Update app.py

app.py

@@ -25,39 +25,43 @@ os.system('python -m spacy download en_core_web_sm')
 nlp = spacy.load("en_core_web_sm")
 
 def historical_generation(prompt, max_new_tokens=600, top_k=50, temperature=0.7, top_p=0.95, repetition_penalty=1.0):
-    prompt = f"### Text ###\n{prompt}"
-    inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True, max_length=1024)
-    input_ids = inputs["input_ids"].to(device)
-    attention_mask = inputs["attention_mask"].to(device)
-
-    output = model.generate(
-        input_ids,
-        attention_mask=attention_mask,
-        max_new_tokens=max_new_tokens,
-        pad_token_id=tokenizer.eos_token_id,
-        top_k=top_k,
-        temperature=temperature,
-        top_p=top_p,
-        do_sample=True,
-        repetition_penalty=repetition_penalty,
-        bos_token_id=tokenizer.bos_token_id,
-        eos_token_id=tokenizer.eos_token_id
-    )
-
-    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
-
-    if "### Correction ###" in generated_text:
-        generated_text = generated_text.split("### Correction ###")[1].strip()
-
-    tokens = tokenizer.tokenize(generated_text)
-
-    highlighted_text = []
-    for token in tokens:
-        clean_token = token.replace("Ġ", "") 
-        token_type = tokenizer.convert_ids_to_tokens([tokenizer.convert_tokens_to_ids(token)])[0].replace("Ġ", "")
-        highlighted_text.append((clean_token, token_type))
-
-    return highlighted_text, generated_text  
+    with torch.no_grad():
+        prompt = f"### Text ###\n{prompt}"
+        inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True, max_length=1024)
+        input_ids = inputs["input_ids"].to(device)
+        attention_mask = inputs["attention_mask"].to(device)
+        
+        output = model.generate(
+            input_ids,
+            attention_mask=attention_mask,
+            max_new_tokens=max_new_tokens,
+            pad_token_id=tokenizer.eos_token_id,
+            top_k=top_k,
+            temperature=temperature,
+            top_p=top_p,
+            do_sample=True,
+            repetition_penalty=repetition_penalty,
+            bos_token_id=tokenizer.bos_token_id,
+            eos_token_id=tokenizer.eos_token_id
+        )
+
+        generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
+
+        if "### Correction ###" in generated_text:
+            generated_text = generated_text.split("### Correction ###")[1].strip()
+
+        tokens = tokenizer.tokenize(generated_text)
+
+        highlighted_text = []
+        for token in tokens:
+            clean_token = token.replace("Ġ", "") 
+            token_type = tokenizer.convert_ids_to_tokens([tokenizer.convert_tokens_to_ids(token)])[0].replace("Ġ", "")
+            highlighted_text.append((clean_token, token_type))
+
+        del inputs, input_ids, attention_mask, output, tokens
+        torch.cuda.empty_cache()
+
+        return highlighted_text, generated_text  
 
 def text_analysis(text):
     doc = nlp(text)