Update app.py

app.py

@@ -1,52 +1,25 @@
 import gradio as gr
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
-from peft import PeftModel
 from threading import Thread
 
 # ------------------------------------------------------------------
 # 1. Model setup
 # ------------------------------------------------------------------
-BASE_MODEL = "unsloth/Qwen3-4B-Instruct-2507"
-LORA_MODEL = "michsethowusu/opani-coder"
+MODEL_ID = "michsethowusu/opani-coder_1b-merged-16bit"
 
 print("Loading tokenizer…")
-tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 
-print("Loading base model…")
-base_model = AutoModelForCausalLM.from_pretrained(
-    BASE_MODEL,
+print("Loading model…")
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
     torch_dtype=torch.float16,
     device_map="auto",
     low_cpu_mem_usage=True,
     trust_remote_code=True
 )
 
-print("Loading LoRA adapters…")
-try:
-    # Try loading with the correct device_map handling
-    model = PeftModel.from_pretrained(
-        base_model, 
-        LORA_MODEL,
-        device_map="auto",
-        torch_dtype=torch.float16
-    )
-    print("Merging LoRA adapters…")
-    model = model.merge_and_unload()
-except Exception as e:
-    print(f"Error loading LoRA model: {e}")
-    print("Attempting alternative loading method…")
-    # Alternative: Load the model differently
-    from peft import AutoPeftModelForCausalLM
-    model = AutoPeftModelForCausalLM.from_pretrained(
-        LORA_MODEL,
-        torch_dtype=torch.float16,
-        device_map="auto",
-        low_cpu_mem_usage=True,
-        trust_remote_code=True
-    )
-    model = model.merge_and_unload()
-
 print("Model ready!")
 
 # ------------------------------------------------------------------
@@ -119,8 +92,8 @@ def bot_respond(history, temperature, top_p, top_k, max_tokens):
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown(
         """
-        # 🇬🇭 Opani Coder
-        A fine-tuned Qwen3-4B model for coding assistance in Twi.
+        # 🇬🇭 Opani Coder 1B
+        A fine-tuned Llama 3.2 1B model (16-bit) for coding assistance in Twi.
         Ask me anything about programming, and I'll help you out!
         """
     )
@@ -128,7 +101,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     chatbot = gr.Chatbot(
         height=500,
         label="Chat History",
-        type="messages",  # NEW FORMAT
+        type="messages",
         avatar_images=(None, "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png"),
     )
 
@@ -144,7 +117,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     with gr.Accordion("⚙️ Generation Parameters", open=False):
         gr.Markdown("*Adjust these settings to control the response style*")
         temperature = gr.Slider(0.1, 2.0, 0.7, step=0.1, label="Temperature")
-        top_p = gr.Slider(0.1, 1.0, 0.8, step=0.05, label="Top P")
+        top_p = gr.Slider(0.1, 1.0, 0.9, step=0.05, label="Top P")
         top_k = gr.Slider(1, 100, 20, step=1, label="Top K")
         max_tokens = gr.Slider(64, 2048, 512, step=64, label="Max Tokens")
 
@@ -195,8 +168,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
         - **Code generation**: temperature 0.5-0.7
 
         ### 📝 About This Model
-        Fine-tuned Qwen3-4B with Unsloth for coding assistance in Twi.  
-        **Model**: [michsethowusu/opani-coder](https://huggingface.co/michsethowusu/opani-coder)
+        Fine-tuned Llama 3.2 1B (16-bit full model) for coding assistance in Twi.  
+        **Model**: [michsethowusu/opani-coder_1b-merged-16bit](https://huggingface.co/michsethowusu/opani-coder_1b-merged-16bit)
         """
     )