feat(ark): support doubao vision series models (#11740)

Author: sinomoe

api/core/model_runtime/model_providers/volcengine_maas/client.py

@@ -1,4 +1,3 @@
-import re
 from collections.abc import Generator
 from typing import Optional, cast
 
@@ -104,17 +103,16 @@ def convert_prompt_message(message: PromptMessage) -> ChatCompletionMessageParam
                     if message_content.type == PromptMessageContentType.TEXT:
                         content.append(
                             ChatCompletionContentPartTextParam(
-                                text=message_content.text,
+                                text=message_content.data,
                                 type="text",
                             )
                         )
                     elif message_content.type == PromptMessageContentType.IMAGE:
                         message_content = cast(ImagePromptMessageContent, message_content)
-                        image_data = re.sub(r"^data:image\/[a-zA-Z]+;base64,", "", message_content.data)
                         content.append(
                             ChatCompletionContentPartImageParam(
                                 image_url=ImageURL(
-                                    url=image_data,
+                                    url=message_content.data,
                                     detail=message_content.detail.value,
                                 ),
                                 type="image_url",

api/core/model_runtime/model_providers/volcengine_maas/llm/llm.py

@@ -132,6 +132,14 @@ def _get_num_tokens_v3(self, messages: list[PromptMessage]) -> int:
         messages_dict = [ArkClientV3.convert_prompt_message(m) for m in messages]
         for message in messages_dict:
             for key, value in message.items():
+                # Ignore tokens for image type
+                if isinstance(value, list):
+                    text = ""
+                    for item in value:
+                        if isinstance(item, dict) and item["type"] == "text":
+                            text += item["text"]
+
+                    value = text
                 num_tokens += self._get_num_tokens_by_gpt2(str(key))
                 num_tokens += self._get_num_tokens_by_gpt2(str(value))
 

api/core/model_runtime/model_providers/volcengine_maas/llm/models.py

@@ -16,6 +16,14 @@ class ModelConfig(BaseModel):
 
 
 configs: dict[str, ModelConfig] = {
+    "Doubao-vision-pro-32k": ModelConfig(
+        properties=ModelProperties(context_size=32768, max_tokens=4096, mode=LLMMode.CHAT),
+        features=[ModelFeature.VISION],
+    ),
+    "Doubao-vision-lite-32k": ModelConfig(
+        properties=ModelProperties(context_size=32768, max_tokens=4096, mode=LLMMode.CHAT),
+        features=[ModelFeature.VISION],
+    ),
     "Doubao-pro-4k": ModelConfig(
         properties=ModelProperties(context_size=4096, max_tokens=4096, mode=LLMMode.CHAT),
         features=[ModelFeature.TOOL_CALL],

api/core/model_runtime/model_providers/volcengine_maas/volcengine_maas.yaml

@@ -118,6 +118,18 @@ model_credential_schema:
       type: select
       required: true
       options:
+        - label:
+            en_US: Doubao-vision-pro-32k
+          value: Doubao-vision-pro-32k
+          show_on:
+            - variable: __model_type
+              value: llm
+        - label:
+            en_US: Doubao-vision-lite-32k
+          value: Doubao-vision-lite-32k
+          show_on:
+            - variable: __model_type
+              value: llm
         - label:
             en_US: Doubao-pro-4k
           value: Doubao-pro-4k