fix: tabs and only use pipeline

app.py

@@ -184,8 +184,9 @@ def plot_report(title, expected, predicted, display_labels):
 
 
 def get_prompt_format(model_name):
+    formatted_text = ""
     if model_name == "TheBloke/Llama-2-13B-chat-GPTQ" or model_name== "TheBloke/Llama-2-7B-Chat-GPTQ":
-        return '''[INST] <<SYS>>
+        formatted_text = '''[INST] <<SYS>>
                 {system_message}
                 <</SYS>>
                 {prompt}[/INST]
@@ -193,13 +194,15 @@ def get_prompt_format(model_name):
                 '''
 
     if model_name == "TheBloke/meditron-7B-GPTQ" or model_name == "TheBloke/meditron-70B-GPTQ":
-        return '''<|im_start|>system
+        formatted_text = '''<|im_start|>system
                 {system_message}<|im_end|>
                 <|im_start|>user
                 {prompt}<|im_end|>
-                <|im_start|>assistant'''
+                <|im_start|>assistant
+                
+                '''
     
-    return ""
+    return formatted_text.replace("\t", "")
 
 def format_prompt(template, system_message, prompt):
     if template == "":

utils/epfl_meditron_utils.py

@@ -19,11 +19,10 @@ def get_llm_response(model_name_or_path, temperature, do_sample, top_p, top_k, m
     print("Formatted prompt:")
     print(formatted_prompt)
     
-    print("\n\n*** Generate:")
-
-    input_ids = tokenizer(formatted_prompt, return_tensors='pt').input_ids.cuda()
-    output = model.generate(inputs=input_ids, temperature=temperature, do_sample=do_sample, top_p=top_p, top_k=top_k, max_new_tokens=max_new_tokens)
-    print(tokenizer.decode(output[0], skip_special_tokens=True)) 
+    #print("\n\n*** Generate:")
+    #input_ids = tokenizer(formatted_prompt, return_tensors='pt').input_ids.cuda()
+    #output = model.generate(inputs=input_ids, temperature=temperature, do_sample=do_sample, top_p=top_p, top_k=top_k, max_new_tokens=max_new_tokens)
+    #print(tokenizer.decode(output[0], skip_special_tokens=True)) 
 
     print("*** Pipeline:")
     pipe = pipeline(