test chatbot

app.py

@@ -1,175 +1,34 @@
-import os
-import sys
-import json
-import time
-import openai
-import pickle
-import argparse
-import requests
-from tqdm import tqdm
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizer
-
-from fastchat.model import load_model, get_conversation_template, add_model_args
-
-from nltk.tag.mapping import _UNIVERSAL_TAGS
-
 import gradio as gr
-from transformers import pipeline
-
-demo = gr.Blocks()
-
-uni_tags = list(_UNIVERSAL_TAGS)
-uni_tags[-1] = 'PUNC'
-
-bio_tags = ['B', 'I', 'O']
-chunk_tags = ['ADJP', 'ADVP', 'CONJP', 'INTJ', 'LST', 'NP', 'O', 'PP', 'PRT', 'SBAR', 'UCP', 'VP']
-
-syntags = ['NP', 'S', 'VP', 'ADJP', 'ADVP', 'SBAR', 'TOP', 'PP', 'POS', 'NAC', "''", 'SINV', 'PRN', 'QP', 'WHNP', 'RB', 'FRAG',
- 'WHADVP', 'NX', 'PRT', 'VBZ', 'VBP', 'MD', 'NN', 'WHPP', 'SQ', 'SBARQ', 'LST', 'INTJ', 'X', 'UCP', 'CONJP', 'NNP', 'CD', 'JJ',
- 'VBD', 'WHADJP', 'PRP', 'RRC', 'NNS', 'SYM', 'CC']
-
-openai.api_key = " "
-
-# determinant vs. determiner
-# https://wikidiff.com/determiner/determinant
-ents_prompt = ['Noun','Verb','Adjective','Adverb','Preposition/Subord','Coordinating Conjunction',# 'Cardinal Number',
-    'Determiner',
-    'Noun Phrase','Verb Phrase','Adjective Phrase','Adverb Phrase','Preposition Phrase','Conjunction Phrase','Coordinate Phrase','Quantitave Phrase','Complex Nominal',
-    'Clause','Dependent Clause','Fragment Clause','T-unit','Complex T-unit',# 'Fragment T-unit',
-][7:]
-ents = ['NN', 'VB', 'JJ', 'RB', 'IN', 'CC', 'DT', 'NP', 'VP', 'ADJP', 'ADVP', 'PP', 'CONJP', 'CP', 'QP', 'CN', 'C', 'DC', 'FC', 'T', 'CT'][7:]
-
-
-ents_prompt_uni_tags = ['Verb', 'Noun', 'Pronoun', 'Adjective', 'Adverb', 'Preposition and Postposition', 'Coordinating Conjunction', 
-                        'Determiner', 'Cardinal Number', 'Particles or other function words', 
-                        'Words that cannot be assigned a POS tag', 'Punctuation']
-
-ents = uni_tags + ents
-ents_prompt = ents_prompt_uni_tags + ents_prompt
-
-for i, j in zip(ents, ents_prompt):
-    print(i, j)
-
-model_mapping = {
-    'gpt3.5': 'gpt2',
-    #'vicuna-7b': 'lmsys/vicuna-7b-v1.3',
-    #'llama-7b': './llama/hf/7B',
-}
-
-with open('sample_uniform_1k_2.txt', 'r') as f:
-    selected_idx = f.readlines()
-selected_idx = [int(i.strip()) for i in selected_idx]#[s:e]
-
-ptb = []
-with open('ptb.jsonl', 'r') as f:
-    for l in f:
-        ptb.append(json.loads(l))
-
-
-## Prompt 1
-template_all = '''Please output the <Noun, Verb, Adjective, Adverb, Preposition/Subord, Coordinating Conjunction, Cardinal Number, Determiner, Noun Phrase, Verb Phrase, Adjective Phrase, Adverb Phrase, Preposition Phrase, Conjunction Phrase, Coordinate Phrase, Quantitave Phrase, Complex Nominal, Clause, Dependent Clause, Fragment Clause, T-unit, Complex T-unit, Fragment T-unit> in the following sentence without any additional text in json format: "{}"'''
-template_single = '''Please output any <{}> in the following sentence one per line without any additional text: "{}"'''
-
-## Prompt 2
-prompt2_pos = '''Please pos tag the following sentence using Universal POS tag set without generating any additional text: {}'''
-prompt2_chunk = '''Please do sentence chunking for the following sentence as in CoNLL 2000 shared task without generating any addtional text: {}'''
-prompt2_parse = '''Generate textual representation of the constituency parse tree of the following sentence using Penn TreeBank tag set without outputing any additional text: {}'''
-
-prompt2_chunk = '''Please chunk the following sentence in CoNLL 2000 format with BIO tags without outputing any additional text: {}'''
-
-## Prompt 3
-with open('demonstration_3_42_pos.txt', 'r') as f:
-    demon_pos = f.read()
-with open('demonstration_3_42_chunk.txt', 'r') as f:
-    demon_chunk = f.read()
-with open('demonstration_3_42_parse.txt', 'r') as f:
-    demon_parse = f.read()
-
-# Your existing code
-theme = gr.themes.Soft()
-
-# issue get request for gpt 3.5
-gpt_pipeline = pipeline(task="text2text-generation", model="gpt2")
-#vicuna7b_pipeline = pipeline(task="text2text-generation", model="lmsys/vicuna-7b-v1.3")
-#llama7b_pipeline = pipeline(task="text2text-generation", model="./llama/hf/7B")
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-# Dropdown options for model and task
-model_options = list(model_mapping.keys())
-task_options = ['POS', 'Chunking'] # remove parsing
+# Load the models and tokenizers
+gpt35_model = AutoModelForCausalLM.from_pretrained("gpt-3.5-turbo-0613")
+gpt35_tokenizer = AutoTokenizer.from_pretrained("gpt-3.5-turbo-0613")
 
+vicuna_model = AutoModelForCausalLM.from_pretrained("lmsys/vicuna-7b-v1.3")
+vicuna_tokenizer = AutoTokenizer.from_pretrained("lmsys/vicuna-7b-v1.3")
 
-# Function to process text based on model and task
-def process_text(tab, text):
-    if tab == 'POS Tab':
-        strategy1_format = template_all.format(text)
-        strategy2_format = prompt2_pos.format(text)
-        strategy3_format = demon_pos
+llama_model = AutoModelForCausalLM.from_pretrained("./llama/hf/7B")
+llama_tokenizer = AutoTokenizer.from_pretrained("./llama/hf/7B")
 
-        vicuna_result1 = gpt_pipeline(strategy1_format)[0]['generated_text']
-        vicuna_result2 = gpt_pipeline(strategy2_format)[0]['generated_text']
-        vicuna_result3 = gpt_pipeline(strategy3_format)[0]['generated_text']
-        
-        return (vicuna_result1, vicuna_result2, vicuna_result3)
-    elif tab == 'Chunk Tab':
-        strategy1_format = template_all.format(text)
-        strategy2_format = prompt2_chunk.format(text)
-        strategy3_format = demon_chunk
+# Define the function for generating responses
+def generate_response(model, tokenizer, prompt):
+    inputs = tokenizer(prompt, return_tensors="pt")
+    outputs = model.generate(**inputs, max_length=100, pad_token_id=tokenizer.eos_token_id)
+    response = tokenizer.decode(outputs[0])
+    return response
 
-        result1 = gpt_pipeline(strategy1_format)[0]['generated_text']
-        result2 = gpt_pipeline(strategy2_format)[0]['generated_text']
-        result3 = gpt_pipeline(strategy3_format)[0]['generated_text']
-        return (result1, result2, result3)
+# Define the Gradio interface
+def chatbot_interface(prompt):
+    gpt35_response = generate_response(gpt35_model, gpt35_tokenizer, prompt)
+    vicuna_response = generate_response(vicuna_model, vicuna_tokenizer, prompt)
+    llama_response = generate_response(llama_model, llama_tokenizer, prompt)
 
-# Gradio interface
-with demo:
-    gr.Markdown("# LLM Evaluator With Linguistic Scrutiny")
-    
-    with gr.Tabs():
-        with gr.TabItem("POS", id="POS Tab"):
-            with gr.Row():
-              gr.Markdown("<center>Vicuna 7b</center>")
-              gr.Markdown("<center> LLaMA-7b </center>")
-              gr.Markdown("<center> GPT 3.5 </center>")
-            with gr.Row():
-                model1_S1_output = gr.Textbox(label="Strategy 1 QA")
-                model2_S1_output = gr.Textbox(label=".")
-                model3_S1_output = gr.Textbox(label=".")
-            with gr.Row():
-                model1_S2_output = gr.Textbox(label="Strategy 2 Instruction")
-                model2_S2_output = gr.Textbox(label=".")
-                model3_S2_output = gr.Textbox(label=".")
-            with gr.Row():
-                model1_S3_output = gr.Textbox(label="Strategy 3 Structured Prompting")
-                model2_S3_output = gr.Textbox(label=".")
-                model3_S3_output = gr.Textbox(label=".")
-            with gr.Row():
-                prompt_POS = gr.Textbox(show_label=False, placeholder="Enter prompt")
-                send_button_POS = gr.Button("Send", scale=0)
-              
-        with gr.TabItem("Chunking", id="Chunk Tab"):
-            with gr.Row():
-                gr.Markdown("<center>Vicuna 7b</center>")
-                gr.Markdown("<center> LLaMA-7b </center>")
-                gr.Markdown("<center> GPT 3.5 </center>")
-            with gr.Row():
-                model1_S1_output = gr.Textbox(label="Strategy 1 QA")
-                model2_S1_output = gr.Textbox(label=".")
-                model3_S1_output = gr.Textbox(label=".")
-            with gr.Row():
-                model1_S2_output = gr.Textbox(label="Strategy 2 Instruction")
-                model2_S2_output = gr.Textbox(label=".")
-                model3_S2_output = gr.Textbox(label=".")
-            with gr.Row():
-                model1_S3_output = gr.Textbox(label="Strategy 3 Structured Prompting")
-                model2_S3_output = gr.Textbox(label=".")
-                model3_S3_output = gr.Textbox(label=".")
-            with gr.Row():
-                prompt_Chunk = gr.Textbox(id="prompt_Chunk", show_label=False, placeholder="Enter prompt")
-                send_button_Chunk = gr.Button("Send", scale=0)
-    
-    send_button_POS.click(process_text, inputs=["POS Tab", prompt_Chunk], outputs=[model1_S1_output, model1_S1_output, model1_S1_output])
-    send_button_Chunk.click(process_text, inputs=["Chunk Tab", prompt_POS], outputs=[model1_S1_output, model1_S1_output, model1_S1_output])
+    return {"GPT-3.5": gpt35_response, "Vicuna-7B": vicuna_response, "Llama-7B": llama_response}
 
-demo.launch()
+iface = gr.Interface(fn=chatbot_interface, 
+                     inputs="text", 
+                     outputs="panel",
+                     title="Chatbot with Three Models")
 
+iface.launch()

placeholder.py

@@ -0,0 +1,175 @@
+import os
+import sys
+import json
+import time
+import openai
+import pickle
+import argparse
+import requests
+from tqdm import tqdm
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM, LlamaTokenizer
+
+from fastchat.model import load_model, get_conversation_template, add_model_args
+
+from nltk.tag.mapping import _UNIVERSAL_TAGS
+
+import gradio as gr
+from transformers import pipeline
+
+demo = gr.Blocks()
+
+uni_tags = list(_UNIVERSAL_TAGS)
+uni_tags[-1] = 'PUNC'
+
+bio_tags = ['B', 'I', 'O']
+chunk_tags = ['ADJP', 'ADVP', 'CONJP', 'INTJ', 'LST', 'NP', 'O', 'PP', 'PRT', 'SBAR', 'UCP', 'VP']
+
+syntags = ['NP', 'S', 'VP', 'ADJP', 'ADVP', 'SBAR', 'TOP', 'PP', 'POS', 'NAC', "''", 'SINV', 'PRN', 'QP', 'WHNP', 'RB', 'FRAG',
+ 'WHADVP', 'NX', 'PRT', 'VBZ', 'VBP', 'MD', 'NN', 'WHPP', 'SQ', 'SBARQ', 'LST', 'INTJ', 'X', 'UCP', 'CONJP', 'NNP', 'CD', 'JJ',
+ 'VBD', 'WHADJP', 'PRP', 'RRC', 'NNS', 'SYM', 'CC']
+
+openai.api_key = " "
+
+# determinant vs. determiner
+# https://wikidiff.com/determiner/determinant
+ents_prompt = ['Noun','Verb','Adjective','Adverb','Preposition/Subord','Coordinating Conjunction',# 'Cardinal Number',
+    'Determiner',
+    'Noun Phrase','Verb Phrase','Adjective Phrase','Adverb Phrase','Preposition Phrase','Conjunction Phrase','Coordinate Phrase','Quantitave Phrase','Complex Nominal',
+    'Clause','Dependent Clause','Fragment Clause','T-unit','Complex T-unit',# 'Fragment T-unit',
+][7:]
+ents = ['NN', 'VB', 'JJ', 'RB', 'IN', 'CC', 'DT', 'NP', 'VP', 'ADJP', 'ADVP', 'PP', 'CONJP', 'CP', 'QP', 'CN', 'C', 'DC', 'FC', 'T', 'CT'][7:]
+
+
+ents_prompt_uni_tags = ['Verb', 'Noun', 'Pronoun', 'Adjective', 'Adverb', 'Preposition and Postposition', 'Coordinating Conjunction', 
+                        'Determiner', 'Cardinal Number', 'Particles or other function words', 
+                        'Words that cannot be assigned a POS tag', 'Punctuation']
+
+ents = uni_tags + ents
+ents_prompt = ents_prompt_uni_tags + ents_prompt
+
+for i, j in zip(ents, ents_prompt):
+    print(i, j)
+
+model_mapping = {
+    'gpt3.5': 'gpt2',
+    #'vicuna-7b': 'lmsys/vicuna-7b-v1.3',
+    #'llama-7b': './llama/hf/7B',
+}
+
+with open('sample_uniform_1k_2.txt', 'r') as f:
+    selected_idx = f.readlines()
+selected_idx = [int(i.strip()) for i in selected_idx]#[s:e]
+
+ptb = []
+with open('ptb.jsonl', 'r') as f:
+    for l in f:
+        ptb.append(json.loads(l))
+
+
+## Prompt 1
+template_all = '''Please output the <Noun, Verb, Adjective, Adverb, Preposition/Subord, Coordinating Conjunction, Cardinal Number, Determiner, Noun Phrase, Verb Phrase, Adjective Phrase, Adverb Phrase, Preposition Phrase, Conjunction Phrase, Coordinate Phrase, Quantitave Phrase, Complex Nominal, Clause, Dependent Clause, Fragment Clause, T-unit, Complex T-unit, Fragment T-unit> in the following sentence without any additional text in json format: "{}"'''
+template_single = '''Please output any <{}> in the following sentence one per line without any additional text: "{}"'''
+
+## Prompt 2
+prompt2_pos = '''Please pos tag the following sentence using Universal POS tag set without generating any additional text: {}'''
+prompt2_chunk = '''Please do sentence chunking for the following sentence as in CoNLL 2000 shared task without generating any addtional text: {}'''
+prompt2_parse = '''Generate textual representation of the constituency parse tree of the following sentence using Penn TreeBank tag set without outputing any additional text: {}'''
+
+prompt2_chunk = '''Please chunk the following sentence in CoNLL 2000 format with BIO tags without outputing any additional text: {}'''
+
+## Prompt 3
+with open('demonstration_3_42_pos.txt', 'r') as f:
+    demon_pos = f.read()
+with open('demonstration_3_42_chunk.txt', 'r') as f:
+    demon_chunk = f.read()
+with open('demonstration_3_42_parse.txt', 'r') as f:
+    demon_parse = f.read()
+
+# Your existing code
+theme = gr.themes.Soft()
+
+# issue get request for gpt 3.5
+gpt_pipeline = pipeline(task="text2text-generation", model="gpt2")
+#vicuna7b_pipeline = pipeline(task="text2text-generation", model="lmsys/vicuna-7b-v1.3")
+#llama7b_pipeline = pipeline(task="text2text-generation", model="./llama/hf/7B")
+
+# Dropdown options for model and task
+model_options = list(model_mapping.keys())
+task_options = ['POS', 'Chunking'] # remove parsing
+
+
+# Function to process text based on model and task
+def process_text(tab, text):
+    if tab == 'POS Tab':
+        strategy1_format = template_all.format(text)
+        strategy2_format = prompt2_pos.format(text)
+        strategy3_format = demon_pos
+
+        vicuna_result1 = gpt_pipeline(strategy1_format)[0]['generated_text']
+        vicuna_result2 = gpt_pipeline(strategy2_format)[0]['generated_text']
+        vicuna_result3 = gpt_pipeline(strategy3_format)[0]['generated_text']
+        
+        return (vicuna_result1, vicuna_result2, vicuna_result3)
+    elif tab == 'Chunk Tab':
+        strategy1_format = template_all.format(text)
+        strategy2_format = prompt2_chunk.format(text)
+        strategy3_format = demon_chunk
+
+        result1 = gpt_pipeline(strategy1_format)[0]['generated_text']
+        result2 = gpt_pipeline(strategy2_format)[0]['generated_text']
+        result3 = gpt_pipeline(strategy3_format)[0]['generated_text']
+        return (result1, result2, result3)
+
+# Gradio interface
+with demo:
+    gr.Markdown("# LLM Evaluator With Linguistic Scrutiny")
+    
+    with gr.Tabs():
+        with gr.TabItem("POS", id="POS Tab"):
+            with gr.Row():
+              gr.Markdown("<center>Vicuna 7b</center>")
+              gr.Markdown("<center> LLaMA-7b </center>")
+              gr.Markdown("<center> GPT 3.5 </center>")
+            with gr.Row():
+                model1_S1_output = gr.Textbox(label="Strategy 1 QA")
+                model2_S1_output = gr.Textbox(label=".")
+                model3_S1_output = gr.Textbox(label=".")
+            with gr.Row():
+                model1_S2_output = gr.Textbox(label="Strategy 2 Instruction")
+                model2_S2_output = gr.Textbox(label=".")
+                model3_S2_output = gr.Textbox(label=".")
+            with gr.Row():
+                model1_S3_output = gr.Textbox(label="Strategy 3 Structured Prompting")
+                model2_S3_output = gr.Textbox(label=".")
+                model3_S3_output = gr.Textbox(label=".")
+            with gr.Row():
+                prompt_POS = gr.Textbox(show_label=False, placeholder="Enter prompt")
+                send_button_POS = gr.Button("Send", scale=0)
+              
+        with gr.TabItem("Chunking", id="Chunk Tab"):
+            with gr.Row():
+                gr.Markdown("<center>Vicuna 7b</center>")
+                gr.Markdown("<center> LLaMA-7b </center>")
+                gr.Markdown("<center> GPT 3.5 </center>")
+            with gr.Row():
+                model1_S1_output = gr.Textbox(label="Strategy 1 QA")
+                model2_S1_output = gr.Textbox(label=".")
+                model3_S1_output = gr.Textbox(label=".")
+            with gr.Row():
+                model1_S2_output = gr.Textbox(label="Strategy 2 Instruction")
+                model2_S2_output = gr.Textbox(label=".")
+                model3_S2_output = gr.Textbox(label=".")
+            with gr.Row():
+                model1_S3_output = gr.Textbox(label="Strategy 3 Structured Prompting")
+                model2_S3_output = gr.Textbox(label=".")
+                model3_S3_output = gr.Textbox(label=".")
+            with gr.Row():
+                prompt_Chunk = gr.Textbox(id="prompt_Chunk", show_label=False, placeholder="Enter prompt")
+                send_button_Chunk = gr.Button("Send", scale=0)
+    
+    send_button_POS.click(process_text, inputs=["POS Tab", prompt_Chunk], outputs=[model1_S1_output, model1_S1_output, model1_S1_output])
+    send_button_Chunk.click(process_text, inputs=["Chunk Tab", prompt_POS], outputs=[model1_S1_output, model1_S1_output, model1_S1_output])
+
+demo.launch()
+