Added files and test

app.py

@@ -1,22 +1,38 @@
 import gradio as gr
 from transformers import pipeline
+import argparse
+from run_llm import main
 
 # Initialize the GPT-2 pipeline
 pipe = pipeline("text-generation", model="gpt2")
 
 theme = gr.themes.Soft()
 
+
+
+
 # Function that generates text based on instruction-based prompting
-def generate_text(input_instruction):
-    # Use the input instruction to generate text
-    generated_text = pipe(input_instruction, max_length=500)[0]['generated_text']
+#def generate_text(input_instruction):
+#    # Use the input instruction to generate text
+#    generated_text = pipe(input_instruction, max_length=500)[0]['generated_text']
+#    return generated_text
+
+# Create a function that takes 3 inputs:
+# - A prompt which will be a random string
+# - From the first dropdown select the task (1,2,3)
+# - From the 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)
+
+def generate_text(prompt, task_number, model_type):
+    generated_text = pipe(prompt, max_length=500)[0]['generated_text']
     return generated_text
 
+
 # Define example instructions for testing
 instruction_examples = [
-    ("Describe the origin of the universe"),
-    ("Explain the concept of artificial intelligence"),
-    ("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"],
 ]
 
 # Function that echoes the input text
@@ -25,45 +41,19 @@ instruction_examples = [
 
 with gr.Interface(
     fn=generate_text,
-    inputs=gr.Textbox(placeholder="Enter text here..."),
-    outputs=gr.Textbox(),
+    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.Textbox("", label="Enter Sentence", key="sentence", placeholder="Enter a sentence..."),
+    ],
+    outputs=[
+        gr.Textbox("", label="Strategy 1 Output", key="output_1", readonly=True),
+        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, 
     live=False,
     title="LLM Evaluator with Linguistic Scrutiny",
     theme=theme
 ) as iface:
-    blocks = gr.Blocks()
-    
-    with gr.Row():
-        vicuna_model_selector = gr.Dropdown(["7b", "13b", "33b"], label="Vicuna Model", placeholder="Select model size")
-        llama_model_selector = gr.Dropdown(["7B", "13B", "30B", "65B"], label="LLaMa Model", placeholder="Select model size")
-        chatgpt_api_key = gr.Textbox(label="ChatGPT API Key", type="password", placeholder="Enter your API key")
-
-    # Strategy 1 - QA-Based Prompting
-    with gr.Accordion("Strategy 1 - QA-Based Prompting", style="font-weight: bold; font-size: 16px;"):
-        with gr.Row():
-            chatgpt_btn = gr.Button("ChatGPT")
-            llama_btn = gr.Button("LLaMA")
-            vicuna_btn = gr.Button("Vicuna")
-            alpaca_btn = gr.Button("Alpaca")
-            flant5_btn = gr.Button("Flan-T5")
-
-    # Strategy 2 - Instruction-Based Prompting
-    with gr.Accordion("Strategy 2 - Instruction-Based Prompting", style="font-weight: bold; font-size: 16px;"):
-        with gr.Row():
-            chatgpt_btn = gr.Button("ChatGPT")
-            llama_btn = gr.Button("LLaMA")
-            vicuna_btn = gr.Button("Vicuna")
-            alpaca_btn = gr.Button("Alpaca")
-            flant5_btn = gr.Button("Flan-T5")
-
-    # Strategy 3 - Structured Prompting
-    with gr.Accordion("Strategy 3 - Structured Prompting", style="font-weight: bold; font-size: 16px;"):
-        with gr.Row():
-            chatgpt_btn = gr.Button("ChatGPT")
-            llama_btn = gr.Button("LLaMA")
-            vicuna_btn = gr.Button("Vicuna")
-            alpaca_btn = gr.Button("Alpaca")
-            flant5_btn = gr.Button("Flan-T5")
-    
     iface.launch()

run_llm.py

@@ -12,9 +12,10 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM,
 
 from fastchat.model import load_model, get_conversation_template, add_model_args
 
-
 from nltk.tag.mapping import _UNIVERSAL_TAGS
 
+import gradio as gr
+
 uni_tags = list(_UNIVERSAL_TAGS)
 uni_tags[-1] = 'PUNC'
 
@@ -127,6 +128,7 @@ def para(m):
 def main(args=None):
 
     gid_list = selected_idx[args.start:args.end]
+    text_to_analyze = "Mr. Guber , by contrast , has been married to one woman for more than 20 years ."
 
 
     if 'gpt3' in args.model_path:
@@ -160,7 +162,7 @@ def main(args=None):
 
     if args.prompt == 1:
         for gid in tqdm(gid_list, desc='Query'):
-            text = ptb[gid]['text']
+            text = text_to_analyze
 
             for eid, ent in enumerate(ents):
                 os.makedirs(f'result/prompt1_qa/{args.model_path}/ptb/per_ent/{ent}', exist_ok=True)
@@ -215,7 +217,7 @@ def main(args=None):
 
     if args.prompt == 2:
         for gid in tqdm(gid_list, desc='Query'):
-            text = ptb[gid]['text']
+            text = text_to_analyze
 
             ## POS tagging
             # if os.path.exists(f'result/prompt2_instruction/pos_tagging/{args.model_path}/ptb/{gid}.txt'):
@@ -300,7 +302,7 @@ def main(args=None):
 
     if args.prompt == 3:
         for gid in tqdm(gid_list, desc='Query'):
-            text = ptb[gid]['text']
+            text = text_to_analyze
             tokens = ptb[gid]['tokens']
             poss = ptb[gid]['uni_poss']