added api url

app.py

@@ -5,28 +5,44 @@ from prompts import PROMPT_LIST
 import streamlit as st
 import random
 
-backend = "http://fastapi:8000/generate"
+headers = {"Authorization": f"Bearer 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"}
+MODELS = {
+    "GPT-2 Small": {
+        "url": "https://api-inference.huggingface.co/models/flax-community/gpt2-small-indonesian"
+    },
+    "GPT-2 Medium": {
+        "url": "https://api-inference.huggingface.co/models/flax-community/gpt2-medium-indonesian"
+    },
+}
+
+
+def query(payload, model_name):
+    data = json.dumps(payload)
+    response = requests.request("POST", MODELS[model_name]["url"], headers=headers, data=data)
+    return json.loads(response.content.decode("utf-8"))
+
 
 def process(text: str,
             model_name: str,
             max_len: int,
             temp: float,
             top_k: int,
-            top_p: float,
-            do_sample: bool,
-            server_url: str):
-    payload = {"text": text,
-               "max_len": max_len,
-               "temp": temp,
-               "top_k": top_k,
-               "top_p": top_p,
-               "do_sample": do_sample,
-               "model_name": model_name}
-    r = requests.post(
-        server_url, data=json.dumps(payload), headers={"Content-Type": "application/json"}, timeout=8000
-    )
-
-    return json.loads(r.text)
+            top_p: float):
+
+    payload = {
+        "inputs": text,
+        "parameters": {
+            "max_new_tokens": max_len,
+            "top_k": top_k,
+            "top_p": top_p,
+            "temperature": temp,
+            "repetition_penalty": 2.0,
+        },
+        "options": {
+            "use_cache": True,
+        }
+    }
+    return query(payload, model_name)
 
 st.set_page_config(page_title="Indonesian GPT-2 Demo")
 
@@ -86,9 +102,8 @@ if st.button("Run"):
                          max_len=max_len,
                          temp=temp,
                          top_k=top_k,
-                         top_p=top_p,
-                         do_sample=do_sample,
-                         server_url=backend)
-        st.write(result["result"].replace("\n", "  \n"))
+                         top_p=top_p)[0]["generated_text"]
+
+        st.write(result.replace("\n", "  \n"))
         st.text("English translation")
-        st.write(translate(result["result"], "en", "id").replace("\n", "  \n"))
+        st.write(translate(result, "en", "id").replace("\n", "  \n"))