Update app.py

app.py

@@ -5,20 +5,25 @@ from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
 model = AutoModelForSeq2SeqLM.from_pretrained("vennify/t5-base-grammar-correction")
 tokenizer = AutoTokenizer.from_pretrained("vennify/t5-base-grammar-correction")
 
-def correct_text(text, max_length, min_length, max_new_tokens, num_beams, temperature, top_p):
+def correct_text(text, max_length, min_length, max_new_tokens, min_new_tokens, num_beams, temperature, top_p):
     inputs = tokenizer.encode("grammar: " + text, return_tensors="pt")
     
-    if max_new_tokens > 0:
-        outputs = model.generate(
-            inputs,
-            max_length=max_length,
-            max_new_tokens=max_new_tokens,
-            min_length=min_length,
-            num_beams=num_beams,
-            temperature=temperature,
-            top_p=top_p,
-            early_stopping=True
-        )
+    if max_new_tokens > 0 or min_new_tokens > 0:
+        if max_new_tokens > 0 and min_new_tokens > 0:
+            outputs = model.generate(
+                inputs,
+                max_new_tokens=max_new_tokens,
+                min_new_tokens=min_new_tokens,
+                num_beams=num_beams,
+                temperature=temperature,
+                top_p=top_p,
+                early_stopping=True,
+                do_sample=True
+            )
+        elif max_new_tokens > 0:
+            outputs = model.generate(inputs, max_new_tokens=max_new_tokens, min_length=min_length, num_beams=num_beams, temperature=temperature, top_p=top_p, early_stopping=True, do_sample=True)
+        else:
+            outputs = model.generate(inputs, max_length=max_length, min_new_tokens=min_new_tokens, num_beams=num_beams, temperature=temperature, top_p=top_p, early_stopping=True, do_sample=True)
     else:
         outputs = model.generate(
             inputs,
@@ -27,19 +32,13 @@ def correct_text(text, max_length, min_length, max_new_tokens, num_beams, temper
             num_beams=num_beams,
             temperature=temperature,
             top_p=top_p,
-            early_stopping=True
+            early_stopping=True,
+            do_sample=True
         )
     
     corrected_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
     yield corrected_text
 
-def respond(prompt, max_length, min_length, max_new_tokens, num_beams, temperature, top_p):
-    # Your response generation logic here
-    #response = correct_text(message, max_length, max_new_tokens, min_length, num_beams, temperature, top_p)
-    #yield response
-    #return f"System message: {system_message}, Max Length: {max_length}, Min Length: {min_length}, Max new tokens: {max_new_tokens}, Num Beams: {num_beams}, Temperature: {temperature}, Top-p: {top_p}"
-    #response = correct_text(prompt, max_length, max_new_tokens, min_length, num_beams, temperature, top_p)
-    return prompt
 
 def update_prompt(prompt):
     return prompt
@@ -68,14 +67,14 @@ with gr.Blocks() as demo:
     with gr.Accordion("Generation Parameters:", open=False):
         max_length = gr.Slider(minimum=1, maximum=256, value=80, step=1, label="Max Length")
         min_length = gr.Slider(minimum=1, maximum=256, value=0,  step=1, label="Min Length")
-        max_tokens = gr.Slider(minimum=0, maximum=256, value=0, step=1, label="Max new tokens")
+        max_tokens = gr.Slider(minimum=0, maximum=256, value=0, step=1, label="Max New Tokens")
+        min_tokens = gr.Slider(minimum=0, maximum=256, value=0, step=1, label="Min New Tokens")
         num_beams = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Num Beams")
         temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
         top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")
 
     
-
-    submitBtn.click(correct_text, [prompt_box, max_length, min_length, max_tokens, num_beams, temperature, top_p], output_box)
+    submitBtn.click(correct_text, [prompt_box, max_length, min_length, max_tokens, min_tokens, num_beams, temperature, top_p], output_box)