prompt.go 2.0 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374
  1. package server
  2. import (
  3. "bytes"
  4. "context"
  5. "log/slog"
  6. "github.com/ollama/ollama/api"
  7. "github.com/ollama/ollama/llm"
  8. "github.com/ollama/ollama/template"
  9. )
  10. type tokenizeFunc func(context.Context, string) ([]int, error)
  11. // chatPrompt accepts a list of messages and returns the prompt and images that should be used for the next chat turn.
  12. // chatPrompt truncates any messages that exceed the context window of the model, making sure to always include 1) the
  13. // latest message and 2) system messages
  14. func chatPrompt(ctx context.Context, m *Model, tokenize tokenizeFunc, opts *api.Options, msgs []api.Message, tools []api.Tool) (prompt string, images []llm.ImageData, _ error) {
  15. var system []api.Message
  16. // always include the last message
  17. n := len(msgs) - 1
  18. // in reverse, find all messages that fit into context window
  19. for i := n - 1; i >= 0; i-- {
  20. system = make([]api.Message, 0)
  21. for j := range i {
  22. if msgs[j].Role == "system" {
  23. system = append(system, msgs[j])
  24. }
  25. }
  26. var b bytes.Buffer
  27. if err := m.Template.Execute(&b, template.Values{Messages: append(system, msgs[i:]...), Tools: tools}); err != nil {
  28. return "", nil, err
  29. }
  30. s, err := tokenize(ctx, b.String())
  31. if err != nil {
  32. return "", nil, err
  33. }
  34. c := len(s)
  35. if m.ProjectorPaths != nil {
  36. for _, m := range msgs[i:] {
  37. // images are represented as 768 sized embeddings
  38. // TODO: get embedding length from project metadata
  39. c += 768 * len(m.Images)
  40. }
  41. }
  42. if c > opts.NumCtx {
  43. slog.Debug("truncating input messages which exceed context length", "truncated", len(msgs[i:]))
  44. break
  45. } else {
  46. n = i
  47. }
  48. }
  49. // truncate any messages that do not fit into the context window
  50. var b bytes.Buffer
  51. if err := m.Template.Execute(&b, template.Values{Messages: append(system, msgs[n:]...), Tools: tools}); err != nil {
  52. return "", nil, err
  53. }
  54. for _, m := range msgs[n:] {
  55. for _, i := range m.Images {
  56. images = append(images, llm.ImageData{
  57. ID: len(images),
  58. Data: i,
  59. })
  60. }
  61. }
  62. return b.String(), images, nil
  63. }