|
| 1 | +package llnotes |
| 2 | + |
| 3 | +import ( |
| 4 | + "fmt" |
| 5 | + "strings" |
| 6 | + |
| 7 | + "github.com/databricks/databricks-sdk-go/service/serving" |
| 8 | + "github.com/databrickslabs/sandbox/go-libs/sed" |
| 9 | +) |
| 10 | + |
| 11 | +type message interface { |
| 12 | + ChatMessage() serving.ChatMessage |
| 13 | +} |
| 14 | + |
| 15 | +type SystemMessage string |
| 16 | + |
| 17 | +func (m SystemMessage) ChatMessage() serving.ChatMessage { |
| 18 | + return serving.ChatMessage{ |
| 19 | + Role: serving.ChatMessageRoleSystem, |
| 20 | + Content: string(m), |
| 21 | + } |
| 22 | +} |
| 23 | + |
| 24 | +type UserMessage string |
| 25 | + |
| 26 | +func (m UserMessage) ChatMessage() serving.ChatMessage { |
| 27 | + return serving.ChatMessage{ |
| 28 | + Role: serving.ChatMessageRoleUser, |
| 29 | + Content: string(m), |
| 30 | + } |
| 31 | +} |
| 32 | + |
| 33 | +type AssistantMessage string |
| 34 | + |
| 35 | +func (m AssistantMessage) ChatMessage() serving.ChatMessage { |
| 36 | + return serving.ChatMessage{ |
| 37 | + Role: serving.ChatMessageRoleAssistant, |
| 38 | + Content: string(m), |
| 39 | + } |
| 40 | +} |
| 41 | + |
| 42 | +type History []message |
| 43 | + |
| 44 | +func (h History) Messages() (out []serving.ChatMessage) { |
| 45 | + for _, v := range h { |
| 46 | + out = append(out, v.ChatMessage()) |
| 47 | + } |
| 48 | + return |
| 49 | +} |
| 50 | + |
| 51 | +func (h History) messageTokens(m message) int { |
| 52 | + // this is good enough approximation of message token count |
| 53 | + content := m.ChatMessage().Content |
| 54 | + return len(strings.Split(content, " ")) |
| 55 | +} |
| 56 | + |
| 57 | +func (h History) truncated(m message, maxTokens int) message { |
| 58 | + // this is good enough approximation of message token count |
| 59 | + cm := m.ChatMessage() |
| 60 | + tokens := strings.Split(cm.Content, " ") |
| 61 | + if len(tokens) < maxTokens { |
| 62 | + return m |
| 63 | + } |
| 64 | + switch m.(type) { |
| 65 | + case UserMessage: |
| 66 | + return UserMessage(strings.Join(tokens[:maxTokens-100], " ")) |
| 67 | + case SystemMessage: |
| 68 | + return SystemMessage(strings.Join(tokens[:maxTokens-100], " ")) |
| 69 | + case AssistantMessage: |
| 70 | + return AssistantMessage(strings.Join(tokens[:maxTokens-100], " ")) |
| 71 | + } |
| 72 | + panic("cannot truncate message") |
| 73 | +} |
| 74 | + |
| 75 | +func (h History) totalTokens() int { |
| 76 | + totalTokens := 0 |
| 77 | + for _, m := range h { |
| 78 | + totalTokens += h.messageTokens(m) |
| 79 | + } |
| 80 | + return totalTokens |
| 81 | +} |
| 82 | + |
| 83 | +func (h History) With(m message) History { |
| 84 | + maxContextSize := 32768 |
| 85 | + increment := h.messageTokens(m) |
| 86 | + if increment > maxContextSize { |
| 87 | + m = h.truncated(m, 32000) |
| 88 | + } |
| 89 | + return append(h, m) |
| 90 | +} |
| 91 | + |
| 92 | +func (h History) Last() string { |
| 93 | + return h[len(h)-1].ChatMessage().Content |
| 94 | +} |
| 95 | + |
| 96 | +func (h History) Excerpt(n int) string { |
| 97 | + var out []string |
| 98 | + oneLine := sed.Rule(`\n|\s+`, ` `) |
| 99 | + for i, v := range h { |
| 100 | + m := v.ChatMessage() |
| 101 | + out = append(out, fmt.Sprintf("(%d/%d) %s: %s", |
| 102 | + i+1, len(h), |
| 103 | + strings.ToUpper(m.Role.String()), |
| 104 | + h.onlyNBytes(oneLine.Apply(m.Content), n))) |
| 105 | + } |
| 106 | + return strings.Join(out, "\n") |
| 107 | +} |
| 108 | + |
| 109 | +func (h History) onlyNBytes(j string, numBytes int) string { |
| 110 | + diff := len([]byte(j)) - numBytes |
| 111 | + if diff > 0 { |
| 112 | + return fmt.Sprintf("%s... (%d more bytes)", j[:numBytes], diff) |
| 113 | + } |
| 114 | + return j |
| 115 | +} |
0 commit comments