refactor: use return full text (#3)

* feat: add completion context

* feat: remove unnecessary completion context

src/main.rs

@@ -38,10 +38,33 @@ struct FimParams {
     suffix: String,
 }
 
+#[derive(Debug, Serialize)]
+struct APIParams {
+    max_new_tokens: u32,
+    temperature: f32,
+    do_sample: bool,
+    top_p: f32,
+    stop: Vec<String>,
+    return_full_text: bool,
+}
+
+impl From<RequestParams> for APIParams {
+    fn from(params: RequestParams) -> Self {
+        Self {
+            max_new_tokens: params.max_new_tokens,
+            temperature: params.temperature,
+            do_sample: params.do_sample,
+            top_p: params.top_p,
+            stop: vec![params.stop_token.clone()],
+            return_full_text: false,
+        }
+    }
+}
+
 #[derive(Serialize)]
 struct APIRequest {
     inputs: String,
-    parameters: RequestParams,
+    parameters: APIParams,
 }
 
 #[derive(Debug, Deserialize)]
@@ -185,7 +208,7 @@ async fn request_completion(
 ) -> Result<Vec<Generation>> {
     let mut req = http_client.post(model).json(&APIRequest {
         inputs: prompt,
-        parameters: request_params,
+        parameters: request_params.into(),
     });
 
     if let Some(api_token) = api_token.clone() {
@@ -200,15 +223,11 @@ async fn request_completion(
     }
 }
 
-fn parse_generations(
-    generations: Vec<Generation>,
-    prompt: &str,
-    stop_token: &str,
-) -> Vec<Completion> {
+fn parse_generations(generations: Vec<Generation>, stop_token: &str) -> Vec<Completion> {
     generations
         .into_iter()
         .map(|g| Completion {
-            generated_text: g.generated_text.replace(prompt, "").replace(stop_token, ""),
+            generated_text: g.generated_text.replace(stop_token, ""),
         })
         .collect()
 }
@@ -243,7 +262,7 @@ impl Backend {
         )
         .await?;
 
-        Ok(parse_generations(result, &prompt, &stop_token))
+        Ok(parse_generations(result, &stop_token))
     }
 }