Update app.py

app.py

@@ -1,5 +1,4 @@
-# app.py – FINAL, BULLETPROOF VERSION (November 2025)
-# Works on ANY HF Space (CPU or GPU), instant output, dual input
+# app.py – FINAL VERSION THAT WORKS TODAY (November 2025)
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 from peft import PeftModel
@@ -8,25 +7,23 @@ import gradio as gr
 BASE_MODEL   = "mistralai/Mistral-7B-Instruct-v0.2"
 LORA_ADAPTER = "rishu834763/java-explainer-lora"
 
-print("Loading Java Explainer (8-bit CPU – guaranteed to start)...")
+print("Loading your Java Explainer (8-bit CPU – guaranteed to work)...")
 
-# 8-bit CPU mode = ~7.5 GB RAM, fits on every HF Space, no offload errors
+# 8-bit CPU = works everywhere, no offload error, no CUDA error
 model = AutoModelForCausalLM.from_pretrained(
     BASE_MODEL,
     load_in_8bit=True,
-    device_map="cpu",                    # Force CPU only → no offload confusion
+    device_map="cpu",
     torch_dtype=torch.float16,
     low_cpu_mem_usage=True,
 )
 
-# Apply your LoRA (adds only ~168 MB)
 model = PeftModel.from_pretrained(model, LORA_ADAPTER)
 
 tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
 if tokenizer.pad_token is None:
     tokenizer.pad_token = tokenizer.eos_token
 
-# Fast, reliable pipeline
 pipe = pipeline(
     "text-generation",
     model=model,
@@ -40,11 +37,8 @@ pipe = pipeline(
 )
 
 SYSTEM_PROMPT = """You are the world's best Java teacher.
-Always respond with:
-• Clear, step-by-step explanation
-• Clean, modern, runnable Java code (Java 17+)
-• Best practices and common pitfalls
-• Fix any bugs or bad patterns"""
+Always respond with clear explanation + clean, modern, runnable Java code.
+Fix bugs, use best practices (Java 17+), records, var, streams, etc."""
 
 def generate(instruction: str, code: str = ""):
     user_input = f"### Instruction:\n{instruction.strip()}\n\n### Code:\n{code.strip()}" if code.strip() else instruction.strip()
@@ -55,31 +49,30 @@ def generate(instruction: str, code: str = ""):
     ]
     
     prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    
     output = pipe(prompt)[0]["generated_text"].strip()
     return output
 
-# Dual-input UI (Instruction + Code)
+# FIXED UI – works with current Gradio (no 'placeholder' in gr.Code)
 with gr.Blocks(theme=gr.themes.Soft(), title="Java Explainer Pro") as demo:
-    gr.Markdown("# Java Explainer Pro\nYour personal senior Java mentor")
+    gr.Markdown("# Java Explainer Pro\nYour personal senior Java mentor is ready")
 
     with gr.Row():
-        with gr.Column(scale=1):
+        with gr.Column():
             instruction = gr.Textbox(
-                label="What do you want?",
-                placeholder="Explain this code / Fix this bug / Make it thread-safe / Convert to records / Best way to read JSON",
+                label="Instruction / Question",
+                placeholder="Explain this code / Fix bug / Make thread-safe / Convert to records / Best way to do X",
                 lines=5
             )
             code_input = gr.Code(
                 label="Java Code (optional)",
                 language="java",
                 lines=15,
-                placeholder="// Paste your code here\npublic class Main {\n    public static void main(String[] args) {\n        System.out.println(\"Hello\");\n    }\n}"
+                value="// Paste your code here or leave empty\npublic class Main {\n    public static void main(String[] args) {\n        System.out.println(\"Hello\");\n    }\n}"
             )
-            submit = gr.Button("Get Answer", variant="primary", size="lg")
+            submit = gr.Button("Get Expert Answer", variant="primary", size="lg")
 
-        with gr.Column(scale=1):
-            output = gr.Markdown(label="Expert Answer")
+        with gr.Column():
+            output = gr.Markdown()
 
     submit.click(generate, [instruction, code_input], output)
     instruction.submit(generate, [instruction, code_input], output)