feat(ollama-adapter): add keepAlive configuration for persistent conn…

…ections, fix #356
ChatLunaLab · Jan 12, 2025 · 6b08e9b · 6b08e9b
1 parent adcf1da
commit 6b08e9b
Show file tree

Hide file tree

Showing 5 changed files with 9 additions and 3 deletions.
diff --git a/packages/ollama-adapter/src/index.ts b/packages/ollama-adapter/src/index.ts
@@ -38,6 +38,7 @@ export interface Config extends ChatLunaPlugin.Config {
     presencePenalty: number
     frequencyPenalty: number
     supportImage: boolean
+    keepAlive: boolean
 }
 
 export const Config: Schema<Config> = Schema.intersect([
@@ -46,7 +47,8 @@ export const Config: Schema<Config> = Schema.intersect([
         apiEndpoints: Schema.array(Schema.string()).default([
             'http://127.0.0.1:11434'
         ]),
-        supportImage: Schema.boolean().default(true)
+        supportImage: Schema.boolean().default(true),
+        keepAlive: Schema.boolean().default(true)
     }),
     Schema.object({
         maxTokens: Schema.number().min(16).max(4096).step(16).default(1024),

diff --git a/packages/ollama-adapter/src/locales/en-US.schema.yml b/packages/ollama-adapter/src/locales/en-US.schema.yml
@@ -2,8 +2,9 @@ $inner:
   - {}
   - $desc: 'API Configuration'
     apiEndpoints:
-      $desc: 'Ollama API endpoints'
-    supportImage: 'Whether the current model supports multi-modal image input'
+        $desc: 'Ollama API endpoints'
+        supportImage: 'Whether the current model supports multi-modal image input'
+        keepAlive: 'Whether to keep the connection alive and keep the model loaded'
 
   - $desc: 'Model Parameters'
     maxTokens: 'Max input tokens (16-4096, multiple of 16). Note: >2000 for 8k+ token models'

diff --git a/packages/ollama-adapter/src/locales/zh-CN.schema.yml b/packages/ollama-adapter/src/locales/zh-CN.schema.yml
@@ -4,6 +4,7 @@ $inner:
     apiEndpoints:
       $desc: Ollama 的请求地址列表。
     supportImage: 目前运行的模型是否支持多模态图片输入。
+    keepAlive: 是否保持长连接，不卸载模型。
 
   - $desc: 模型配置
     maxTokens: 输入的最大上下文 Token（16~4096，必须是 16 的倍数）。注意：仅当您使用的模型最大 Token 为 8000 及以上时，才建议设置超过 2000 token。

diff --git a/packages/ollama-adapter/src/requester.ts b/packages/ollama-adapter/src/requester.ts
@@ -46,6 +46,7 @@ export class OllamaRequester
                         params.input,
                         this._plugin.config.supportImage
                     ),
+                    keep_alive: this._plugin.config.keepAlive ? -1 : undefined,
                     options: {
                         temperature: params.temperature,
                         // top_k: params.n,

diff --git a/packages/ollama-adapter/src/types.ts b/packages/ollama-adapter/src/types.ts
@@ -6,6 +6,7 @@ export interface OllamaRequest {
         top_p: number
         stop: string
     }
+    keep_alive?: number
     messages: OllamaMessage[]
     stream: boolean
 }