Update app.py

app.py

@@ -1,44 +1,17 @@
 import gradio as gr
-import torch
 from transformers import pipeline, set_seed
 
 # Setzen eines Seeds für Reproduzierbarkeit
 set_seed(42)
 
-# Laden des GPT-Modells mit Hugging Face Pipeline
-model = pipeline("text-generation", model="Loewolf/GPT_1")
+# Laden des GPT-Modells mit Hugging Face Pipeline für CPU
+model = pipeline("text-generation", model="Loewolf/GPT_1", device=-1)  # device=-1 für CPU
 tokenizer = model.tokenizer
 
 def generate_text(input_text, temp, top_k, top_p, length):
-    # Konvertieren des Eingabetextes in Token-IDs
-    input_ids = tokenizer.encode(input_text, return_tensors="pt")
-
-    # Erstellung der Attention-Mask
-    attention_mask = torch.ones(input_ids.shape, dtype=torch.bool)
-
-    # Einstellung der maximalen Länge
-    max_length = model.model.config.n_positions if len(input_ids[0]) > model.model.config.n_positions else len(input_ids[0]) + length
-
     # Textgenerierung mit spezifischen Parametern
-    beam_output = model.model.generate(
-        input_ids,
-        attention_mask=attention_mask,
-        max_length=max_length,
-        min_length=4,
-        num_beams=5,
-        no_repeat_ngram_size=2,
-        early_stopping=True,
-        temperature=temp,
-        top_p=top_p,
-        top_k=top_k,
-        length_penalty=2.0,
-        do_sample=True,
-        eos_token_id=tokenizer.eos_token_id,
-        pad_token_id=tokenizer.eos_token_id
-    )
-
-    # Konvertieren der generierten Token-IDs zurück in Text
-    return tokenizer.decode(beam_output[0], skip_special_tokens=True)
+    generated_texts = model(input_text, max_length=length, temperature=temp, top_k=top_k, top_p=top_p, num_return_sequences=1)
+    return generated_texts[0]['generated_text']
 
 def chat_with_model(user_input, history, temperature, top_k, top_p, length, system_prompt):
     combined_input = f"{history}\nNutzer: {user_input}\n{system_prompt}:"
@@ -49,15 +22,17 @@ def chat_with_model(user_input, history, temperature, top_k, top_p, length, syst
 # Erstellen der Gradio-Schnittstelle
 with gr.Blocks() as demo:
     with gr.Row():
-        history = gr.Textbox(label="Chatverlauf", lines=10, interactive=False)
-        user_input = gr.Textbox(label="Deine Nachricht")
-        system_prompt = gr.Textbox(label="System Prompt", value="Löwolf GPT")
-        with gr.Column(scale=1):
+        with gr.Column():
+            history = gr.Textbox(label="Chatverlauf", lines=10, interactive=False)
+            user_input = gr.Textbox(label="Deine Nachricht")
+            submit_btn = gr.Button("Senden")
+        with gr.Column():
+            system_prompt = gr.Textbox(label="System Prompt", value="Löwolf GPT")
             temperature = gr.Slider(minimum=0, maximum=1, step=0.01, label="Temperature", value=0.9)
             top_k = gr.Slider(minimum=0, maximum=100, step=1, label="Top K", value=50)
             top_p = gr.Slider(minimum=0, maximum=1, step=0.01, label="Top P", value=0.9)
             length = gr.Slider(minimum=1, maximum=100, step=1, label="Länge", value=20)
-        submit_btn = gr.Button("Senden")
+
     submit_btn.click(
         chat_with_model,
         inputs=[user_input, history, temperature, top_k, top_p, length, system_prompt],