Standardize llama.cpp endpoint with baseURL (#1305)

* Standardize llama.cpp endpoint with `baseURL`

* cleaner

src/lib/server/endpoints/llamacpp/endpointLlamacpp.ts

@@ -9,7 +9,8 @@ export const endpointLlamacppParametersSchema = z.object({
 	weight: z.number().int().positive().default(1),
 	model: z.any(),
 	type: z.literal("llamacpp"),
-	url: z.string().url().default("http://127.0.0.1:8080"),
+	url: z.string().url().default("http://127.0.0.1:8080"), // legacy, feel free to remove in breaking change update
+	baseURL: z.string().url().optional(),
 	accessToken: z
 		.string()
 		.min(1)
@@ -19,7 +20,7 @@ export const endpointLlamacppParametersSchema = z.object({
 export function endpointLlamacpp(
 	input: z.input<typeof endpointLlamacppParametersSchema>
 ): Endpoint {
-	const { url, model } = endpointLlamacppParametersSchema.parse(input);
+	const { baseURL, url, model } = endpointLlamacppParametersSchema.parse(input);
 	return async ({ messages, preprompt, continueMessage, generateSettings }) => {
 		const prompt = await buildPrompt({
 			messages,
@@ -30,7 +31,7 @@ export function endpointLlamacpp(
 
 		const parameters = { ...model.parameters, ...generateSettings };
 
-		const r = await fetch(`${url}/completion`, {
+		const r = await fetch(`${baseURL ?? url}/completion`, {
 			method: "POST",
 			headers: {
 				"Content-Type": "application/json",