Final update for version 2 application

app.py

@@ -1,22 +1,22 @@
-import os
+#this is version two with flagging features
 import gradio as gr
 from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 
 # initialize the environment
 model_name = 'anugrahap/gpt2-indo-textgen'
 HF_TOKEN = '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'
-callback = gr.HuggingFaceDatasetSaver(HF_TOKEN, "output-gpt2-indo-textgen")
+hf_writer = gr.HuggingFaceDatasetSaver(HF_TOKEN, "output-gpt2-indo-textgen")
 
-# define the tokenization method
-tokenizer = AutoTokenizer.from_pretrained(model_name,
-                                          model_max_length=1e30,
-                                          padding_side='right',
-                                          return_tensors='pt')
+# # define the tokenization method
+# tokenizer = AutoTokenizer.from_pretrained(model_name,
+#                                           model_max_length=1e30,
+#                                           padding_side='right',
+#                                           return_tensors='pt')
 
-# add the EOS token as PAD token to avoid warnings
-model = AutoModelForCausalLM.from_pretrained(model_name, pad_token_id=tokenizer.eos_token_id)
+# # add the EOS token as PAD token to avoid warnings
+# model = AutoModelForCausalLM.from_pretrained(model_name, pad_token_id=tokenizer.eos_token_id)
 
-generator = pipeline('text-generation', model=model, tokenizer=tokenizer)
+generator = pipeline('text-generation', model=model_name)
 
 # create the decoder parameter to generate the text
 def single_generation(text,min_length,max_length,temperature,top_k,top_p,num_beams,repetition_penalty,do_sample):
@@ -66,56 +66,20 @@ def single_generation(text,min_length,max_length,temperature,top_k,top_p,num_bea
     else:
         return error_unknown
 
-# create the decoder parameter to generate the text
-def multiple_generation(text,min_length,max_length,temperature,top_k,top_p,num_beams,repetition_penalty,do_sample):
-    # create local variable for error parameter    
-    error_rep=ValueError(f"ERROR: repetition penalty cannot be lower than one! Given rep penalty = {repetition_penalty}")
-    error_temp=ValueError(f"ERROR: temperature cannot be zero or lower! Given temperature = {temperature}")
-    error_minmax=ValueError(f"ERROR: min length must be lower than or equal to max length! Given min length = {min_length}")
-    error_numbeams_type=TypeError(f"ERROR: number of beams must be an integer not {type(num_beams)}")
-    error_topk_type=TypeError(f"ERROR: top k must be an integer not {type(top_k)}")
-    error_minmax_type=TypeError(f"ERROR: min length and max length must be an integer not {type(min_length)} and {type(max_length)}")   
-    error_empty=ValueError("ERROR: Input Text cannot be empty!")
-    error_unknown=TypeError("Unknown Error.")
 
-    if text != '':
-        if type(min_length) == int and type(max_length) == int:
-            if type(top_k) == int:
-                if type(num_beams) == int:
-                    if min_length <= max_length:
-                        if temperature > 0:
-                            if repetition_penalty >= 1:
-                                result = generator(text,
-                                                   min_length=min_length,
-                                                   max_length=max_length,
-                                                   temperature=temperature,
-                                                   top_k=top_k,
-                                                   top_p=top_p,
-                                                   num_beams=num_beams,
-                                                   repetition_penalty=repetition_penalty,
-                                                   do_sample=do_sample,
-                                                   no_repeat_ngram_size=2,
-                                                   num_return_sequences=3)
-                                return result[0]["generated_text"], result[1]["generated_text"], result[2]["generated_text"],
-                            elif repetition_penalty < 1:
-                                return error_rep,error_rep,error_rep
-                        elif temperature <= 0:
-                            return error_temp,error_temp,error_temp
-                    elif min_length > max_length:
-                        return error_minmax,error_minmax,error_minmax
-                elif type(num_beams) != int:
-                    return error_numbeams_type,error_numbeams_type,error_numbeams_type
-            elif type(top_k) != int:
-                return error_topk_type,error_topk_type,error_topk_type
-        elif type(min_length) != int or type(max_length) != int:
-            return  error_minmax_type,error_minmax_type,error_minmax_type
-    elif text == '':
-        return error_empty,error_empty,error_empty
-    else:
-        return error_unknown,error_unknown,error_unknown
+# create the variable needed for the gradio app
+forinput=[gr.Textbox(lines=5, label="Input Text"),
+        gr.Slider(label="Min Length", minimum=10, maximum=50, value=10, step=5),
+        gr.Slider(label="Max Length", minimum=10, maximum=100, value=30, step=10),
+        gr.Number(label="Temperature Sampling", value=1.5),
+        gr.Slider(label="Top K Sampling", minimum=0, maximum=100, value=30, step=5),
+        gr.Slider(label="Top P Sampling", minimum=0.01, maximum=1, value=0.93),
+        gr.Slider(label="Number of Beams", minimum=1, maximum=10, value=5, step=1),
+        gr.Number(label="Rep Penalty", value=2.0),
+        gr.Dropdown(label="Do Sample?", choices=[True,False], value=True, multiselect=False)]
 
+foroutput=gr.Textbox(lines=5, max_lines=50, label="Generated Text with Greedy/Beam Search Decoding")
 
-# create the baseline examples
 examples = [
     ["Indonesia adalah negara kepulauan", 10, 30, 1.0, 25, 0.92, 5, 2.0, True],
     ["Indonesia adalah negara kepulauan", 10, 30, 1.0, 25, 0.92, 5, 1.0, False],
@@ -124,9 +88,7 @@ examples = [
     ["Pemandangan di pantai kuta Bali sangatlah indah.", 30, 50, 0.5, 40, 0.98, 10, 1.0, True],
     ["Pemandangan di pantai kuta Bali sangatlah indah.", 10, 30, 1.5, 30, 0.93, 5, 2.0, True]]
 
-# using gradio block to create the interface
-with gr.Blocks(title="GPT-2 Indonesian Text Generation Playground", theme='Default') as app:
-    gr.Markdown("""
+title = """
     <style>
     .center {
     display: block;
@@ -147,66 +109,31 @@ with gr.Blocks(title="GPT-2 Indonesian Text Generation Playground", theme='Defau
     border="0" 
     class="center"
     style="height: 100px; width: 100px;"/>
-    <h1>GPT-2 Indonesian Text Generation Playground</h1>""")
-
-    gr.Markdown("<p><i>This project is a part of thesis requirement of Anugrah Akbar Praramadhan</i></p>")
-
-    with gr.Tabs():
-        #single generation
-        with gr.TabItem("Single Generation"):
-            with gr.Row():
-                with gr.Column():
-                    input1=[gr.Textbox(lines=5, label="Input Text"),
-                           gr.Slider(label="Min Length", minimum=10, maximum=50, value=10, step=5),
-                           gr.Slider(label="Max Length", minimum=10, maximum=100, value=30, step=10),
-                           gr.Number(label="Temperature Sampling", value=1.5),
-                           gr.Slider(label="Top K Sampling", minimum=0, maximum=100, value=30, step=5),
-                           gr.Slider(label="Top P Sampling", minimum=0.01, maximum=1, value=0.93),
-                           gr.Slider(label="Number of Beams", minimum=1, maximum=10, value=5, step=1),
-                           gr.Number(label="Rep Penalty", value=2.0),
-                           gr.Dropdown(label="Do Sample?", choices=[True,False], value=True, multiselect=False)]
-
-                with gr.Column():
-                    output1=gr.Textbox(lines=5, max_lines=50, label="Generated Text with Greedy/Beam Search Decoding")
-                    button1=gr.Button("Run the model")
-                    button1.click(fn=single_generation, inputs=input1, outputs=output1, show_progress=True)
-                    flag_btn = gr.Button("Flag")
-            
-            callback.setup([input1,output1],"Flagged Data Points")
-            flag_btn.click(lambda *args: callback.flag(args), input1, output1, preprocess=False)
-            gr.Examples(examples, inputs=input1)
-
-        #multiple generation
-        with gr.TabItem("Multiple Generation"):
-            with gr.Row():
-                with gr.Column():
-                    input2=[gr.Textbox(lines=5, label="Input Text"),
-                            gr.Slider(label="Min Length", minimum=10, maximum=50, value=10, step=5),
-                            gr.Slider(label="Max Length", minimum=10, maximum=100, value=30, step=10),
-                            gr.Number(label="Temperature Sampling", value=1.5),
-                            gr.Slider(label="Top K Sampling", minimum=0, maximum=100, value=30, step=5),
-                            gr.Slider(label="Top P Sampling", minimum=0.01, maximum=1, value=0.93),
-                            gr.Slider(label="Number of Beams", minimum=1, maximum=10, value=5, step=1),
-                            gr.Number(label="Rep Penalty", value=2.0),
-                            gr.Dropdown(label="Do Sample?", choices=[True,False], value=True, multiselect=False)]
-                with gr.Column():
-                    output2=[gr.Textbox(lines=5, max_lines=50, label="#1 Generated Text with Greedy/Beam Search Decoding"),
-                            gr.Textbox(lines=5, max_lines=50, label="#2 Generated Text with Greedy/Beam Search Decoding"),
-                            gr.Textbox(lines=5, max_lines=50, label="#3 Generated Text with Greedy/Beam Search Decoding")]
-                    button2=gr.Button("Run the model")
-                    button2.click(fn=multiple_generation, inputs=input2, outputs=output2, show_progress=True)
-                    flag_btn = gr.Button("Flag")
-
-            callback.setup([input2,output2],"Flagged Data Points")
-            flag_btn.click(lambda *args: callback.flag(args), input2, output2, preprocess=False)
-            gr.Examples(examples, inputs=input2)
-    
-    gr.Markdown("""<p style='text-align: center'>Copyright Anugrah Akbar Praramadhan 2023 <br>
+    <h1>GPT-2 Indonesian Text Generation Playground</h1>"""
+
+description = "<p><i>This project is a part of thesis requirement of Anugrah Akbar Praramadhan</i></p>"
+
+article = """<p style='text-align: center'>Copyright Anugrah Akbar Praramadhan 2023 <br>
     <p style='text-align: center'> Trained on Indo4B Benchmark Dataset of Indonesian language Wikipedia with a Causal Language Modeling (CLM) objective <br>
     <p style='text-align: center'><a href='https://huggingface.co/anugrahap/gpt2-indo-textgen' target='_blank'>Link to the Trained Model</a><br>
     <p style='text-align: center'><a href='https://huggingface.co/spaces/anugrahap/gpt2-indo-text-gen/tree/main' target='_blank'>Link to the Project Repository</a><br>
+    <p style='text-align: center'><a href='https://huggingface.co/datasets/anugrahap/output-gpt2-indo-textgen/' target='_blank'>Link to the Autosaved Generated Output</a><br>
     <p style='text-align: center'><a href='https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf' target='_blank'>Original Paper</a>
-    """)
+    """
+
+# using gradio interfaces
+app = gr.Interface(
+    fn=single_generation,
+    inputs=forinput,
+    outputs=foroutput,
+    examples=examples,
+    title=title,
+    description=description,
+    article=article,
+    allow_flagging='manual',
+    flagging_options=['Well Performed', 'Inappropriate Word Selection', 'Wordy', 'Strange Word', 'Others'],
+    flagging_callback=hf_writer)
+
 
 if __name__=='__main__':
     app.launch()