diff --git a/src/pages/llmodels/config/types.ts b/src/pages/llmodels/config/types.ts
index 1ae718ae..532ef75d 100644
--- a/src/pages/llmodels/config/types.ts
+++ b/src/pages/llmodels/config/types.ts
@@ -299,8 +299,13 @@ export interface BackendGroupOption {
 }
 
 export interface BackendOption {
+  value: string;
   label: string;
-  options: BackendGroupOption[];
+  title?: string;
+  default_backend_param: string[];
+  default_version: string;
+  isBuiltIn: boolean;
+  versions: { label: string; value: string; title?: string }[];
 }
 
 export interface AccessControlFormData {
diff --git a/src/pages/llmodels/forms/backend.tsx b/src/pages/llmodels/forms/backend.tsx
index 9ec7f3d5..5f6bc9bb 100644
--- a/src/pages/llmodels/forms/backend.tsx
+++ b/src/pages/llmodels/forms/backend.tsx
@@ -37,15 +37,35 @@ const BackendFields: React.FC = () => {
     return getBackendParamsTips(backend);
   }, [backend]);
 
+  const backendGroupedOptions = useMemo(() => {
+    const builtInBackends = backendOptions?.filter((item) => item.isBuiltIn);
+    const customBackends = backendOptions?.filter((item) => !item.isBuiltIn);
+
+    const options = [];
+
+    if (builtInBackends && builtInBackends.length > 0) {
+      options.push({
+        label: intl.formatMessage({ id: 'backend.builtin' }),
+        options: builtInBackends
+      });
+    }
+
+    if (customBackends && customBackends.length > 0) {
+      options.push({
+        label: intl.formatMessage({ id: 'backend.custom' }),
+        options: customBackends
+      });
+    }
+    return options;
+  }, [backendOptions, intl]);
+
   const backendVersions = useMemo(() => {
     if (!backend || backend === backendOptionsMap.custom) {
       return [];
     }
 
     // find the backend item from backendOptions
-    const backendItem = backendOptions
-      .flatMap((group) => group.options)
-      .find((item) => item.value === backend);
+    const backendItem = backendOptions.find((item) => item.value === backend);
 
     const versions = backendItem?.versions || [];
 
@@ -105,7 +125,7 @@ const BackendFields: React.FC = () => {
           onChange={onBackendChange}
           label={intl.formatMessage({ id: 'models.form.backend' })}
           description={<TooltipList list={backendTipsList}></TooltipList>}
-          options={backendOptions}
+          options={backendGroupedOptions}
           optionRender={optionRender}
           labelRender={labelRender}
         ></SealSelect>
diff --git a/src/pages/llmodels/forms/index.tsx b/src/pages/llmodels/forms/index.tsx
index 6432c10b..df8a577b 100644
--- a/src/pages/llmodels/forms/index.tsx
+++ b/src/pages/llmodels/forms/index.tsx
@@ -16,7 +16,7 @@ import {
 import { backendOptionsMap } from '../config/backend-parameters';
 import { FormContext } from '../config/form-context';
 import {
-  BackendGroupOption,
+  BackendOption,
   DeployFormKey,
   FormData,
   SourceType
@@ -188,7 +188,7 @@ const DataForm: React.FC<DataFormProps> = forwardRef((props, ref) => {
     };
   };
 
-  const updateKVCacheConfig = (backend: string, option: BackendGroupOption) => {
+  const updateKVCacheConfig = (backend: string, option: BackendOption) => {
     if (
       !option.isBuiltIn &&
       [backendOptionsMap.SGLang, backendOptionsMap.vllm].includes(backend)
@@ -202,15 +202,10 @@ const DataForm: React.FC<DataFormProps> = forwardRef((props, ref) => {
     return {};
   };
 
-  const handleBackendChange = async (
-    val: string,
-    option: BackendGroupOption
-  ) => {
-    console.log('handleBackendChange option===>', val, option);
-    const isGGUF = checkIsGGUF();
+  const handleBackendChange = async (val: string, option: BackendOption) => {
     form.setFieldsValue({
       env: null,
-      backend_version: option.default_version || '',
+      backend_version: '', // don't set default version here, let the user select it
       backend_parameters: option.default_backend_param || [],
       ...updateKVCacheConfig(val, option),
       ...updateGPUSelector(val)
diff --git a/src/pages/llmodels/forms/kv-cache.tsx b/src/pages/llmodels/forms/kv-cache.tsx
index ea65bb36..a9e1df75 100644
--- a/src/pages/llmodels/forms/kv-cache.tsx
+++ b/src/pages/llmodels/forms/kv-cache.tsx
@@ -3,7 +3,6 @@ import SealInputNumber from '@/components/seal-form/input-number';
 import { useIntl } from '@umijs/max';
 import { Form } from 'antd';
 import { useMemo, useRef } from 'react';
-import { DeployFormKeyMap } from '../config';
 import { backendOptionsMap } from '../config/backend-parameters';
 import { useFormContext } from '../config/form-context';
 import { FormData } from '../config/types';
@@ -36,9 +35,8 @@ const KVCacheForm = () => {
       (item) => item.value === backend
     );
 
-    // FIXME：remove formKey check after all built-in backends support KV cache
     return (
-      (currentBackend?.isBuiltIn || formKey === DeployFormKeyMap.CATALOG) &&
+      currentBackend?.isBuiltIn &&
       [backendOptionsMap.SGLang, backendOptionsMap.vllm].includes(
         backend as string
       )
diff --git a/src/pages/llmodels/hooks/use-query-backends.ts b/src/pages/llmodels/hooks/use-query-backends.ts
index 9084acb2..52d5629e 100644
--- a/src/pages/llmodels/hooks/use-query-backends.ts
+++ b/src/pages/llmodels/hooks/use-query-backends.ts
@@ -3,7 +3,7 @@ import { useIntl } from '@umijs/max';
 import { useAtom } from 'jotai';
 import { queryBackendList } from '../apis';
 import { backendOptionsMap } from '../config/backend-parameters';
-import { BackendGroupOption } from '../config/types';
+import { BackendOption } from '../config/types';
 
 export default function useQueryBackends() {
   const [backendOptions, setBackendOptions] = useAtom(backendOptionsAtom);
@@ -12,7 +12,7 @@ export default function useQueryBackends() {
   const getBackendOptions = async (params?: { cluster_id: number }) => {
     try {
       const res = await queryBackendList(params);
-      const list: BackendGroupOption[] = res?.items?.map((item) => {
+      const list: BackendOption[] = res?.items?.map((item) => {
         return {
           value: item.backend_name,
           label:
@@ -33,27 +33,9 @@ export default function useQueryBackends() {
           }))
         };
       });
-      const builtInBackends = list?.filter((item) => item.isBuiltIn);
-      const customBackends = list?.filter((item) => !item.isBuiltIn);
-
-      const options = [];
-
-      if (builtInBackends && builtInBackends.length > 0) {
-        options.push({
-          label: intl.formatMessage({ id: 'backend.builtin' }),
-          options: builtInBackends
-        });
-      }
-
-      if (customBackends && customBackends.length > 0) {
-        options.push({
-          label: intl.formatMessage({ id: 'backend.custom' }),
-          options: customBackends
-        });
-      }
 
       if (res?.items) {
-        setBackendOptions(options);
+        setBackendOptions(list);
       }
       return list || [];
     } catch (error) {