fix(openai-embedding): add dimensions request parameter toggle

astrbot/core/config/default.py

@@ -1818,6 +1818,7 @@
                         "embedding_api_base": "",
                         "embedding_model": "",
                         "embedding_dimensions": 1024,
+                        "embedding_dimensions_as_request_param": False,
                         "timeout": 20,
                         "proxy": "",
                     },
@@ -2190,6 +2191,12 @@
                         "hint": "嵌入向量的维度。根据模型不同，可能需要调整，请参考具体模型的文档。此配置项请务必填写正确，否则将导致向量数据库无法正常工作。",
                         "_special": "get_embedding_dim",
                     },
+                    "embedding_dimensions_as_request_param": {
+                        "description": "将嵌入维度作为请求参数发送",
+                        "type": "bool",
+                        "hint": "开启后会把嵌入维度作为 dimensions 参数发送给 OpenAI-compatible Embedding API。仅在模型明确支持该参数时开启。",
+                        "invisible": True,
+                    },
                     "embedding_model": {
                         "description": "嵌入模型",
                         "type": "string",

astrbot/core/provider/sources/openai_embedding_source.py

@@ -64,15 +64,23 @@ async def get_embeddings(self, text: list[str]) -> list[list[float]]:
 
     def _embedding_kwargs(self) -> dict:
         """构建嵌入请求的可选参数"""
-        kwargs = {}
-        if "embedding_dimensions" in self.provider_config:
-            try:
-                kwargs["dimensions"] = int(self.provider_config["embedding_dimensions"])
-            except (ValueError, TypeError):
-                logger.warning(
-                    f"embedding_dimensions in embedding configs is not a valid integer: '{self.provider_config['embedding_dimensions']}', ignored."
-                )
-        return kwargs
+        if not self.provider_config.get("embedding_dimensions_as_request_param", False):
+            return {}
+
+        if "embedding_dimensions" not in self.provider_config:
+            return {}
+
+        try:
+            dimensions = int(self.provider_config["embedding_dimensions"])
+        except (ValueError, TypeError):
+            logger.warning(
+                f"embedding_dimensions in embedding configs is not a valid integer: '{self.provider_config['embedding_dimensions']}', ignored."
+            )
+            return {}
+
+        if dimensions <= 0:
+            return {}
+        return {"dimensions": dimensions}
 
     def get_dim(self) -> int:
         """获取向量的维度"""

dashboard/src/components/shared/AstrBotConfig.vue

@@ -272,6 +272,7 @@ function hasVisibleItemsAfter(items, currentIndex) {
                 :plugin-name="pluginName"
                 :plugin-i18n="pluginI18n"
                 :config-key="getItemPath(key)"
+                :config-root="iterable"
                 :loading="loadingEmbeddingDim"
                 :show-fullscreen-btn="!!metadata[metadataKey].items[key]?.editor_mode"
                 @get-embedding-dim="getEmbeddingDimensions(iterable)"
@@ -322,6 +323,7 @@ function hasVisibleItemsAfter(items, currentIndex) {
             :plugin-name="pluginName"
             :plugin-i18n="pluginI18n"
             :config-key="getItemPath(metadataKey)"
+            :config-root="iterable"
           />
         </v-col>
       </v-row>

dashboard/src/components/shared/ConfigItemRenderer.vue

@@ -73,6 +73,25 @@
         >
           {{ t('core.common.autoDetect') }}
         </v-btn>
+        <v-tooltip
+          v-if="hasEmbeddingDimensionsRequestParam"
+          :text="t('core.common.embeddingDimensionsRequestParamHint')"
+          location="top"
+        >
+          <template #activator="{ props: tooltipProps }">
+            <v-switch
+              v-bind="tooltipProps"
+              :model-value="Boolean(configRoot.embedding_dimensions_as_request_param)"
+              @update:model-value="setEmbeddingDimensionsRequestParam"
+              :label="t('core.common.embeddingDimensionsRequestParam')"
+              color="primary"
+              inset
+              density="compact"
+              hide-details
+              class="dimensions-param-switch ml-1"
+            ></v-switch>
+          </template>
+        </v-tooltip>
       </div>
     </template>
 
@@ -307,11 +326,24 @@ const listSelectItems = computed(() =>
     : []
 )
 
+const hasEmbeddingDimensionsRequestParam = computed(() =>
+  props.configRoot
+  && Object.prototype.hasOwnProperty.call(
+    props.configRoot,
+    'embedding_dimensions_as_request_param'
+  )
+)
+
 function toNumber(val) {
   const n = parseFloat(val)
   return isNaN(n) ? 0 : n
 }
 
+function setEmbeddingDimensionsRequestParam(val) {
+  if (!props.configRoot) return
+  props.configRoot.embedding_dimensions_as_request_param = Boolean(val)
+}
+
 function getLabel(itemMeta, index, option) {
   const labels = getTranslatedLabels(itemMeta)
   return labels ? labels[index] : option

dashboard/src/i18n/locales/en-US/core/common.json

@@ -38,6 +38,8 @@
   "no": "No",
   "imagePreview": "Image Preview",
   "autoDetect": "Auto Detect",
+  "embeddingDimensionsRequestParam": "Request Param",
+  "embeddingDimensionsRequestParamHint": "When enabled, the embedding dimension is sent as the API dimensions parameter. Disabled by default; enable only for models that explicitly support it.",
   "dialog": {
     "confirmTitle": "Confirm Action",
     "confirmMessage": "Are you sure you want to perform this action?",

dashboard/src/i18n/locales/ru-RU/core/common.json

@@ -38,6 +38,8 @@
     "no": "Нет",
     "imagePreview": "Предпросмотр изображения",
     "autoDetect": "Автоопределение",
+    "embeddingDimensionsRequestParam": "Параметр API",
+    "embeddingDimensionsRequestParamHint": "Если включено, размерность эмбеддинга отправляется как параметр dimensions. По умолчанию выключено; включайте только для моделей, которые явно поддерживают этот параметр.",
     "dialog": {
         "confirmTitle": "Подтверждение",
         "confirmMessage": "Вы уверены, что хотите выполнить это действие?",
@@ -130,4 +132,4 @@
             "subtitle": "Файл FIRST_NOTICE.md не найден или пуст."
         }
     }
-}
+}

dashboard/src/i18n/locales/zh-CN/core/common.json

@@ -38,6 +38,8 @@
   "no": "否",
   "imagePreview": "图片预览",
   "autoDetect": "自动检测",
+  "embeddingDimensionsRequestParam": "请求参数",
+  "embeddingDimensionsRequestParamHint": "开启后会把嵌入维度作为 dimensions 参数发送给 API。默认关闭；仅在模型明确支持时开启。",
   "dialog": {
     "confirmTitle": "确认操作",
     "confirmMessage": "你确定要执行此操作吗？",

tests/test_openai_embedding_source.py

@@ -0,0 +1,104 @@
+from types import SimpleNamespace
+
+import pytest
+
+from astrbot.core.provider.sources.openai_embedding_source import (
+    OpenAIEmbeddingProvider,
+)
+
+
+class FakeEmbeddingsClient:
+    def __init__(self):
+        self.calls = []
+
+    async def create(self, **kwargs):
+        self.calls.append(kwargs)
+        input_value = kwargs["input"]
+        if isinstance(input_value, list):
+            data = [
+                SimpleNamespace(embedding=[float(index), 0.0, 1.0])
+                for index, _ in enumerate(input_value)
+            ]
+        else:
+            data = [SimpleNamespace(embedding=[1.0, 2.0, 3.0])]
+        return SimpleNamespace(data=data)
+
+
+@pytest.mark.asyncio
+async def test_openai_embedding_does_not_send_dimensions_by_default():
+    provider = OpenAIEmbeddingProvider(
+        {
+            "id": "openai-compatible-embedding",
+            "embedding_api_key": "test-key",
+            "embedding_api_base": "https://example.com/v1",
+            "embedding_model": "BAAI/bge-m3",
+            "embedding_dimensions": 1024,
+        },
+        {},
+    )
+    fake_embeddings = FakeEmbeddingsClient()
+    provider.client = SimpleNamespace(embeddings=fake_embeddings)
+
+    embedding = await provider.get_embedding("hello")
+    embeddings = await provider.get_embeddings(["hello", "world"])
+
+    assert embedding == [1.0, 2.0, 3.0]
+    assert embeddings == [[0.0, 0.0, 1.0], [1.0, 0.0, 1.0]]
+    assert provider.get_dim() == 1024
+    assert fake_embeddings.calls == [
+        {"input": "hello", "model": "BAAI/bge-m3"},
+        {"input": ["hello", "world"], "model": "BAAI/bge-m3"},
+    ]
+
+
+@pytest.mark.asyncio
+async def test_openai_embedding_sends_dimensions_when_explicitly_enabled():
+    provider = OpenAIEmbeddingProvider(
+        {
+            "id": "openai-compatible-embedding",
+            "embedding_api_key": "test-key",
+            "embedding_api_base": "https://api.openai.com/v1",
+            "embedding_model": "text-embedding-3-small",
+            "embedding_dimensions": 512,
+            "embedding_dimensions_as_request_param": True,
+        },
+        {},
+    )
+    fake_embeddings = FakeEmbeddingsClient()
+    provider.client = SimpleNamespace(embeddings=fake_embeddings)
+
+    embedding = await provider.get_embedding("hello")
+
+    assert embedding == [1.0, 2.0, 3.0]
+    assert provider.get_dim() == 512
+    assert fake_embeddings.calls == [
+        {
+            "input": "hello",
+            "model": "text-embedding-3-small",
+            "dimensions": 512,
+        },
+    ]
+
+
+@pytest.mark.asyncio
+async def test_openai_embedding_omits_dimensions_when_dimension_not_configured():
+    provider = OpenAIEmbeddingProvider(
+        {
+            "id": "openai-compatible-embedding",
+            "embedding_api_key": "test-key",
+            "embedding_api_base": "https://example.com/v1",
+            "embedding_model": "BAAI/bge-m3",
+            "embedding_dimensions_as_request_param": True,
+        },
+        {},
+    )
+    fake_embeddings = FakeEmbeddingsClient()
+    provider.client = SimpleNamespace(embeddings=fake_embeddings)
+
+    embeddings = await provider.get_embeddings(["hello", "world"])
+
+    assert embeddings == [[0.0, 0.0, 1.0], [1.0, 0.0, 1.0]]
+    assert provider.get_dim() == 0
+    assert fake_embeddings.calls == [
+        {"input": ["hello", "world"], "model": "BAAI/bge-m3"},
+    ]