model.go 4.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249
  1. package model
  2. import (
  3. "errors"
  4. "fmt"
  5. "image"
  6. _ "image/jpeg"
  7. _ "image/png"
  8. "log/slog"
  9. "os"
  10. "reflect"
  11. "strconv"
  12. "strings"
  13. _ "golang.org/x/image/bmp"
  14. _ "golang.org/x/image/tiff"
  15. _ "golang.org/x/image/webp"
  16. "github.com/ollama/ollama/kvcache"
  17. "github.com/ollama/ollama/ml"
  18. _ "github.com/ollama/ollama/ml/backend"
  19. )
  20. type Options struct {
  21. Inputs []int32
  22. Positions []int32
  23. Sequences []int
  24. Outputs []int32
  25. Images []image.Image
  26. }
  27. type config struct {
  28. Cache kvcache.Cache
  29. }
  30. type Base struct {
  31. b ml.Backend
  32. config
  33. }
  34. func (m *Base) Backend() ml.Backend {
  35. return m.b
  36. }
  37. func (m *Base) Config() config {
  38. return m.config
  39. }
  40. type Model interface {
  41. Forward(ml.Context, Options) (ml.Tensor, error)
  42. Backend() ml.Backend
  43. Config() config
  44. }
  45. var models = make(map[string]func(ml.Config) (Model, error))
  46. func Register(name string, f func(ml.Config) (Model, error)) {
  47. if _, ok := models[name]; ok {
  48. panic("model: model already registered")
  49. }
  50. models[name] = f
  51. }
  52. func New(s string) (Model, error) {
  53. r, err := os.Open(s)
  54. if err != nil {
  55. return nil, err
  56. }
  57. defer r.Close()
  58. b, err := ml.NewBackend(r)
  59. if err != nil {
  60. return nil, err
  61. }
  62. arch := b.Config().Architecture()
  63. f, ok := models[arch]
  64. if !ok {
  65. return nil, fmt.Errorf("unsupported model architecture %q", arch)
  66. }
  67. m, err := f(b.Config())
  68. if err != nil {
  69. return nil, err
  70. }
  71. base := Base{b: b, config: m.Config()}
  72. v := reflect.ValueOf(m)
  73. v.Elem().Set(populateFields(base, v.Elem()))
  74. return m, nil
  75. }
  76. func populateFields(base Base, v reflect.Value, tags ...Tag) reflect.Value {
  77. t := v.Type()
  78. if t.Kind() == reflect.Struct {
  79. allNil := true
  80. for i := range t.NumField() {
  81. tt := t.Field(i).Type
  82. vv := v.Field(i)
  83. if !vv.CanSet() {
  84. continue
  85. }
  86. // make a copy
  87. tagsCopy := tags
  88. if tag := t.Field(i).Tag.Get("gguf"); tag != "" {
  89. tagsCopy = append(tagsCopy, ParseTags(tag))
  90. }
  91. if tt == reflect.TypeOf((*Base)(nil)).Elem() {
  92. vv.Set(reflect.ValueOf(base))
  93. } else if tt == reflect.TypeOf((*ml.Tensor)(nil)).Elem() {
  94. var fn func([]Tag) [][]string
  95. fn = func(tags []Tag) (values [][]string) {
  96. if len(tags) < 1 {
  97. return nil
  98. }
  99. values = [][]string{{tags[0].Name}}
  100. for _, alt := range tags[0].Alternate {
  101. values = append(values, []string{alt})
  102. }
  103. for i, value := range values {
  104. for _, rest := range fn(tags[1:]) {
  105. value = append(value, rest...)
  106. }
  107. values[i] = value
  108. }
  109. return values
  110. }
  111. names := fn(tagsCopy)
  112. for _, name := range names {
  113. if tensor := base.Backend().Get(strings.Join(name, ".")); tensor != nil {
  114. slog.Debug("found tensor", "", tensor)
  115. vv.Set(reflect.ValueOf(tensor))
  116. break
  117. }
  118. }
  119. } else if tt.Kind() == reflect.Pointer || tt.Kind() == reflect.Interface {
  120. setPointer(base, vv, tagsCopy)
  121. } else if tt.Kind() == reflect.Slice || tt.Kind() == reflect.Array {
  122. for i := range vv.Len() {
  123. vvv := vv.Index(i)
  124. if vvv.Kind() == reflect.Pointer || vvv.Kind() == reflect.Interface {
  125. setPointer(base, vvv, append(tagsCopy, Tag{Name: strconv.Itoa(i)}))
  126. } else {
  127. vvv.Set(populateFields(base, vvv, append(tagsCopy, Tag{Name: strconv.Itoa(i)})...))
  128. }
  129. }
  130. }
  131. if !canNil(tt) || !vv.IsNil() {
  132. allNil = false
  133. }
  134. }
  135. if allNil {
  136. return reflect.Zero(t)
  137. }
  138. }
  139. return v
  140. }
  141. func setPointer(base Base, v reflect.Value, tags []Tag) {
  142. vv := v
  143. if v.Kind() == reflect.Interface {
  144. if v.IsNil() {
  145. return
  146. }
  147. vv = vv.Elem()
  148. }
  149. vv = vv.Elem()
  150. if v.IsNil() {
  151. vv = reflect.New(v.Type().Elem()).Elem()
  152. }
  153. if f := populateFields(base, vv, tags...); f.CanAddr() {
  154. v.Set(f.Addr())
  155. }
  156. }
  157. type Tag struct {
  158. Name string
  159. Alternate []string
  160. }
  161. func ParseTags(s string) (tag Tag) {
  162. parts := strings.Split(s, ",")
  163. if len(parts) > 0 {
  164. tag.Name = parts[0]
  165. for _, part := range parts[1:] {
  166. if value, ok := strings.CutPrefix(part, "alt:"); ok {
  167. tag.Alternate = append(tag.Alternate, value)
  168. }
  169. }
  170. }
  171. return
  172. }
  173. func canNil(t reflect.Type) bool {
  174. return t.Kind() == reflect.Chan ||
  175. t.Kind() == reflect.Func ||
  176. t.Kind() == reflect.Interface ||
  177. t.Kind() == reflect.Map ||
  178. t.Kind() == reflect.Pointer ||
  179. t.Kind() == reflect.Slice
  180. }
  181. func Forward(ctx ml.Context, m Model, opts Options) (ml.Tensor, error) {
  182. if len(opts.Positions) != len(opts.Sequences) {
  183. return nil, fmt.Errorf("length of positions (%v) must match length of seqs (%v)", len(opts.Positions), len(opts.Sequences))
  184. }
  185. if len(opts.Positions) < 1 {
  186. return nil, errors.New("batch size cannot be less than 1")
  187. }
  188. cache := m.Config().Cache
  189. if cache != nil {
  190. err := cache.StartForward(ctx, opts.Positions, opts.Sequences)
  191. if err != nil {
  192. return nil, err
  193. }
  194. }
  195. t, err := m.Forward(ctx, opts)
  196. if err != nil {
  197. return nil, err
  198. }
  199. ctx.Forward(t)
  200. ctx.Compute(t)
  201. return t, nil
  202. }