feat(ark): support partial assistant message (#670)

Author: meguminnnnnnnnn

components/model/ark/go.mod

@@ -9,7 +9,7 @@ require (
 	github.com/eino-contrib/jsonschema v1.0.3
 	github.com/smartystreets/goconvey v1.8.1
 	github.com/stretchr/testify v1.11.1
-	github.com/volcengine/volcengine-go-sdk v1.1.49
+	github.com/volcengine/volcengine-go-sdk v1.2.9
 
 )
 

components/model/ark/go.sum

@@ -138,6 +138,8 @@ github.com/volcengine/volc-sdk-golang v1.0.23 h1:anOslb2Qp6ywnsbyq9jqR0ljuO63kg9
 github.com/volcengine/volc-sdk-golang v1.0.23/go.mod h1:AfG/PZRUkHJ9inETvbjNifTDgut25Wbkm2QoYBTbvyU=
 github.com/volcengine/volcengine-go-sdk v1.1.49 h1:jkk3Zt6uFGiZshrVshsdRvadzuHIf4nLkekIZM+wLkY=
 github.com/volcengine/volcengine-go-sdk v1.1.49/go.mod h1:oxoVo+A17kvkwPkIeIHPVLjSw7EQAm+l/Vau1YGHN+A=
+github.com/volcengine/volcengine-go-sdk v1.2.9 h1:du2gnImtyWXKkQFnJW/GXCs+UBibGGOXIbP1Ams2pB8=
+github.com/volcengine/volcengine-go-sdk v1.2.9/go.mod h1:oxoVo+A17kvkwPkIeIHPVLjSw7EQAm+l/Vau1YGHN+A=
 github.com/wk8/go-ordered-map/v2 v2.1.8 h1:5h/BUHu93oj4gIdvHHHGsScSTMijfx5PeYkE/fJgbpc=
 github.com/wk8/go-ordered-map/v2 v2.1.8/go.mod h1:5nJHM5DyteebpVlHnWMV0rPz6Zp7+xBAnxjb1X5vnTw=
 github.com/x-cray/logrus-prefixed-formatter v0.5.2 h1:00txxvfBM9muc0jiLIEAkAcIMJzfthRT6usrui8uGmg=

components/model/ark/message_extra.go

@@ -30,6 +30,7 @@ const (
 	keyOfResponseID            = "ark-response-id"
 	keyOfResponseCacheExpireAt = "ark-response-cache-expire-at"
 	keyOfServiceTier           = "ark-service-tier"
+	keyOfPartial               = "ark-partial"
 	ImageSizeKey               = "seedream-image-size"
 )
 
@@ -162,14 +163,7 @@ func InvalidateMessageCaches(messages []*schema.Message) error {
 			continue
 		}
 
-		// there may be concurrency
-		extra := make(map[string]any, len(msg.Extra))
-		for k, v := range msg.Extra {
-			extra[k] = v
-		}
-
-		delete(extra, keyOfResponseCacheExpireAt)
-		msg.Extra = extra
+		delete(msg.Extra, keyOfResponseCacheExpireAt)
 	}
 	return nil
 }
@@ -380,3 +374,21 @@ func getImageSize(extra map[string]any) (string, bool) {
 	}
 	return size, true
 }
+
+// SetPartial marks the message as a partial message to enable continuation (prefill) mode.
+// By pre-filling part of the assistant role's content, it guides and controls the model
+// to continue generating from existing text fragments and maintain consistency in role-play scenarios.
+// To use this, set the role of the last message in the input list to assistant and call SetPartial
+// on it. The model will then continue writing based on the message's content.
+// Only available for ResponsesAPI.
+func SetPartial(msg *schema.Message) {
+	setMsgExtra(msg, keyOfPartial, true)
+}
+
+func getPartial(msg *schema.Message) bool {
+	v, ok := getMsgExtraValue[bool](msg, keyOfPartial)
+	if !ok {
+		return false
+	}
+	return v
+}

components/model/ark/responses_api.go

@@ -793,6 +793,11 @@ func (cm *ResponsesAPIChatModel) toArkAssistantRoleItemInputMessage(msg *schema.
 		Role: responses.MessageRole_assistant,
 	}
 
+	if getPartial(msg) {
+		b := true
+		inputItemMessage.Partial = &b
+	}
+
 	if len(msg.UserInputMultiContent) > 0 {
 		return nil, fmt.Errorf("if assistant role, UserInputMultiContent cannot be set")
 	}