Digest != "" {
if spinner != nil {
spinner.Stop()
}
2023-11-14 16:33:24 -08:00
bar, ok := bars[resp.Digest]
if !ok {
bar = progress.NewBar(fmt.Sprintf("pulling %s...", resp.Digest[7:19]), resp.Total, resp.Completed)
2023-11-14 16:33:24 -08:00
bars[resp.Digest] = bar
p.Add(resp.Digest, bar)
}
bar.Set(resp.Completed)
} else if status != resp.Status {
if spinner != nil {
spinner.Stop()
}
2023-11-14 16:33:24 -08:00
status = resp.Status
spinner = progress.NewSpinner(status)
p.Add(status, spinner)
}
return nil
}
2023-07-07 10:22:37 -04:00
2023-11-14 16:33:24 -08:00
request := api.PullRequest{Name: args[0], Insecure: insecure}
2023-10-12 15:56:40 -07:00
if err := client.Pull(cmd.Context(), &request, fn); err != nil {
2023-11-14 16:33:24 -08:00
return err
}
return nil
2023-07-06 13:49:31 -07:00
}
type generateContextKey string
type runOptions struct {
2024-01-25 12:12:36 -08:00
Model string
ParentModel string
Prompt string
Messages []api.Message
WordWrap bool
Format string
System string
Images []api.ImageData
Options map[string]interface{}
MultiModal bool
KeepAlive *api.Duration
}
type displayResponseState struct {
lineLength int
wordBuffer string
}
func displayResponse(content string, wordWrap bool, state *displayResponseState) {
termWidth, _, _ := term.GetSize(int(os.Stdout.Fd()))
if wordWrap && termWidth >= 10 {
for _, ch := range content {
2024-05-15 17:24:17 -07:00
if state.lineLength+1 > termWidth-5 {
if runewidth.StringWidth(state.wordBuffer) > termWidth-10 {
fmt.Printf("%s%c", state.wordBuffer, ch)
state.wordBuffer = ""
state.lineLength = 0
continue
}
// backtrack the length of the last word and clear to the end of the line
a := runewidth.StringWidth(state.wordBuffer)
if a > 0 {
2024-05-30 10:38:07 -07:00
fmt.Printf("\x1b[%dD", a)
}
fmt.Printf("\x1b[K\n")
fmt.Printf("%s%c", state.wordBuffer, ch)
chWidth := runewidth.RuneWidth(ch)
state.lineLength = runewidth.StringWidth(state.wordBuffer) + chWidth
} else {
fmt.Print(string(ch))
state.lineLength += runewidth.RuneWidth(ch)
if runewidth.RuneWidth(ch) >= 2 {
state.wordBuffer = ""
continue
2024-05-15 17:24:17 -07:00
}
switch ch {
case ' ':
state.wordBuffer = ""
case '\n':
state.lineLength = 0
default:
state.wordBuffer += string(ch)
}
}
}
} else {
fmt.Printf("%s%s", state.wordBuffer, content)
if len(state.wordBuffer) > 0 {
state.wordBuffer = ""
}
}
}
func chat(cmd *cobra.Command, opts runOptions) (*api.Message, error) {
client, err := api.ClientFromEnvironment()
if err != nil {
return nil, err
}
p := progress.NewProgress(os.Stderr)
defer p.StopAndClear()
spinner := progress.NewSpinner("")
p.Add("", spinner)
cancelCtx, cancel := context.WithCancel(cmd.Context())
defer cancel()
sigChan := make(chan os.Signal, 1)
signal.Notify(sigChan, syscall.SIGINT)
go func() {
<-sigChan
cancel()
}()
var state *displayResponseState = &displayResponseState{}
var latest api.ChatResponse
var fullResponse strings.Builder
var role string
fn := func(response api.ChatResponse) error {
p.StopAndClear()
latest = response
role = response.Message.Role
content := response.Message.Content
fullResponse.WriteString(content)
displayResponse(content, opts.WordWrap, state)
return nil
}
2024-12-06 14:13:15 -08:00
if opts.Format == "json" {
opts.Format = `"` + opts.Format + `"`
}
req := &api.ChatRequest{
Model: opts.Model,
Messages: opts.Messages,
Format: json.RawMessage(opts.Format),
Options: opts.Options,
}
if opts.KeepAlive != nil {
req.KeepAlive = opts.KeepAlive
}
if err := client.Chat(cancelCtx, req, fn); err != nil {
if errors.Is(err, context.Canceled) {
return nil, nil
}
return nil, err
}
if len(opts.Messages) > 0 {
fmt.Println()
fmt.Println()
}
verbose, err := cmd.Flags().GetBool("verbose")
if err != nil {
return nil, err
}
if verbose {
latest.Summary()
}
return &api.Message{Role: role, Content: fullResponse.String()}, nil
}
func generate(cmd *cobra.Command, opts runOptions) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
2023-09-18 12:26:56 -07:00
if err != nil {
return err
2023-09-18 12:26:56 -07:00
}
2023-07-07 10:12:58 -07:00
2023-11-14 16:58:51 -08:00
p := progress.NewProgress(os.Stderr)
defer p.StopAndClear()
2023-11-14 16:58:51 -08:00
spinner := progress.NewSpinner("")
p.Add("", spinner)
var latest api.GenerateResponse
generateContext, ok := cmd.Context().Value(generateContextKey("context")).([]int)
if !ok {
generateContext = []int{}
}
2023-10-12 15:56:40 -07:00
ctx, cancel := context.WithCancel(cmd.Context())
defer cancel()
sigChan := make(chan os.Signal, 1)
signal.Notify(sigChan, syscall.SIGINT)
go func() {
<-sigChan
cancel()
}()
var state *displayResponseState = &displayResponseState{}
fn := func(response api.GenerateResponse) error {
2023-11-14 16:58:51 -08:00
p.StopAndClear()
2023-09-18 12:26:56 -07:00
latest = response
content := response.Response
2023-07-12 18:18:06 -07:00
displayResponse(content, opts.WordWrap, state)
2023-09-18 12:26:56 -07:00
return nil
}
2023-07-07 14:04:43 -07:00
if opts.MultiModal {
opts.Prompt, opts.Images, err = extractFileData(opts.Prompt)
if err != nil {
return err
}
}
2024-12-06 14:13:15 -08:00
if opts.Format == "json" {
opts.Format = `"` + opts.Format + `"`
}
2023-10-12 15:56:40 -07:00
request := api.GenerateRequest{
Model: opts.Model,
Prompt: opts.Prompt,
Context: generateContext,
Images: opts.Images,
Format: json.RawMessage(opts.Format),
System: opts.System,
Options: opts.Options,
KeepAlive: opts.KeepAlive,
2023-10-12 15:56:40 -07:00
}
if err := client.Generate(ctx, &request, fn); err != nil {
if errors.Is(err, context.Canceled) {
return nil
2023-07-07 14:04:43 -07:00
}
return err
2023-09-18 12:26:56 -07:00
}
if opts.Prompt != "" {
2023-07-07 10:12:58 -07:00
fmt.Println()
fmt.Println()
2023-09-18 12:26:56 -07:00
}
2023-07-12 18:18:06 -07:00
if !latest.Done {
return nil
}
2023-09-18 12:26:56 -07:00
verbose, err := cmd.Flags().GetBool("verbose")
if err != nil {
return err
2023-09-18 12:26:56 -07:00
}
2023-07-18 11:59:42 -07:00
2023-09-18 12:26:56 -07:00
if verbose {
latest.Summary()
2023-07-07 10:12:58 -07:00
}
2023-07-06 13:49:31 -07:00
ctx = context.WithValue(cmd.Context(), generateContextKey("context"), latest.Context)
cmd.SetContext(ctx)
return nil
2023-07-06 13:49:31 -07:00
}
func RunServer(_ *cobra.Command, _ []string) error {
2023-09-20 17:49:48 -07:00
if err := initializeKeypair(); err != nil {
return err
}
2024-07-03 16:44:57 -07:00
ln, err := net.Listen("tcp", envconfig.Host().Host)
if err != nil {
return err
}
2023-07-04 00:47:00 -04:00
err = server.Serve(ln)
if errors.Is(err, http.ErrServerClosed) {
return nil
}
return err
2023-07-04 00:47:00 -04:00
}
func initializeKeypair() error {
home, err := os.UserHomeDir()
if err != nil {
return err
}
privKeyPath := filepath.Join(home, ".ollama", "id_ed25519")
pubKeyPath := filepath.Join(home, ".ollama", "id_ed25519.pub")
_, err = os.Stat(privKeyPath)
if os.IsNotExist(err) {
fmt.Printf("Couldn't find '%s'. Generating new private key.\n", privKeyPath)
cryptoPublicKey, cryptoPrivateKey, err := ed25519.GenerateKey(rand.Reader)
if err != nil {
return err
}
privateKeyBytes, err := ssh.MarshalPrivateKey(cryptoPrivateKey, "")
if err != nil {
return err
}
if err := os.MkdirAll(filepath.Dir(privKeyPath), 0o755); err != nil {
return fmt.Errorf("could not create directory %w", err)
}
if err := os.WriteFile(privKeyPath, pem.EncodeToMemory(privateKeyBytes), 0o600); err != nil {
return err
}
sshPublicKey, err := ssh.NewPublicKey(cryptoPublicKey)
if err != nil {
return err
}
publicKeyBytes := ssh.MarshalAuthorizedKey(sshPublicKey)
if err := os.WriteFile(pubKeyPath, publicKeyBytes, 0o644); err != nil {
return err
}
fmt.Printf("Your new public key is: \n\n%s\n", publicKeyBytes)
}
return nil
}
2023-10-12 15:56:40 -07:00
func checkServerHeartbeat(cmd *cobra.Command, _ []string) error {
2023-10-09 12:18:26 -07:00
client, err := api.ClientFromEnvironment()
if err != nil {
return err
}
2023-10-12 15:56:40 -07:00
if err := client.Heartbeat(cmd.Context()); err != nil {
if !strings.Contains(err.Error(), " refused") {
2023-07-31 17:38:10 -04:00
return err
}
if err := startApp(cmd.Context(), client); err != nil {
2024-08-01 14:52:15 -07:00
return errors.New("could not connect to ollama app, is it running?")
}
}
return nil
}
2023-11-22 09:41:02 -08:00
func versionHandler(cmd *cobra.Command, _ []string) {
client, err := api.ClientFromEnvironment()
if err != nil {
return
}
serverVersion, err := client.Version(cmd.Context())
if err != nil {
2023-12-01 12:10:27 -08:00
fmt.Println("Warning: could not connect to a running Ollama instance")
}
if serverVersion != "" {
fmt.Printf("ollama version is %s\n", serverVersion)
2023-11-22 09:41:02 -08:00
}
if serverVersion != version.Version {
2023-12-01 12:10:27 -08:00
fmt.Printf("Warning: client version is %s\n", version.Version)
}
2023-11-22 09:41:02 -08:00
}
func appendEnvDocs(cmd *cobra.Command, envs []envconfig.EnvVar) {
if len(envs) == 0 {
return
}
envUsage := `
Environment Variables:
`
for _, e := range envs {
envUsage += fmt.Sprintf(" %-24s %s\n", e.Name, e.Description)
}
cmd.SetUsageTemplate(cmd.UsageTemplate() + envUsage)
}
func NewCLI() *cobra.Command {
log.SetFlags(log.LstdFlags | log.Lshortfile)
2023-11-22 09:41:02 -08:00
cobra.EnableCommandSorting = false
if runtime.GOOS == "windows" && term.IsTerminal(int(os.Stdout.Fd())) {
console.ConsoleFromFile(os.Stdin) //nolint:errcheck
}
rootCmd := &cobra.Command{
2023-08-14 11:15:53 -07:00
Use: "ollama",
Short: "Large language model runner",
SilenceUsage: true,
SilenceErrors: true,
CompletionOptions: cobra.CompletionOptions{
DisableDefaultCmd: true,
},
2023-11-22 09:41:02 -08:00
Run: func(cmd *cobra.Command, args []string) {
if version, _ := cmd.Flags().GetBool("version"); version {
versionHandler(cmd, args)
return
}
cmd.Print(cmd.UsageString())
},
}
2023-11-22 09:41:02 -08:00
rootCmd.Flags().BoolP("version", "v", false, "Show version information")
createCmd := &cobra.Command{
Use: "create MODEL",
Short: "Create a model from a Modelfile",
2023-10-18 11:57:22 -07:00
Args: cobra.ExactArgs(1),
PreRunE: checkServerHeartbeat,
RunE: CreateHandler,
}
createCmd.Flags().StringP("file", "f", "", "Name of the Modelfile (default \"Modelfile\"")
createCmd.Flags().StringP("quantize", "q", "", "Quantize model to this level (e.g. q4_0)")
2023-09-06 11:04:17 -07:00
showCmd := &cobra.Command{
Use: "show MODEL",
Short: "Show information for a model",
2023-10-18 11:57:22 -07:00
Args: cobra.ExactArgs(1),
2023-09-06 11:04:17 -07:00
PreRunE: checkServerHeartbeat,
RunE: ShowHandler,
}
showCmd.Flags().Bool("license", false, "Show license of a model")
showCmd.Flags().Bool("modelfile", false, "Show Modelfile of a model")
showCmd.Flags().Bool("parameters", false, "Show parameters of a model")
showCmd.Flags().Bool("template", false, "Show template of a model")
showCmd.Flags().Bool("system", false, "Show system message of a model")
2023-09-06 11:04:17 -07:00
runCmd := &cobra.Command{
Use: "run MODEL [PROMPT]",
Short: "Run a model",
Args: cobra.MinimumNArgs(1),
PreRunE: checkServerHeartbeat,
RunE: RunHandler,
}
runCmd.Flags().String("keepalive", "", "Duration to keep a model loaded (e.g. 5m)")
2023-07-12 18:18:06 -07:00
runCmd.Flags().Bool("verbose", false, "Show timings for response")
2023-08-21 21:56:56 -07:00
runCmd.Flags().Bool("insecure", false, "Use an insecure registry")
runCmd.Flags().Bool("nowordwrap", false, "Don't wrap words to the next line automatically")
2023-11-13 21:54:02 -05:00
runCmd.Flags().String("format", "", "Response format (e.g. json)")
2024-09-11 16:36:21 -07:00
stopCmd := &cobra.Command{
Use: "stop MODEL",
Short: "Stop a running model",
Args: cobra.ExactArgs(1),
PreRunE: checkServerHeartbeat,
RunE: StopHandler,
}
serveCmd := &cobra.Command{
Use: "serve",
Aliases: []string{"start"},
Short: "Start ollama",
2023-10-18 11:57:22 -07:00
Args: cobra.ExactArgs(0),
2023-07-06 13:49:31 -07:00
RunE: RunServer,
}
pullCmd := &cobra.Command{
Use: "pull MODEL",
Short: "Pull a model from a registry",
2023-10-18 11:57:22 -07:00
Args: cobra.ExactArgs(1),
PreRunE: checkServerHeartbeat,
RunE: PullHandler,
}
pullCmd.Flags().Bool("insecure", false, "Use an insecure registry")
pushCmd := &cobra.Command{
Use: "push MODEL",
Short: "Push a model to a registry",
2023-10-18 11:57:22 -07:00
Args: cobra.ExactArgs(1),
PreRunE: checkServerHeartbeat,
RunE: PushHandler,
}
pushCmd.Flags().Bool("insecure", false, "Use an insecure registry")
2023-07-18 09:09:45 -07:00
listCmd := &cobra.Command{
Use: "list",
2023-07-20 15:28:27 -07:00
Aliases: []string{"ls"},
Short: "List models",
PreRunE: checkServerHeartbeat,
RunE: ListHandler,
2023-07-20 16:09:23 -07:00
}
psCmd := &cobra.Command{
Use: "ps",
Short: "List running models",
PreRunE: checkServerHeartbeat,
RunE: ListRunningHandler,
}
2023-07-24 11:27:28 -04:00
copyCmd := &cobra.Command{
2024-05-01 12:39:05 -07:00
Use: "cp SOURCE DESTINATION",
Short: "Copy a model",
2023-10-18 11:57:22 -07:00
Args: cobra.ExactArgs(2),
PreRunE: checkServerHeartbeat,
RunE: CopyHandler,
2023-07-24 11:27:28 -04:00
}
2023-07-20 16:09:23 -07:00
deleteCmd := &cobra.Command{
2023-10-18 11:57:22 -07:00
Use: "rm MODEL [MODEL...]",
Short: "Remove a model",
Args: cobra.MinimumNArgs(1),
PreRunE: checkServerHeartbeat,
RunE: DeleteHandler,
2023-07-18 09:09:45 -07:00
}
build: Make target improvements (#7499) * llama: wire up builtin runner This adds a new entrypoint into the ollama CLI to run the cgo built runner. On Mac arm64, this will have GPU support, but on all other platforms it will be the lowest common denominator CPU build. After we fully transition to the new Go runners more tech-debt can be removed and we can stop building the "default" runner via make and rely on the builtin always. * build: Make target improvements Add a few new targets and help for building locally. This also adjusts the runner lookup to favor local builds, then runners relative to the executable, and finally payloads. * Support customized CPU flags for runners This implements a simplified custom CPU flags pattern for the runners. When built without overrides, the runner name contains the vector flag we check for (AVX) to ensure we don't try to run on unsupported systems and crash. If the user builds a customized set, we omit the naming scheme and don't check for compatibility. This avoids checking requirements at runtime, so that logic has been removed as well. This can be used to build GPU runners with no vector flags, or CPU/GPU runners with additional flags (e.g. AVX512) enabled. * Use relative paths If the user checks out the repo in a path that contains spaces, make gets really confused so use relative paths for everything in-repo to avoid breakage. * Remove payloads from main binary * install: clean up prior libraries This removes support for v0.3.6 and older versions (before the tar bundle) and ensures we clean up prior libraries before extracting the bundle(s). Without this change, runners and dependent libraries could leak when we update and lead to subtle runtime errors.
2024-12-10 09:47:19 -08:00
runnerCmd := &cobra.Command{
Use: "runner",
Hidden: true,
RunE: func(cmd *cobra.Command, args []string) error {
return runner.Execute(os.Args[1:])
},
FParseErrWhitelist: cobra.FParseErrWhitelist{UnknownFlags: true},
}
runnerCmd.SetHelpFunc(func(cmd *cobra.Command, args []string) {
_ = runner.Execute(args[1:])
})
envVars := envconfig.AsMap()
envs := []envconfig.EnvVar{envVars["OLLAMA_HOST"]}
for _, cmd := range []*cobra.Command{
createCmd,
showCmd,
runCmd,
2024-09-11 16:36:21 -07:00
stopCmd,
pullCmd,
pushCmd,
listCmd,
psCmd,
copyCmd,
deleteCmd,
serveCmd,
} {
switch cmd {
case runCmd:
appendEnvDocs(cmd, []envconfig.EnvVar{envVars["OLLAMA_HOST"], envVars["OLLAMA_NOHISTORY"]})
case serveCmd:
appendEnvDocs(cmd, []envconfig.EnvVar{
envVars["OLLAMA_DEBUG"],
envVars["OLLAMA_HOST"],
envVars["OLLAMA_KEEP_ALIVE"],
envVars["OLLAMA_MAX_LOADED_MODELS"],
envVars["OLLAMA_MAX_QUEUE"],
envVars["OLLAMA_MODELS"],
envVars["OLLAMA_NUM_PARALLEL"],
envVars["OLLAMA_NOPRUNE"],
envVars["OLLAMA_ORIGINS"],
2024-07-23 15:14:28 -07:00
envVars["OLLAMA_SCHED_SPREAD"],
envVars["OLLAMA_TMPDIR"],
envVars["OLLAMA_FLASH_ATTENTION"],
envVars["OLLAMA_KV_CACHE_TYPE"],
envVars["OLLAMA_LLM_LIBRARY"],
envVars["OLLAMA_GPU_OVERHEAD"],
envVars["OLLAMA_LOAD_TIMEOUT"],
})
default:
appendEnvDocs(cmd, envs)
}
}
rootCmd.AddCommand(
serveCmd,
createCmd,
2023-09-06 11:04:17 -07:00
showCmd,
runCmd,
2024-09-11 16:36:21 -07:00
stopCmd,
pullCmd,
pushCmd,
2023-07-18 09:09:45 -07:00
listCmd,
psCmd,
2023-07-24 11:27:28 -04:00
copyCmd,
2023-07-20 16:09:23 -07:00
deleteCmd,
build: Make target improvements (#7499) * llama: wire up builtin runner This adds a new entrypoint into the ollama CLI to run the cgo built runner. On Mac arm64, this will have GPU support, but on all other platforms it will be the lowest common denominator CPU build. After we fully transition to the new Go runners more tech-debt can be removed and we can stop building the "default" runner via make and rely on the builtin always. * build: Make target improvements Add a few new targets and help for building locally. This also adjusts the runner lookup to favor local builds, then runners relative to the executable, and finally payloads. * Support customized CPU flags for runners This implements a simplified custom CPU flags pattern for the runners. When built without overrides, the runner name contains the vector flag we check for (AVX) to ensure we don't try to run on unsupported systems and crash. If the user builds a customized set, we omit the naming scheme and don't check for compatibility. This avoids checking requirements at runtime, so that logic has been removed as well. This can be used to build GPU runners with no vector flags, or CPU/GPU runners with additional flags (e.g. AVX512) enabled. * Use relative paths If the user checks out the repo in a path that contains spaces, make gets really confused so use relative paths for everything in-repo to avoid breakage. * Remove payloads from main binary * install: clean up prior libraries This removes support for v0.3.6 and older versions (before the tar bundle) and ensures we clean up prior libraries before extracting the bundle(s). Without this change, runners and dependent libraries could leak when we update and lead to subtle runtime errors.
2024-12-10 09:47:19 -08:00
runnerCmd,
)
return rootCmd
}