switch from Inference API to local model with tokenizer (openai-community/gpt2)

app.py

@@ -1,57 +1,27 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
-import os
 import re
-#from transformers import Qwen2_5OmniForConditionalGeneration
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
 
-#qwenModel = Qwen2_5OmniForConditionalGeneration.from_pretrained(
-#    "Qwen/Qwen2.5-Omni-7B",
-#    device_map="auto",
-#    torch_dtype=torch.bfloat16,
-#    attn_implementation="flash_attention_2",
-#)
+# Modell und Tokenizer laden
+tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")
+model = AutoModelForCausalLM.from_pretrained("openai-community/gpt2")
 
-# Load API key from environment variables
-#HF_API_TOKEN = os.getenv("HUG_TOKEN_READ2")
-HF_INF_KEY = os.getenv("HF_INF_PROVIDERS_TOKEN")
-#HF_INF_KEY = os.getenv("INF_NEBIUS")
-
-# Hugging Face Inference API Client
-#client = InferenceClient(model="mistralai/Mistral-7B-Instruct-v0.1", token=HF_API_TOKEN)
-#client = InferenceClient(model="openGPT-X/Teuken-7B-instruct-commercial-v0.4", token=HF_API_TOKEN)
-#client = InferenceClient(model=qwenModel, token=HF_API_TOKEN)
-#client = InferenceClient(model="microsoft/Phi-4-mini-instruct", token=HF_API_TOKEN)
-#client = InferenceClient(model="openai-community/gpt2", token=HF_API_TOKEN)
-client = InferenceClient(
-    provider="nebius",
-    api_key=HF_INF_KEY
-)
-
-# Function to translate text into emojis
 def text_to_emoji(text):
     # remove special characters
     text_cleaned = re.sub(r"[.,!?;:]", "", text)
-    
-    prompt = f"Convert this sentence into an emoji-sequence which conveys a similar meaning and return only the emojis, no explanation:\n\n\"{text_cleaned}\""
-    #response = client.text_generation(prompt, max_new_tokens=50)
-    #return response
+    prompt = f"Convert the following sentence into an emoji-sequence which conveys a similar meaning and return only the emojis, no explanation:\n\n\"{text_cleaned}\""
 
-    completion = client.chat.completions.create(
-        model="mistralai/Mistral-Small-3.1-24B-Instruct-2503",
-        messages=[
-            {
-                "role": "user",
-                "content": [
-                    {
-                        "type": "text",
-                        "text": prompt
-                    }
-                ]
-            }
-        ],
-        max_tokens=25
-    )
-    return completion.choices[0].message
+    # Tokenisieren
+    inputs = tokenizer(prompt, return_tensors="pt")
+    
+    # Antwort generieren
+    outputs = model.generate(**inputs, max_new_tokens=25, do_sample=True, temperature=0.7)
+    
+    # Antwort decodieren
+    result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    return result
 
 # Gradio UI
 iface = gr.Interface(