mirror of
https://github.com/ggerganov/whisper.cpp.git
synced 2023-11-04 02:52:44 +03:00
bindings : initial import of golang bindings (#287)
* Initial import of golang bindings * Updated makefile rules * Updated bindings * Makefile update to add in more tests
This commit is contained in:
30
bindings/go/examples/go-model-download/context.go
Executable file
30
bindings/go/examples/go-model-download/context.go
Executable file
@@ -0,0 +1,30 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"context"
|
||||
"os"
|
||||
"os/signal"
|
||||
)
|
||||
|
||||
// ContextForSignal returns a context object which is cancelled when a signal
|
||||
// is received. It returns nil if no signal parameter is provided
|
||||
func ContextForSignal(signals ...os.Signal) context.Context {
|
||||
if len(signals) == 0 {
|
||||
return nil
|
||||
}
|
||||
|
||||
ch := make(chan os.Signal)
|
||||
ctx, cancel := context.WithCancel(context.Background())
|
||||
|
||||
// Send message on channel when signal received
|
||||
signal.Notify(ch, signals...)
|
||||
|
||||
// When any signal received, call cancel
|
||||
go func() {
|
||||
<-ch
|
||||
cancel()
|
||||
}()
|
||||
|
||||
// Return success
|
||||
return ctx
|
||||
}
|
||||
206
bindings/go/examples/go-model-download/main.go
Executable file
206
bindings/go/examples/go-model-download/main.go
Executable file
@@ -0,0 +1,206 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"context"
|
||||
"flag"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"net/url"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"syscall"
|
||||
"time"
|
||||
)
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// CONSTANTS
|
||||
|
||||
const (
|
||||
srcUrl = "https://huggingface.co/" // The location of the models
|
||||
srcPathPrefix = "/datasets/ggerganov/whisper.cpp/resolve/main/ggml" // Filename prefix
|
||||
srcExt = ".bin" // Filename extension
|
||||
bufSize = 1024 * 64 // Size of the buffer used for downloading the model
|
||||
)
|
||||
|
||||
var (
|
||||
// The models which will be downloaded, if no model is specified as an argument
|
||||
modelNames = []string{"tiny.en", "tiny", "base.en", "base", "small.en", "small", "medium.en", "medium", "large-v1", "large"}
|
||||
)
|
||||
|
||||
var (
|
||||
// The output folder. When not set, use current working directory.
|
||||
flagOut = flag.String("out", "", "Output folder")
|
||||
|
||||
// HTTP timeout parameter - will timeout if takes longer than this to download a model
|
||||
flagTimeout = flag.Duration("timeout", 30*time.Minute, "HTTP timeout")
|
||||
|
||||
// Quiet parameter - will not print progress if set
|
||||
flagQuiet = flag.Bool("quiet", false, "Quiet mode")
|
||||
)
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// MAIN
|
||||
|
||||
func main() {
|
||||
flag.Usage = func() {
|
||||
name := filepath.Base(flag.CommandLine.Name())
|
||||
fmt.Fprintf(flag.CommandLine.Output(), "Usage: %s [options] <model>\n\n", name)
|
||||
flag.PrintDefaults()
|
||||
}
|
||||
flag.Parse()
|
||||
|
||||
// Get output path
|
||||
out, err := GetOut()
|
||||
if err != nil {
|
||||
fmt.Fprintln(os.Stderr, "Error:", err)
|
||||
os.Exit(-1)
|
||||
}
|
||||
|
||||
// Create context which quits on SIGINT or SIGQUIT
|
||||
ctx := ContextForSignal(os.Interrupt, syscall.SIGQUIT)
|
||||
|
||||
// Progress filehandle
|
||||
progress := os.Stdout
|
||||
if *flagQuiet {
|
||||
progress, err = os.Open(os.DevNull)
|
||||
if err != nil {
|
||||
fmt.Fprintln(os.Stderr, "Error:", err)
|
||||
os.Exit(-1)
|
||||
}
|
||||
defer progress.Close()
|
||||
}
|
||||
|
||||
// Download models - exit on error or interrupt
|
||||
for _, model := range GetModels() {
|
||||
url, err := URLForModel(model)
|
||||
if err != nil {
|
||||
fmt.Fprintln(os.Stderr, "Error:", err)
|
||||
continue
|
||||
} else if path, err := Download(ctx, progress, url, out); err == nil || err == io.EOF {
|
||||
continue
|
||||
} else if err == context.Canceled {
|
||||
os.Remove(path)
|
||||
fmt.Fprintln(progress, "\nInterrupted")
|
||||
break
|
||||
} else if err == context.DeadlineExceeded {
|
||||
os.Remove(path)
|
||||
fmt.Fprintln(progress, "Timeout downloading model")
|
||||
continue
|
||||
} else {
|
||||
os.Remove(path)
|
||||
fmt.Fprintln(os.Stderr, "Error:", err)
|
||||
break
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// PUBLIC METHODS
|
||||
|
||||
// GetOut returns the path to the output directory
|
||||
func GetOut() (string, error) {
|
||||
if *flagOut == "" {
|
||||
return os.Getwd()
|
||||
}
|
||||
if info, err := os.Stat(*flagOut); err != nil {
|
||||
return "", err
|
||||
} else if !info.IsDir() {
|
||||
return "", fmt.Errorf("not a directory: %s", info.Name())
|
||||
} else {
|
||||
return *flagOut, nil
|
||||
}
|
||||
}
|
||||
|
||||
// GetModels returns the list of models to download
|
||||
func GetModels() []string {
|
||||
if flag.NArg() == 0 {
|
||||
return modelNames
|
||||
} else {
|
||||
return flag.Args()
|
||||
}
|
||||
}
|
||||
|
||||
// URLForModel returns the URL for the given model on huggingface.co
|
||||
func URLForModel(model string) (string, error) {
|
||||
url, err := url.Parse(srcUrl)
|
||||
if err != nil {
|
||||
return "", err
|
||||
} else {
|
||||
url.Path = srcPathPrefix + "-" + model + srcExt
|
||||
}
|
||||
return url.String(), nil
|
||||
}
|
||||
|
||||
// Download downloads the model from the given URL to the given output directory
|
||||
func Download(ctx context.Context, p io.Writer, model, out string) (string, error) {
|
||||
// Create HTTP client
|
||||
client := http.Client{
|
||||
Timeout: *flagTimeout,
|
||||
}
|
||||
|
||||
// Initiate the download
|
||||
req, err := http.NewRequest("GET", model, nil)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
resp, err := client.Do(req)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
defer resp.Body.Close()
|
||||
if resp.StatusCode != http.StatusOK {
|
||||
return "", fmt.Errorf("%s: %s", model, resp.Status)
|
||||
}
|
||||
|
||||
// If output file exists and is the same size as the model, skip
|
||||
path := filepath.Join(out, filepath.Base(model))
|
||||
if info, err := os.Stat(path); err == nil && info.Size() == resp.ContentLength {
|
||||
fmt.Fprintln(p, "Skipping", model, "as it already exists")
|
||||
return "", nil
|
||||
}
|
||||
|
||||
// Create file
|
||||
w, err := os.Create(path)
|
||||
if err != nil {
|
||||
return "", err
|
||||
}
|
||||
defer w.Close()
|
||||
|
||||
// Report
|
||||
fmt.Fprintln(p, "Downloading", model, "to", out)
|
||||
|
||||
// Progressively download the model
|
||||
data := make([]byte, bufSize)
|
||||
count, pct := int64(0), int64(0)
|
||||
ticker := time.NewTicker(5 * time.Second)
|
||||
for {
|
||||
select {
|
||||
case <-ctx.Done():
|
||||
// Cancelled, return error
|
||||
return path, ctx.Err()
|
||||
case <-ticker.C:
|
||||
pct = DownloadReport(p, pct, count, resp.ContentLength)
|
||||
default:
|
||||
// Read body
|
||||
n, err := resp.Body.Read(data)
|
||||
if err != nil {
|
||||
DownloadReport(p, pct, count, resp.ContentLength)
|
||||
return path, err
|
||||
} else if m, err := w.Write(data[:n]); err != nil {
|
||||
return path, err
|
||||
} else {
|
||||
count += int64(m)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Report periodically reports the download progress when percentage changes
|
||||
func DownloadReport(w io.Writer, pct, count, total int64) int64 {
|
||||
pct_ := count * 100 / total
|
||||
if pct_ > pct {
|
||||
fmt.Fprintf(w, " ...%d MB written (%d%%)\n", count/1e6, pct_)
|
||||
}
|
||||
return pct_
|
||||
}
|
||||
61
bindings/go/examples/go-whisper/flags.go
Executable file
61
bindings/go/examples/go-whisper/flags.go
Executable file
@@ -0,0 +1,61 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"flag"
|
||||
)
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// TYPES
|
||||
|
||||
type Flags struct {
|
||||
*flag.FlagSet
|
||||
}
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// LIFECYCLE
|
||||
|
||||
func NewFlags(name string, args []string) (*Flags, error) {
|
||||
flags := &Flags{
|
||||
FlagSet: flag.NewFlagSet(name, flag.ContinueOnError),
|
||||
}
|
||||
|
||||
// Register the command line arguments
|
||||
registerFlags(flags)
|
||||
|
||||
// Parse command line
|
||||
if err := flags.Parse(args); err != nil {
|
||||
return nil, err
|
||||
}
|
||||
|
||||
// Return success
|
||||
return flags, nil
|
||||
}
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// PUBLIC METHODS
|
||||
|
||||
func (flags *Flags) GetModel() string {
|
||||
return flags.Lookup("model").Value.String()
|
||||
}
|
||||
|
||||
func (flags *Flags) GetLanguage() string {
|
||||
return flags.Lookup("language").Value.String()
|
||||
}
|
||||
|
||||
func (flags *Flags) IsSpeedup() bool {
|
||||
return flags.Lookup("speedup").Value.String() == "true"
|
||||
}
|
||||
|
||||
func (flags *Flags) IsTokens() bool {
|
||||
return flags.Lookup("tokens").Value.String() == "true"
|
||||
}
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////
|
||||
// PRIVATE METHODS
|
||||
|
||||
func registerFlags(flag *Flags) {
|
||||
flag.String("model", "", "Path to the model file")
|
||||
flag.String("language", "", "Language")
|
||||
flag.Bool("speedup", false, "Enable speedup")
|
||||
flag.Bool("tokens", false, "Display tokens")
|
||||
}
|
||||
44
bindings/go/examples/go-whisper/main.go
Executable file
44
bindings/go/examples/go-whisper/main.go
Executable file
@@ -0,0 +1,44 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"flag"
|
||||
"fmt"
|
||||
"os"
|
||||
"path/filepath"
|
||||
|
||||
// Packages
|
||||
whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
|
||||
)
|
||||
|
||||
func main() {
|
||||
flags, err := NewFlags(filepath.Base(os.Args[0]), os.Args[1:])
|
||||
if err == flag.ErrHelp {
|
||||
os.Exit(0)
|
||||
} else if err != nil {
|
||||
fmt.Fprintln(os.Stderr, err)
|
||||
os.Exit(1)
|
||||
} else if flags.GetModel() == "" {
|
||||
fmt.Fprintln(os.Stderr, "Use -model flag to specify which model file to use")
|
||||
os.Exit(1)
|
||||
} else if flags.NArg() == 0 {
|
||||
fmt.Fprintln(os.Stderr, "No input files specified")
|
||||
os.Exit(1)
|
||||
}
|
||||
|
||||
// Load model
|
||||
model, err := whisper.New(flags.GetModel())
|
||||
if err != nil {
|
||||
fmt.Fprintln(os.Stderr, err)
|
||||
os.Exit(1)
|
||||
}
|
||||
defer model.Close()
|
||||
|
||||
// Process files
|
||||
for _, filename := range flags.Args() {
|
||||
fmt.Println("Processing", filename)
|
||||
if err := Process(model, filename, flags.GetLanguage(), flags.IsSpeedup(), flags.IsTokens()); err != nil {
|
||||
fmt.Fprintln(os.Stderr, err)
|
||||
continue
|
||||
}
|
||||
}
|
||||
}
|
||||
80
bindings/go/examples/go-whisper/process.go
Executable file
80
bindings/go/examples/go-whisper/process.go
Executable file
@@ -0,0 +1,80 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"io"
|
||||
"os"
|
||||
"time"
|
||||
|
||||
// Package imports
|
||||
whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
|
||||
wav "github.com/go-audio/wav"
|
||||
)
|
||||
|
||||
func Process(model whisper.Model, path string, lang string, speedup, tokens bool) error {
|
||||
var data []float32
|
||||
|
||||
// Create processing context
|
||||
context, err := model.NewContext()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
// Open the file
|
||||
fh, err := os.Open(path)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
defer fh.Close()
|
||||
|
||||
// Decode the WAV file
|
||||
dec := wav.NewDecoder(fh)
|
||||
if buf, err := dec.FullPCMBuffer(); err != nil {
|
||||
return err
|
||||
} else if dec.SampleRate != whisper.SampleRate {
|
||||
return fmt.Errorf("unsupported sample rate: %d", dec.SampleRate)
|
||||
} else if dec.NumChans != 1 {
|
||||
return fmt.Errorf("unsupported number of channels: %d", dec.NumChans)
|
||||
} else {
|
||||
data = buf.AsFloat32Buffer().Data
|
||||
}
|
||||
|
||||
// Set the parameters
|
||||
var cb whisper.SegmentCallback
|
||||
if lang != "" {
|
||||
if err := context.SetLanguage(lang); err != nil {
|
||||
return err
|
||||
}
|
||||
}
|
||||
if speedup {
|
||||
context.SetSpeedup(true)
|
||||
}
|
||||
if tokens {
|
||||
cb = func(segment whisper.Segment) {
|
||||
fmt.Printf("%02d [%6s->%6s] ", segment.Num, segment.Start.Truncate(time.Millisecond), segment.End.Truncate(time.Millisecond))
|
||||
for _, token := range segment.Tokens {
|
||||
fmt.Printf("%q ", token.Text)
|
||||
}
|
||||
fmt.Println("")
|
||||
}
|
||||
}
|
||||
|
||||
// Process the data
|
||||
if err := context.Process(data, cb); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
// Print out the results
|
||||
for {
|
||||
segment, err := context.NextSegment()
|
||||
if err == io.EOF {
|
||||
break
|
||||
} else if err != nil {
|
||||
return err
|
||||
}
|
||||
fmt.Printf("[%6s->%6s] %s\n", segment.Start.Truncate(time.Millisecond), segment.End.Truncate(time.Millisecond), segment.Text)
|
||||
}
|
||||
|
||||
// Return success
|
||||
return nil
|
||||
}
|
||||
Reference in New Issue
Block a user