fix(chat): include history in model fallback

This commit is contained in:
c
2026-05-01 13:23:50 +00:00
committed by lyingbug
parent 8955bf1aa8
commit 48aeafcc3f
3 changed files with 162 additions and 10 deletions
@@ -75,11 +75,7 @@ func prepareMessagesWithHistory(chatManage *types.ChatManage) []chat.Message {
{Role: "system", Content: systemPrompt},
}
// Add conversation history (already limited by maxRounds in load_history/rewrite plugins)
for _, history := range chatManage.History {
chatMessages = append(chatMessages, chat.Message{Role: "user", Content: history.Query})
chatMessages = append(chatMessages, chat.Message{Role: "assistant", Content: history.Answer})
}
chatMessages = AppendHistoryMessages(chatMessages, chatManage.History)
// Add current user message. Only include images when the chat model supports
// vision; non-vision models rely on the text description in UserContent.
@@ -92,6 +88,16 @@ func prepareMessagesWithHistory(chatManage *types.ChatManage) []chat.Message {
return chatMessages
}
// AppendHistoryMessages appends prior Q&A rounds in chronological order.
// History is already filtered and truncated upstream by the load_history plugin.
func AppendHistoryMessages(messages []chat.Message, history []*types.History) []chat.Message {
for _, history := range history {
messages = append(messages, chat.Message{Role: "user", Content: history.Query})
messages = append(messages, chat.Message{Role: "assistant", Content: history.Answer})
}
return messages
}
// loadAndProcessHistory fetches recent messages, groups them into Q&A pairs,
// strips <think> tags from assistant answers, sorts by recency, and limits to maxRounds.
// fetchCount controls how many raw messages to fetch (typically maxRounds*2+10).
@@ -740,11 +740,7 @@ func (s *sessionService) handleModelFallback(ctx context.Context, chatManage *ty
}
// Start streaming response
userMsg := chat.Message{Role: "user", Content: promptContent}
if chatManage.ChatModelSupportsVision && len(chatManage.Images) > 0 {
userMsg.Images = chatManage.Images
}
responseChan, err := chatModel.ChatStream(ctx, []chat.Message{userMsg}, opt)
responseChan, err := chatModel.ChatStream(ctx, buildFallbackMessages(chatManage, promptContent), opt)
if err != nil {
logger.Errorf(ctx, "Failed to start streaming fallback response: %v, falling back to fixed response", err)
s.handleFixedFallback(ctx, chatManage)
@@ -761,6 +757,18 @@ func (s *sessionService) handleModelFallback(ctx context.Context, chatManage *ty
go s.consumeFallbackStream(ctx, chatManage, responseChan)
}
func buildFallbackMessages(chatManage *types.ChatManage, promptContent string) []chat.Message {
messages := make([]chat.Message, 0, len(chatManage.History)*2+1)
messages = chatpipeline.AppendHistoryMessages(messages, chatManage.History)
userMsg := chat.Message{Role: "user", Content: promptContent}
if chatManage.ChatModelSupportsVision && len(chatManage.Images) > 0 {
userMsg.Images = chatManage.Images
}
return append(messages, userMsg)
}
// renderFallbackPrompt renders the fallback prompt template with query and image context.
func (s *sessionService) renderFallbackPrompt(ctx context.Context, chatManage *types.ChatManage) (string, error) {
query := chatManage.Query
@@ -0,0 +1,138 @@
package service
import (
"context"
"testing"
"github.com/Tencent/WeKnora/internal/event"
"github.com/Tencent/WeKnora/internal/models/asr"
"github.com/Tencent/WeKnora/internal/models/chat"
"github.com/Tencent/WeKnora/internal/models/embedding"
"github.com/Tencent/WeKnora/internal/models/rerank"
"github.com/Tencent/WeKnora/internal/models/vlm"
"github.com/Tencent/WeKnora/internal/types"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
type captureChatModel struct {
lastMessages []chat.Message
}
func (m *captureChatModel) Chat(
context.Context,
[]chat.Message,
*chat.ChatOptions,
) (*types.ChatResponse, error) {
return nil, nil
}
func (m *captureChatModel) ChatStream(
_ context.Context,
messages []chat.Message,
_ *chat.ChatOptions,
) (<-chan types.StreamResponse, error) {
m.lastMessages = append([]chat.Message(nil), messages...)
ch := make(chan types.StreamResponse, 1)
ch <- types.StreamResponse{
ResponseType: types.ResponseTypeAnswer,
Content: "ok",
Done: true,
}
close(ch)
return ch, nil
}
func (m *captureChatModel) GetModelName() string { return "capture" }
func (m *captureChatModel) GetModelID() string { return "capture" }
type stubModelService struct {
chatModel chat.Chat
}
func (s *stubModelService) CreateModel(context.Context, *types.Model) error {
return nil
}
func (s *stubModelService) GetModelByID(context.Context, string) (*types.Model, error) {
return nil, nil
}
func (s *stubModelService) ListModels(context.Context) ([]*types.Model, error) {
return nil, nil
}
func (s *stubModelService) UpdateModel(context.Context, *types.Model) error {
return nil
}
func (s *stubModelService) DeleteModel(context.Context, string) error {
return nil
}
func (s *stubModelService) GetEmbeddingModel(context.Context, string) (embedding.Embedder, error) {
return nil, nil
}
func (s *stubModelService) GetEmbeddingModelForTenant(context.Context, string, uint64) (embedding.Embedder, error) {
return nil, nil
}
func (s *stubModelService) GetRerankModel(context.Context, string) (rerank.Reranker, error) {
return nil, nil
}
func (s *stubModelService) GetChatModel(context.Context, string) (chat.Chat, error) {
return s.chatModel, nil
}
func (s *stubModelService) GetVLMModel(context.Context, string) (vlm.VLM, error) {
return nil, nil
}
func (s *stubModelService) GetASRModel(context.Context, string) (asr.ASR, error) {
return nil, nil
}
func TestHandleModelFallback_IncludesHistoryMessages(t *testing.T) {
chatModel := &captureChatModel{}
svc := &sessionService{
modelService: &stubModelService{chatModel: chatModel},
}
bus := event.NewEventBus()
cm := &types.ChatManage{
PipelineRequest: types.PipelineRequest{
SessionID: "session-1",
Query: "现在还能继续讲吗?",
ChatModelID: "chat-model",
FallbackPrompt: "Answer the latest user question: {{query}}",
SummaryConfig: types.SummaryConfig{
Temperature: 0.2,
},
Language: "zh-CN",
},
PipelineState: types.PipelineState{
History: []*types.History{
{
Query: "先介绍一下 WeKnora",
Answer: "WeKnora 是一个知识库问答系统。",
},
},
},
PipelineContext: types.PipelineContext{
EventBus: bus.AsEventBusInterface(),
},
}
svc.handleModelFallback(context.Background(), cm)
require.Len(t, chatModel.lastMessages, 3)
assert.Equal(t, "user", chatModel.lastMessages[0].Role)
assert.Equal(t, "先介绍一下 WeKnora", chatModel.lastMessages[0].Content)
assert.Equal(t, "assistant", chatModel.lastMessages[1].Role)
assert.Equal(t, "WeKnora 是一个知识库问答系统。", chatModel.lastMessages[1].Content)
assert.Equal(t, "user", chatModel.lastMessages[2].Role)
assert.Contains(t, chatModel.lastMessages[2].Content, "现在还能继续讲吗?")
}