debuggin llama

llama.py

@@ -22,28 +22,28 @@ following this structure: 'Der Kunde ... und erwartet ...'. The summaries need t
 def load_llama_model():
     """Load Llama model and tokenizer with optimized settings"""
     tokenizer = AutoTokenizer.from_pretrained("Walid777/llama3-8b-emails-summarization")
-    model = AutoModelForCausalLM.from_pretrained(
-        "Walid777/llama3-8b-emails-summarization",
-        device_map="auto",
-        torch_dtype="auto"
-    )
+    model = AutoModelForCausalLM.from_pretrained( "Walid777/llama3-8b-emails-summarization")
     return model, tokenizer
 
 def generate_llama_summary(email, model, tokenizer, prompt_template):
     """Generate summary using structured prompt template"""
-    formatted_prompt = prompt_template.format(email)
+    formatted_prompt = prompt_template.format(email, "")
     
     inputs = tokenizer(
-        formatted_prompt,
+        [formatted_prompt],
         return_tensors="pt"
-    ).to(model.device)
+    ).to("cuda")
     
     outputs = model.generate(
         **inputs,
-        max_new_tokens=128,
-        temperature=0.7,
-        pad_token_id=tokenizer.eos_token_id
+        max_new_tokens=128
     )
     
-    full_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
-    return full_text.split("### Summary:")[-1].strip()
+    summary = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    if "### Summary:" in summary:
+        summary = summary.split("### Summary:")[-1].strip()
+    else:
+        summary = "Error: Could not extract summary"
+    return summary
+
+