use-inference-endpoint (#2)

- Use inference endpoint (accd839142573dcfe443faf93a4f5723adfec111)
- Remove token (ac7a676452d8ac749eca96d58258b619f75be198)


Co-authored-by: Lysandre <lysandre@users.noreply.huggingface.co>
Co-authored-by: Lysandre <LysandreJik@users.noreply.huggingface.co>

front/js-src/Api.ts

@@ -15,39 +15,40 @@ export class Api {
 		// `http://coconut-proxy.huggingface.test`
 		// `http://coconuthf.eastus.cloudapp.azure.com:6006`
 		// "http://localhost:6006"
-		`https://transformer.huggingface.co`
+		// `https://transformer.huggingface.co`
+		`https://wvnbdkpga3d4aev9.us-east-1.aws.endpoints.huggingface.cloud`
 	;
 	static shared = new Api();
 	
-	private path(p: string): string {
-		return `${Api.ENDPOINT}/${p}`;
-	}
-	
 	private async postAutocomplete(
 		params: {
-			context:         string;
-			model_size?:     string; /// 'small' | 'medium',
-			top_p?:          number; /// float between 0 and 1
-			temperature?:    number; /// float between 0 and 100
-			step_size?:      number;
-			kl_scale?:       number;
-			gm_scale?:       number;
-			num_iterations?: number;
-			gen_length?:     number;
-			max_time?:       number; /// <- if we want to limit the response time. (in sec)
-			bow_or_discrim?: string;
-			use_sampling?:   boolean;
+			inputs: {
+				context:         string;
+				kwargs: {
+					model_size?:     string; /// 'small' | 'medium',
+					top_p?:          number; /// float between 0 and 1
+					temperature?:    number; /// float between 0 and 100
+					step_size?:      number;
+					kl_scale?:       number;
+					gm_scale?:       number;
+					num_iterations?: number;
+					gen_length?:     number;
+					max_time?:       number; /// <- if we want to limit the response time. (in sec)
+					bow_or_discrim?: string;
+					use_sampling?:   boolean;
+				}
+			}
 		}
-	): Promise<AutocompleteOutput> {
-		
-		const path = this.path(`autocomplete/${params.model_size || ""}`);
-		
-		const response = await fetch(path, {
+	): Promise<any> {
+		const init = {
 			method: 'POST',
 			headers: { 'Content-Type': 'application/json' },
 			body: JSON.stringify(params),
-		});
-		return await response.json() as AutocompleteOutput;
+		}
+
+		const response = await fetch(Api.ENDPOINT, init)
+		const result = JSON.parse(await response.json())
+		return result
 	}
 	
 	/**
@@ -86,21 +87,26 @@ export class Api {
 		const use_sampling = (
 			document.querySelector<HTMLInputElement>('.decoder-settings input[name=use_sampling]') || {}
 		).checked;
-		
-		return this.postAutocomplete({
-			...params,
-			model_size,
-			top_p,
-			temperature,
-			step_size,
-			kl_scale,
-			gm_scale,
-			num_iterations,
-			gen_length,
-			max_time,
-			bow_or_discrim,
-			use_sampling,
-		});
+
+		const sent_params = {
+			inputs: {
+				...params,
+				kwargs: {
+					model_size,
+					top_p,
+					temperature,
+					step_size,
+					kl_scale,
+					gm_scale,
+					num_iterations,
+					gen_length,
+					max_time,
+					bow_or_discrim,
+					use_sampling,
+				}
+			}
+		}
+		return this.postAutocomplete(sent_params);
 	}
 	
 	/**

nginx.conf

@@ -36,7 +36,7 @@ http {
         }
 
         location /autocomplete {
-            proxy_pass         $NGINX_NEURALGENV2_URL;
+            proxy_pass         https://yvkpzfc4acovo7gs.us-east-1.aws.endpoints.huggingface.cloud;
             proxy_set_header   Host      $host;
             proxy_set_header   X-Real-IP $remote_addr;
             proxy_set_header   X-Forwarded-For $proxy_add_x_forwarded_for;
@@ -47,31 +47,5 @@ http {
             add_header         Access-Control-Allow-Headers Content-Type always; # <- circumvent cors for Firefox
             # mirror /mirror_autocomplete;
         }
-
-        location ~ ^/autocomplete/(gpt2\/xl) {
-            # was turned off.
-            proxy_pass         $NGINX_AUTOCOMPLETE_URL;
-            proxy_set_header   Host      $host;
-            proxy_set_header   X-Real-IP $remote_addr;
-            proxy_set_header   X-Forwarded-For $proxy_add_x_forwarded_for;
-            proxy_set_header   X-Forwarded-Proto $scheme;
-            proxy_buffering    off;
-            proxy_http_version 1.1;
-            add_header         X-Jacqueline True always;
-            add_header         Access-Control-Allow-Headers Content-Type always; # <- circumvent cors for Firefox
-        }
-
-        location ~ ^/autocomplete/(ctrl|pplm) {
-            # was turned off.
-            proxy_pass         $NGINX_CTRL_PPLM_URL;
-            proxy_set_header   Host      $host;
-            proxy_set_header   X-Real-IP $remote_addr;
-            proxy_set_header   X-Forwarded-For $proxy_add_x_forwarded_for;
-            proxy_set_header   X-Forwarded-Proto $scheme;
-            proxy_buffering    off;
-            proxy_http_version 1.1;
-            add_header         X-Jacinthe True always;
-            add_header         Access-Control-Allow-Headers Content-Type always; # <- circumvent cors for Firefox
-        }
     }
 }