feat: add extended_kv_cache

2025-10-14 20:09:43 +08:00
parent 620c113fae
commit 83403a1a1a
13 changed files with 277 additions and 142 deletions
@@ -76,6 +76,12 @@ export interface FormData {
  optimize_long_prompt: boolean;
  enable_speculative_decoding: boolean;
  cluster_id: number;
+  extended_kv_cache: {
+    enabled: boolean;
+    chunk_size: number;
+    max_local_cpu_size: number;
+    remote_url: string;
+  };
 }

 interface ComputedResourceClaim {
@@ -16,6 +16,7 @@ import { useFormContext } from '../config/form-context';
 import { FormData } from '../config/types';
 import Backend from '../forms/backend';
 import BackendParametersList from '../forms/backend-parameters-list';
+import ScheduleTypeForm from '../forms/schedule-type';

 const placementStrategyTips = [
  {
@@ -88,6 +89,7 @@ const AdvanceConfig = () => {
          options={modelCategories}
        ></SealSelect>
      </Form.Item>
+      <ScheduleTypeForm></ScheduleTypeForm>
      <Backend></Backend>
      {scheduleType === ScheduleValueMap.Auto && (
        <>
@@ -169,7 +171,7 @@ const AdvanceConfig = () => {
        [backendOptionsMap.vllm, backendOptionsMap.ascendMindie].includes(
          backend
        ) && (
-          <div style={{ paddingBottom: 22, paddingLeft: 10 }}>
+          <div style={{ paddingBottom: 22 }}>
            <Form.Item<FormData>
              name="distributed_inference_across_workers"
              valuePropName="checked"
@@ -187,7 +189,7 @@ const AdvanceConfig = () => {
            </Form.Item>
          </div>
        )}
-      <div style={{ paddingBottom: 22, paddingLeft: 10 }}>
+      <div style={{ paddingBottom: 22 }}>
        <Form.Item<FormData>
          name="restart_on_error"
          valuePropName="checked"
@@ -191,6 +191,12 @@ const DataForm: React.FC<DataFormProps> = forwardRef((props, ref) => {
          categories: null,
          restart_on_error: true,
          distributed_inference_across_workers: true,
+          extended_kv_cache: {
+            enabled: false,
+            chunk_size: 256,
+            max_local_cpu_size: 10,
+            remote_url: ''
+          },
          ...initialValues
        }}
      >
@@ -254,6 +260,28 @@ const DataForm: React.FC<DataFormProps> = forwardRef((props, ref) => {
            ></SealSelect>
          }
        </Form.Item>
+        <Form.Item<FormData>
+          name="replicas"
+          rules={[
+            {
+              required: true,
+              message: getRuleMessage('input', 'models.form.replicas')
+            }
+          ]}
+        >
+          <SealInput.Number
+            style={{ width: '100%' }}
+            label={intl.formatMessage({
+              id: 'models.form.replicas'
+            })}
+            required
+            description={intl.formatMessage(
+              { id: 'models.form.replicas.tips' },
+              { api: `${window.location.origin}/v1` }
+            )}
+            min={0}
+          ></SealInput.Number>
+        </Form.Item>
        <CatalogFrom></CatalogFrom>
        <Form.Item<FormData> name="description">
          <SealInput.TextArea
@@ -0,0 +1,60 @@
+import CheckboxField from '@/components/seal-form/checkbox-field';
+import SealInputNumber from '@/components/seal-form/input-number';
+import SealInput from '@/components/seal-form/seal-input';
+import { useIntl } from '@umijs/max';
+import { Form } from 'antd';
+import { FormData } from '../config/types';
+
+const KVCacheForm = () => {
+  const intl = useIntl();
+  const form = Form.useFormInstance();
+  const kvCacheEnabled = Form.useWatch(['extended_kv_cache', 'enabled'], form);
+
+  return (
+    <>
+      <div style={{ paddingBottom: 22 }}>
+        <Form.Item<FormData>
+          name={['extended_kv_cache', 'enabled']}
+          valuePropName="checked"
+          style={{ padding: '0 10px', marginBottom: 0 }}
+          noStyle
+        >
+          <CheckboxField
+            label={intl.formatMessage({ id: 'models.form.extendedkvcache' })}
+          ></CheckboxField>
+        </Form.Item>
+      </div>
+      {kvCacheEnabled && (
+        <>
+          <Form.Item<FormData>
+            name={['extended_kv_cache', 'max_local_cpu_size']}
+          >
+            <SealInputNumber
+              label={intl.formatMessage({ id: 'models.form.maxCPUSize' })}
+              min={0}
+              step={1}
+              precision={0}
+            />
+          </Form.Item>
+          <Form.Item<FormData> name={['extended_kv_cache', 'chunk_size']}>
+            <SealInputNumber
+              label={intl.formatMessage({ id: 'models.form.chunkSize' })}
+              min={0}
+              step={1}
+            />
+          </Form.Item>
+          <Form.Item<FormData> name={['extended_kv_cache', 'remote_url']}>
+            <SealInput.Input
+              label={intl.formatMessage({ id: 'models.form.remoteURL' })}
+              min={0}
+              step={1}
+              placeholder="protocol://host:port"
+            />
+          </Form.Item>
+        </>
+      )}
+    </>
+  );
+};
+
+export default KVCacheForm;
@@ -1,143 +1,15 @@
-import SealCascader from '@/components/seal-form/seal-cascader';
-import SealInput from '@/components/seal-form/seal-input';
-import SealSelect from '@/components/seal-form/seal-select';
-import TooltipList from '@/components/tooltip-list';
-import useAppUtils from '@/hooks/use-app-utils';
 import { useIntl } from '@umijs/max';
 import { Form } from 'antd';
 import React from 'react';
-import GPUCard from '../components/gpu-card';
-import { scheduleList, ScheduleValueMap } from '../config';
-import { backendOptionsMap } from '../config/backend-parameters';
-import { useCatalogFormContext, useFormContext } from '../config/form-context';
-import { FormData } from '../config/types';
-
-const scheduleTypeTips = [
-  {
-    title: {
-      text: 'models.form.scheduletype.auto',
-      locale: true
-    },
-    tips: 'models.form.scheduletype.auto.tips'
-  },
-  {
-    title: {
-      text: 'models.form.scheduletype.manual',
-      locale: true
-    },
-    tips: 'models.form.scheduletype.manual.tips'
-  }
-];
+import KVCacheForm from './kv-cache';

 const Performance: React.FC = () => {
  const intl = useIntl();
-  const { onValuesChange, gpuOptions } = useFormContext();
-  const { onQuantizationChange } = useCatalogFormContext();
-  const { getRuleMessage } = useAppUtils();
  const form = Form.useFormInstance();
-  const scheduleType = Form.useWatch('scheduleType', form);
-
-  const handleScheduleTypeChange = (value: string) => {
-    if (value === ScheduleValueMap.Auto) {
-      onValuesChange?.({}, form.getFieldsValue());
-    }
-  };
-
-  const handleOnQuantizationChange = (val: any) => {
-    onQuantizationChange?.(val);
-  };
-
-  const handleBeforeGpuSelectorChange = (gpuIds: any[]) => {};
-
-  const handleGpuSelectorChange = (value: any[]) => {
-    handleBeforeGpuSelectorChange(value);
-    onValuesChange?.({}, form.getFieldsValue());
-  };

  return (
    <>
-      <Form.Item<FormData>
-        name="replicas"
-        rules={[
-          {
-            required: true,
-            message: getRuleMessage('input', 'models.form.replicas')
-          }
-        ]}
-      >
-        <SealInput.Number
-          style={{ width: '100%' }}
-          label={intl.formatMessage({
-            id: 'models.form.replicas'
-          })}
-          required
-          description={intl.formatMessage(
-            { id: 'models.form.replicas.tips' },
-            { api: `${window.location.origin}/v1` }
-          )}
-          min={0}
-        ></SealInput.Number>
-      </Form.Item>
-      <Form.Item name="scheduleType">
-        <SealSelect
-          onChange={handleScheduleTypeChange}
-          label={intl.formatMessage({ id: 'models.form.scheduletype' })}
-          description={<TooltipList list={scheduleTypeTips}></TooltipList>}
-          options={scheduleList}
-        ></SealSelect>
-      </Form.Item>
-      {scheduleType === ScheduleValueMap.SpecificGPUType && (
-        <>
-          <Form.Item name={['gpu_selector', 'gpu_type']}>
-            <SealSelect
-              label={intl.formatMessage({ id: 'models.form.gpuType' })}
-              options={[]}
-            ></SealSelect>
-          </Form.Item>
-          <Form.Item name={['gpu_selector', 'gpu_count']}>
-            <SealSelect
-              label={intl.formatMessage({ id: 'models.form.gpuCount' })}
-              options={[]}
-            ></SealSelect>
-          </Form.Item>
-        </>
-      )}
-      {scheduleType === ScheduleValueMap.Manual &&
-        !form.getFieldValue('fix_gpu_type') && (
-          <>
-            <Form.Item
-              name={['gpu_selector', 'gpu_ids']}
-              rules={[
-                {
-                  required: true,
-                  message: getRuleMessage('select', 'models.form.gpuselector')
-                }
-              ]}
-            >
-              <SealCascader
-                required
-                showSearch
-                expandTrigger="hover"
-                multiple={
-                  form.getFieldValue('backend') !== backendOptionsMap.voxBox
-                }
-                classNames={{
-                  popup: {
-                    root: 'cascader-popup-wrapper gpu-selector'
-                  }
-                }}
-                maxTagCount={1}
-                label={intl.formatMessage({ id: 'models.form.gpuselector' })}
-                options={gpuOptions}
-                showCheckedStrategy="SHOW_CHILD"
-                value={form.getFieldValue(['gpu_selector', 'gpu_ids'])}
-                optionNode={GPUCard}
-                getPopupContainer={(triggerNode) => triggerNode.parentNode}
-                onChange={handleGpuSelectorChange}
-              ></SealCascader>
-            </Form.Item>
-          </>
-        )}
+      <KVCacheForm></KVCacheForm>
      {/* <div style={{ paddingBottom: 22, paddingLeft: 10 }}>
        <Form.Item<FormData>
          name="optimize_long_prompt"
@@ -0,0 +1,124 @@
+import SealCascader from '@/components/seal-form/seal-cascader';
+import SealSelect from '@/components/seal-form/seal-select';
+import TooltipList from '@/components/tooltip-list';
+import useAppUtils from '@/hooks/use-app-utils';
+import { useIntl } from '@umijs/max';
+import { Form } from 'antd';
+import React from 'react';
+import GPUCard from '../components/gpu-card';
+import { scheduleList, ScheduleValueMap } from '../config';
+import { backendOptionsMap } from '../config/backend-parameters';
+import { useCatalogFormContext, useFormContext } from '../config/form-context';
+
+const scheduleTypeTips = [
+  {
+    title: {
+      text: 'models.form.scheduletype.auto',
+      locale: true
+    },
+    tips: 'models.form.scheduletype.auto.tips'
+  },
+  {
+    title: {
+      text: 'models.form.scheduletype.manual',
+      locale: true
+    },
+    tips: 'models.form.scheduletype.manual.tips'
+  }
+];
+
+const ScheduleTypeForm: React.FC = () => {
+  const intl = useIntl();
+  const { onValuesChange, gpuOptions } = useFormContext();
+  const { onQuantizationChange } = useCatalogFormContext();
+  const { getRuleMessage } = useAppUtils();
+  const form = Form.useFormInstance();
+  const scheduleType = Form.useWatch('scheduleType', form);
+
+  const handleScheduleTypeChange = (value: string) => {
+    if (value === ScheduleValueMap.Auto) {
+      onValuesChange?.({}, form.getFieldsValue());
+    }
+  };
+
+  const handleOnQuantizationChange = (val: any) => {
+    onQuantizationChange?.(val);
+  };
+
+  const handleBeforeGpuSelectorChange = (gpuIds: any[]) => {};
+
+  const handleGpuSelectorChange = (value: any[]) => {
+    handleBeforeGpuSelectorChange(value);
+    onValuesChange?.({}, form.getFieldsValue());
+  };
+
+  return (
+    <>
+      <Form.Item name="scheduleType">
+        <SealSelect
+          onChange={handleScheduleTypeChange}
+          label={intl.formatMessage({ id: 'models.form.scheduletype' })}
+          description={<TooltipList list={scheduleTypeTips}></TooltipList>}
+          options={scheduleList}
+        ></SealSelect>
+      </Form.Item>
+      {scheduleType === ScheduleValueMap.SpecificGPUType && (
+        <>
+          <Form.Item name={['gpu_selector', 'gpu_type']}>
+            <SealSelect
+              label={intl.formatMessage({ id: 'models.form.gpuType' })}
+              options={[]}
+            ></SealSelect>
+          </Form.Item>
+          <Form.Item name={['gpu_selector', 'gpu_count']}>
+            <SealSelect
+              label={intl.formatMessage({ id: 'models.form.gpuCount' })}
+              options={[]}
+            ></SealSelect>
+          </Form.Item>
+        </>
+      )}
+      {scheduleType === ScheduleValueMap.Manual &&
+        !form.getFieldValue('fix_gpu_type') && (
+          <>
+            <Form.Item
+              name={['gpu_selector', 'gpu_ids']}
+              rules={[
+                {
+                  required: true,
+                  message: getRuleMessage('select', 'models.form.gpuselector')
+                }
+              ]}
+            >
+              <SealCascader
+                required
+                showSearch
+                expandTrigger="hover"
+                multiple={
+                  form.getFieldValue('backend') !== backendOptionsMap.voxBox
+                }
+                classNames={{
+                  popup: {
+                    root: 'cascader-popup-wrapper gpu-selector'
+                  }
+                }}
+                maxTagCount={1}
+                label={intl.formatMessage({ id: 'models.form.gpuselector' })}
+                options={gpuOptions}
+                showCheckedStrategy="SHOW_CHILD"
+                value={form.getFieldValue(['gpu_selector', 'gpu_ids'])}
+                optionNode={GPUCard}
+                getPopupContainer={(triggerNode) => triggerNode.parentNode}
+                notFoundContent={intl.formatMessage({
+                  id: 'models.form.gpus.notfound'
+                })}
+                onChange={handleGpuSelectorChange}
+              ></SealCascader>
+            </Form.Item>
+          </>
+        )}
+    </>
+  );
+};
+
+export default ScheduleTypeForm;