Update app.py

Different text generation model "distilgpt2"

app.py

@@ -163,41 +163,59 @@
 # which could capture long-term dependencies in text. Finally, we came to transformers, whose decoder architecture became popular for generative models 
 # used for generating text as an example.
 
-from transformers import GPT2LMHeadModel,GPT2Tokenizer
+# from transformers import GPT2LMHeadModel,GPT2Tokenizer
+# import gradio as grad
+
+# mdl = GPT2LMHeadModel.from_pretrained('gpt2')
+# gpt2_tkn=GPT2Tokenizer.from_pretrained('gpt2')
+
+# def generate(starting_text):
+#     tkn_ids = gpt2_tkn.encode(starting_text, return_tensors = 'pt')
+    
+#     # When no specific parameter is specified, the model performs a greedy search to find the next word, which entails selecting the word from all of the
+#     # alternatives that has the highest probability of being correct. This process is deterministic in nature, which means that resultant text is the same
+#     # as before if we use the same parameters.
+
+#     # The num_beams parameter does a beam search: it returns the sequences that have the highest probability, and then, when it comes time to
+#     # choose, it picks the one that has the highest probability.
+
+#     # The do_sample parameter select the next word at random from the probability distribution.
+
+#     # The temperature parameter controls the level of greed that the generative model exhibits.
+#     # If the temperature is low, the probabilities of sample classes other than the one with the highest log probability will be low. 
+#     # As a result, the model will probably output the text that is most correct, but it will be rather monotonous and contain only a small amount of variation.
+#     # If the temperature is high, the model has a greater chance of outputting different words than those with the highest probability. 
+#     # The generated text will feature a greater variety of topics, but there is also an increased likelihood that it will generate nonsense text and 
+#     # contain grammatical errors.
+
+#     # With less temperature (1.5 --> 0.1), the output becomes less variational.
+#     gpt2_tensors = mdl.generate(tkn_ids, max_length=100, no_repeat_ngram_size=True, num_beams=3, do_sample=True, temperature=0.1)
+#     response=""
+#     #response = gpt2_tensors
+#     for i, x in enumerate(gpt2_tensors):
+#         response=response+f"{i}: {gpt2_tkn.decode(x, skip_special_tokens=True)}" # Decode tensors into text
+#     return gpt2_tensors, response
+
+# txt=grad.Textbox(lines=1, label="English", placeholder="English Text here")
+# out_tensors=grad.Textbox(lines=1, label="Generated Tensors")
+# out_text=grad.Textbox(lines=1, label="Generated Text")
+
+# grad.Interface(generate, inputs=txt, outputs=[out_tensors, out_text]).launch()
+
+#-----------------------------------------------------------------------------------
+# 9. Text Generation: different model "distilgpt2"
+
+from transformers import pipeline, set_seed
 import gradio as grad
 
-mdl = GPT2LMHeadModel.from_pretrained('gpt2')
-gpt2_tkn=GPT2Tokenizer.from_pretrained('gpt2')
+gpt2_pipe = pipeline('text-generation', model='distilgpt2')
+set_seed(42)
 
 def generate(starting_text):
-    tkn_ids = gpt2_tkn.encode(starting_text, return_tensors = 'pt')
-    
-    # When no specific parameter is specified, the model performs a greedy search to find the next word, which entails selecting the word from all of the
-    # alternatives that has the highest probability of being correct. This process is deterministic in nature, which means that resultant text is the same
-    # as before if we use the same parameters.
-
-    # The num_beams parameter does a beam search: it returns the sequences that have the highest probability, and then, when it comes time to
-    # choose, it picks the one that has the highest probability.
-
-    # The do_sample parameter select the next word at random from the probability distribution.
-
-    # The temperature parameter controls the level of greed that the generative model exhibits.
-    # If the temperature is low, the probabilities of sample classes other than the one with the highest log probability will be low. 
-    # As a result, the model will probably output the text that is most correct, but it will be rather monotonous and contain only a small amount of variation.
-    # If the temperature is high, the model has a greater chance of outputting different words than those with the highest probability. 
-    # The generated text will feature a greater variety of topics, but there is also an increased likelihood that it will generate nonsense text and 
-    # contain grammatical errors.
-
-    # With less temperature (1.5 --> 0.1), the output becomes less variational.
-    gpt2_tensors = mdl.generate(tkn_ids, max_length=100, no_repeat_ngram_size=True, num_beams=3, do_sample=True, temperature=0.1)
-    response=""
-    #response = gpt2_tensors
-    for i, x in enumerate(gpt2_tensors):
-        response=response+f"{i}: {gpt2_tkn.decode(x, skip_special_tokens=True)}" # Decode tensors into text
-    return gpt2_tensors, response
+    response= gpt2_pipe(starting_text, max_length=20, num_return_sequences=5)
+    return response
 
 txt=grad.Textbox(lines=1, label="English", placeholder="English Text here")
-out_tensors=grad.Textbox(lines=1, label="Generated Tensors")
-out_text=grad.Textbox(lines=1, label="Generated Text")
+out=grad.Textbox(lines=1, label="Generated Text")
 
-grad.Interface(generate, inputs=txt, outputs=[out_tensors, out_text]).launch()
+grad.Interface(generate, inputs=txt, outputs=out).launch()