Update app.py

app.py

@@ -4,43 +4,54 @@ import gradio as gr
 
 # Download the model from Hugging Face
 model_name = "johnpaulbin/articulate-V1-Q8_0-GGUF"
-model_file = "articulate-V1-Q8_0.gguf"  # Replace with the actual GGUF file name from the repository
+model_file = "articulate-V1-q8_0.gguf"  # Verify the exact file name in the repository
 model_path = hf_hub_download(repo_id=model_name, filename=model_file)
 
-# Initialize the Llama model with llama-cpp-python
+# Initialize the Llama model
 llm = Llama(
     model_path=model_path,
-    n_ctx=1024,           # Context length (adjust as needed)
-    n_threads=2,         # Number of CPU threads
-    n_gpu_layers=0        # Run on CPU only (no GPU in free Spaces tier)
+    n_ctx=1028,      # Context length
+    n_threads=2,    # Number of CPU threads
+    n_gpu_layers=0   # Run on CPU only
 )
 
-# Define the chat function for Gradio
-def chat(message, history):
-    # Build the message list with history and current user input
-    messages = []
-    for user_msg, assistant_msg in history:
-        messages.append({"role": "user", "content": user_msg})
-        messages.append({"role": "assistant", "content": assistant_msg})
-    messages.append({"role": "user", "content": message})
+# Define the translation function
+def translate(direction, text):
+    # Determine source and target languages based on direction
+    if direction == "English to Spanish":
+        source_lang = "ENGLISH"
+        target_lang = "SPANISH"
+    elif direction == "Spanish to English":
+        source_lang = "SPANISH"
+        target_lang = "ENGLISH"
+    else:
+        return "Invalid direction"
     
-    # Perform inference with greedy decoding
-    response = llm.create_chat_completion(
-        messages=messages,
-        max_tokens=100,    # Limit output length
-        top_k=1,          # Greedy decoding: select the top token
-        temperature=0.01  # Low temperature for determinism (top_k=1 is sufficient)
+    # Construct the prompt for raw completion
+    prompt = f"[{source_lang}]{text}[{target_lang}]"
+    
+    # Generate completion with deterministic settings (greedy decoding)
+    response = llm.create_completion(
+        prompt,
+        max_tokens=200,    # Limit output length
+        temperature=0,     # Greedy decoding
+        top_k=1            # Select the most probable token
     )
     
     # Extract and return the generated text
-    generated_text = response['choices'][0]['message']['content']
-    return generated_text
+    return response['choices'][0]['text'].strip()
 
-# Create the Gradio ChatInterface
-iface = gr.ChatInterface(
-    fn=chat,
-    title="Articulate V1 Chatbot",
-    description="Chat with the Articulate V1 model (Llama 3-based) using greedy decoding."
+# Define the Gradio interface
+direction_options = ["English to Spanish", "Spanish to English"]
+iface = gr.Interface(
+    fn=translate,
+    inputs=[
+        gr.Dropdown(choices=direction_options, label="Translation Direction"),
+        gr.Textbox(lines=5, label="Input Text")
+    ],
+    outputs=gr.Textbox(lines=5, label="Translation"),
+    title="Translation App",
+    description="Translate text between English and Spanish using the Articulate V1 model."
 )
 
 # Launch the app