Test

app.py

@@ -1,48 +1,28 @@
-import os
+# app.py
+
 import gradio as gr
-from run_llm import main
-import argparse
+from run_llm import run_llm_interface
 
 theme = gr.themes.Soft()
 
-def run_llm_interface(model, task, sentence):
-    args = argparse.Namespace(
-        model_path=model,
-        prompt=task,
-        start=0,
-        end=1  # Set to 1 to process a single sentence
-    )
-    main(args)
-    
-    # Create directories if they don't exist
-    os.makedirs(f'result/prompt1_qa/{model}/ptb/per_ent/NOUN', exist_ok=True)
-    os.makedirs(f'result/prompt2_instruction/chunking/{model}/ptb', exist_ok=True)
-    os.makedirs(f'result/prompt3_structured_prompt/chunking/{model}/ptb', exist_ok=True)
-    
-    # Read the outputs from the result files
-    with open(f'result/prompt1_qa/{model}/ptb/per_ent/NOUN/0.txt', 'r') as f:
-        output_1 = f.read()
-    
-    with open(f'result/prompt2_instruction/chunking/{model}/ptb/0.txt', 'r') as f:
-        output_2 = f.read()
-    
-    with open(f'result/prompt3_structured_prompt/chunking/{model}/ptb/0.txt', 'r') as f:
-        output_3 = f.read()
-
-    return {"output_1": output_1, "output_2": output_2, "output_3": output_3}
+# 3 inputs:
+# - An input text which will be a random string
+# - First dropdown to select the task (POS, Chunking, Parsing)
+# - Second dropdown select the model type
+# use run_llm.py to feed the models and then output 3 results in 3 output boxes, one for each strategy (strategy 1, 2 and 3)
 
 # Define example instructions for testing
 instruction_examples = [
-    ["gpt3.5", "POS Tagging", "Describe the origin of the universe"],
-    ["vicuna-7b", "Chunking", "Explain the concept of artificial intelligence"],
-    ["fastchat-t5", "Parsing", "Describe the most common types of cancer"],
+    ["Describe the origin of the universe"],
+    ["Explain the concept of artificial intelligence"],
+    ["Describe the most common types of cancer"],
 ]
 
 with gr.Interface(
     fn=run_llm_interface,
     inputs=[
-        gr.Dropdown(['gpt3.5', 'vicuna-7b', 'vicuna-13b', 'fastchat-t5', 'llama-7b', 'llama-13b', 'llama-30b', 'alpaca'], label="Select Model", default='gpt3.5', key="model"),
-        gr.Dropdown(['POS Tagging', 'Chunking', 'Parsing'], label="Select Task", default='POS Tagging', key="task"),
+        gr.Dropdown(['gpt3.5', 'vicuna-7b', 'vicuna-13b', 'fastchat-t5', 'llama-7b', 'llama-13b', 'llama-30b', 'alpaca'], label="Select Model", default='gpt3.5', key="model_path"),
+        gr.Dropdown(['POS Tagging', 'Chunking', 'Parsing'], label="Select Task", default='POS Tagging', key="prompt"),
         gr.Textbox("", label="Enter Sentence", key="sentence", placeholder="Enter a sentence..."),
     ],
     outputs=[
@@ -50,7 +30,7 @@ with gr.Interface(
         gr.Textbox("", label="Strategy 2 Output", key="output_2", readonly=True),
         gr.Textbox("", label="Strategy 3 Output", key="output_3", readonly=True),
     ],
-    examples=instruction_examples, 
+    examples=instruction_examples,
     live=False,
     title="LLM Evaluator with Linguistic Scrutiny",
     theme=theme

run_llm.py

@@ -426,9 +426,9 @@ def fastchat(prompt, model, tokenizer):
         output_ids, skip_special_tokens=True, spaces_between_special_tokens=False
     )
 
-    print('Empty system message')
-    print(f"{conv.roles[0]}: {msg}")
-    print(f"{conv.roles[1]}: {outputs}")
+    #print('Empty system message')
+    #print(f"{conv.roles[0]}: {msg}")
+    #print(f"{conv.roles[1]}: {outputs}")
 
     return outputs
 
@@ -446,6 +446,25 @@ def gpt3(prompt):
 
         return None
 
+def run_llm_interface(model_path, prompt, sentence):
+    import argparse
+    from run_llm import main
+
+    # Construct arguments
+    args = argparse.Namespace(
+        model_path=model_path,
+        temperature=0.7,
+        repetition_penalty=1.0,
+        max_new_tokens=512,
+        debug=False,
+        message="Hello! Who are you?",
+        start=0,
+        end=1000,
+        prompt=prompt,
+    )
+
+    # Run the main function
+    main(args=args)
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()