t

app.py

@@ -33,6 +33,8 @@ top_p = gr.inputs.Slider(minimum=0, maximum=1.0,
 
 generator = pipeline('text-generation', model='gpt2')
 
+gpt_j_generator = pipeline('text-generation', model='GPT-J 6B')
+
 
 title = "GPT-J-6B"
 
@@ -99,24 +101,30 @@ def f(context, temperature, top_p, max_length, model_idx, SPACE_VERIFICATION_KEY
         # maybe try "0" instead or 1, or "1"
         # use GPT-J-6B
         if model_idx == 0:
-            if main_gpt_j_api_up:
-                # for this api, a length of > 250 instantly errors, so use a while loop or something
-                # that would fetch results in chunks of 250
-                # NOTE change so it uses previous generated input every time
-                generated_text = context #""
-                while (max_length > 0):
-                    payload = {"inputs": generated_text, "parameters": {"max_new_tokens": max_length>250 and 250 or max_length, "temperature": temperature, "top_p": top_p}}
-                    response = requests.request("POST", API_URL, data=json.dumps(payload), headers=headers)
-                    context = json.loads(response.content.decode("utf-8"))#[0]['generated_text']
-                    # context = get_generated_text(generated_context)
-                    # handle inconsistent inference API
-                    if 'generated_text' in context[0]:
-                        context = context[0]['generated_text']
-                    else:
-                        context = context[0][0]['generated_text']
+            # just use regular pipeline models man leave APIs
+
+            set_seed(2**31)
+            generated_text = gpt_j_generator(context, max_length=896, max_new_tokens=max_length, top_p=top_p, temperature=temperature, num_return_sequences=1)
+
+            return get_generated_text(generated_text)
+            # if main_gpt_j_api_up:
+            #     # for this api, a length of > 250 instantly errors, so use a while loop or something
+            #     # that would fetch results in chunks of 250
+            #     # NOTE change so it uses previous generated input every time
+            #     generated_text = context #""
+            #     while (max_length > 0):
+            #         payload = {"inputs": generated_text, "parameters": {"max_new_tokens": max_length>250 and 250 or max_length, "temperature": temperature, "top_p": top_p}}
+            #         response = requests.request("POST", API_URL, data=json.dumps(payload), headers=headers)
+            #         context = json.loads(response.content.decode("utf-8"))#[0]['generated_text']
+            #         # context = get_generated_text(generated_context)
+            #         # handle inconsistent inference API
+            #         if 'generated_text' in context[0]:
+            #             context = context[0]['generated_text']
+            #         else:
+            #             context = context[0][0]['generated_text']
                         
-                    generated_text += context
-                    max_length -= 250
+            #         generated_text += context
+            #         max_length -= 250
 
                 # payload = {"inputs": context, "parameters":{
                 #     "max_new_tokens":max_length, "temperature":temperature, "top_p":top_p}}