to(cuda)

lora.py

@@ -8,7 +8,7 @@ from unsloth import FastLanguageModel
 
 load_in_4bit = True
 
-peft_model_id = "ID2223JR/lora_model"
+peft_model_id = "ID2223JR/recipe_model"
 
 model, tokenizer = FastLanguageModel.from_pretrained(
     model_name=peft_model_id,
@@ -51,10 +51,6 @@ def submit_to_model():
     ingredients_list.clear()
 
     messages = [
-        {
-            "role": "system",
-            "content": "You are a world-renowned chef, celebrated for your expertise in creating delectable dishes from diverse cuisines. You have a vast knowledge of ingredients, cooking techniques, and dietary preferences. Your role is to suggest personalized recipes based on the ingredients available, dietary restrictions, or specific meal requests. Please provide clear, step-by-step instructions and any useful tips to enhance the dish's flavor or presentation. Begin by introducing the recipe and why it’s a great choice.",
-        },
         {"role": "user", "content": prompt},
     ]
     inputs = tokenizer.apply_chat_template(
@@ -74,6 +70,8 @@ def submit_to_model():
     content = ""
     for text in text_streamer:
         print(text)
+        if text.endswith("<|eot_id|>"):
+            text = text.replace("<|eot_id|>", "")
         content += text
         yield content
 
@@ -90,8 +88,9 @@ def app():
         add_button = gr.Button("Add Ingredient", interactive=False)
         output = gr.Textbox(label="Ingredients List", lines=10, interactive=False)
 
+        submit_button = gr.Button("Give me a meal!")
+
         with gr.Row():
-            submit_button = gr.Button("Submit")
             model_output = gr.Textbox(
                 label="Recipe Suggestion", lines=10, interactive=False
             )