Update app.py

app.py

@@ -5,23 +5,22 @@ import torch
 from langgraph.graph import StateGraph, START, END
 from langchain.schema import HumanMessage
 from langchain_groq import ChatGroq
-from langsmith import traceable
+from langsmith import traceable  # ✅ Added LangSmith for Debugging
 from typing import TypedDict
 from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
 
-# Load API Keys
-GROQ_API_KEY = os.getenv("GROQ_API_KEY")
+# ✅ Load API keys from Hugging Face Secrets
+GROQ_API_KEY = os.getenv("GROQ_API_KEY")  
 LANGSMITH_API_KEY = os.getenv("LANGSMITH_API_KEY")
 
-# Set environment variables
-os.environ["GROQ_API_KEY"] = GROQ_API_KEY
+# ✅ Set LangSmith Debugging
 os.environ["LANGCHAIN_TRACING_V2"] = "true"
 os.environ["LANGCHAIN_API_KEY"] = LANGSMITH_API_KEY
 
-# Initialize Groq LLM
+# ✅ Initialize Groq LLM (for content generation)
 llm = ChatGroq(groq_api_key=GROQ_API_KEY, model_name="mixtral-8x7b-32768")
 
-# Define State for LangGraph
+# ✅ Define State for LangGraph
 class State(TypedDict):
     topic: str
     titles: list
@@ -32,19 +31,19 @@ class State(TypedDict):
     tone: str
     language: str
 
-# Function to generate multiple blog titles
-@traceable(name="Generate Titles")
+# ✅ Function to generate multiple blog titles using Groq
+@traceable(name="Generate Titles")  # ✅ Debugging with LangSmith
 def generate_titles(data):
     topic = data.get("topic", "")
     prompt = f"Generate three short and catchy blog titles for the topic: {topic}. Each title should be under 10 words. Separate them with new lines."
     
     response = llm([HumanMessage(content=prompt)])
-    titles = response.content.strip().split("\n")  # Get three titles as a list
+    titles = response.content.strip().split("\n")  
     
-    return {"titles": titles}  # No default selection
+    return {"titles": titles, "selected_title": titles[0]}  
 
-# Function to generate blog content with user-selected title
-@traceable(name="Generate Content")
+# ✅ Function to generate blog content with tone using Groq
+@traceable(name="Generate Content")  # ✅ Debugging with LangSmith
 def generate_content(data):
     title = data.get("selected_title", "")
     tone = data.get("tone", "Neutral")
@@ -53,8 +52,8 @@ def generate_content(data):
     response = llm([HumanMessage(content=prompt)])
     return {"content": response.content.strip()}
 
-# Function to generate summary
-@traceable(name="Generate Summary")
+# ✅ Function to generate summary using Groq
+@traceable(name="Generate Summary")  # ✅ Debugging with LangSmith
 def generate_summary(data):
     content = data.get("content", "")
     prompt = f"Summarize this blog post in a short and engaging way: {content}"
@@ -62,12 +61,16 @@ def generate_summary(data):
     response = llm([HumanMessage(content=prompt)])
     return {"summary": response.content.strip()}
 
-# Load translation model
-model_name = "facebook/nllb-200-distilled-600M"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+# ✅ Load translation model (NLLB-200)
+def load_translation_model():
+    model_name = "facebook/nllb-200-distilled-600M"
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+    return tokenizer, model
 
-# Language codes
+tokenizer, model = load_translation_model()
+
+# ✅ Language codes for NLLB-200
 language_codes = {
     "English": "eng_Latn",
     "Hindi": "hin_Deva",
@@ -76,74 +79,109 @@ language_codes = {
     "French": "fra_Latn"
 }
 
-# Function to translate blog content
-@traceable(name="Translate Content")
+# ✅ Function to translate blog content using NLLB-200
+@traceable(name="Translate Content")  # ✅ Debugging with LangSmith
 def translate_content(data):
     content = data.get("content", "")
     language = data.get("language", "English")
 
     if language == "English":
-        return {"translated_content": content}  # No translation needed
+        return {"translated_content": content}
+
+    tgt_lang = language_codes.get(language, "eng_Latn")  
+
+    # ✅ Split content into smaller chunks (Avoids token limit issues)
+    max_length = 512  
+    sentences = content.split(". ")  
+    chunks = []
+    current_chunk = ""
+
+    for sentence in sentences:
+        if len(current_chunk) + len(sentence) < max_length:
+            current_chunk += sentence + ". "
+        else:
+            chunks.append(current_chunk.strip())
+            current_chunk = sentence + ". "
 
-    tgt_lang = language_codes.get(language, "eng_Latn")
-    inputs = tokenizer(content, return_tensors="pt", padding=True, truncation=True)
-    translated_tokens = model.generate(**inputs, forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_lang))
-    translated_text = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
+    if current_chunk:
+        chunks.append(current_chunk.strip())
 
-    return {"translated_content": translated_text.strip()}
+    # ✅ Translate each chunk separately and combine results
+    translated_chunks = []
+    for chunk in chunks:
+        inputs = tokenizer(chunk, return_tensors="pt", padding=True, truncation=True)
+        translated_tokens = model.generate(**inputs, forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_lang))
+        translated_text = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
+        translated_chunks.append(translated_text.strip())
 
-# Create LangGraph Workflow
+    full_translation = " ".join(translated_chunks)
+
+    return {"translated_content": full_translation}
+
+# ✅ Create LangGraph Workflow
 def make_blog_generation_graph():
+    """Create a LangGraph workflow for Blog Generation"""
     graph_workflow = StateGraph(State)
-    
+
+    # Define Nodes
     graph_workflow.add_node("title_generation", generate_titles)
     graph_workflow.add_node("content_generation", generate_content)
     graph_workflow.add_node("summary_generation", generate_summary)
-    graph_workflow.add_node("translation", translate_content)
-    
+    graph_workflow.add_node("translation", translate_content)  
+
+    # Define Execution Order
     graph_workflow.add_edge(START, "title_generation")
     graph_workflow.add_edge("title_generation", "content_generation")
     graph_workflow.add_edge("content_generation", "summary_generation")
     graph_workflow.add_edge("content_generation", "translation")
     graph_workflow.add_edge("summary_generation", END)
     graph_workflow.add_edge("translation", END)
-    
+
     return graph_workflow.compile()
 
-# Gradio Interface
-def generate_blog(topic, tone, language, selected_title):
+# ✅ Function to generate blog content (Fixed)
+def generate_blog(topic, tone, language):
     try:
         if not topic:
             return "⚠️ Please enter a topic.", "", "", "", ""
-        
+
         blog_agent = make_blog_generation_graph()
-        result = blog_agent.invoke({"topic": topic, "tone": tone, "language": language, "selected_title": selected_title})
-        
-        return result["titles"], selected_title, result["content"], result["summary"], result["translated_content"]
+        result = blog_agent.invoke({"topic": topic, "tone": tone, "language": language})
+
+        return result["titles"], result["selected_title"], result["content"], result["summary"], result["translated_content"]
 
     except Exception as e:
         error_message = f"⚠️ Error: {str(e)}\n{traceback.format_exc()}"
         return error_message, "", "", "", ""
 
+# ✅ Gradio UI
 with gr.Blocks() as app:
-    gr.Markdown("""
-    ### 🌍 Why Translate?
-    We provide translation to make the blog content **accessible to a global audience**.
-    - 🗣️ **Multilingual Support** – Read blogs in your preferred language.
-    - 🌎 **Expand Reach** – Reach international readers.
-    - ✅ **Better Understanding** – Enjoy content in a language you're comfortable with.
-    - 🤖 **AI-Powered Accuracy** – Uses advanced AI models for precise translation.
-    """)
-    
-    topic_input = gr.Textbox(label="Enter a topic")
-    tone_input = gr.Dropdown(["Neutral", "Formal", "Casual"], label="Select Blog Tone")
-    language_input = gr.Dropdown(["English", "Hindi", "Telugu", "Spanish", "French"], label="Translate Blog To")
-    title_output = gr.Dropdown(label="Select Blog Title")
-    content_output = gr.Textbox(label="Generated Blog Content")
-    summary_output = gr.Textbox(label="Blog Summary")
-    translation_output = gr.Textbox(label="Translated Blog Content")
-
-    generate_button = gr.Button("Generate Blog")
-    generate_button.click(generate_blog, [topic_input, tone_input, language_input, title_output], [title_output, content_output, summary_output, translation_output])
-    
+    gr.Markdown(
+        """
+        ### 🌍 Why Translate?  
+        - 🗣️ **Multilingual Support**  
+        - 🌎 **Expand Reach**  
+        - ✅ **Better Understanding**  
+        - 🤖 **AI-Powered Accuracy**  
+        """
+    )
+
+    gr.Interface(
+        fn=generate_blog,
+        inputs=[
+            gr.Textbox(label="Enter a topic for your blog"),
+            gr.Dropdown(["Neutral", "Formal", "Casual", "Persuasive", "Humorous"], label="Select Blog Tone", value="Neutral"),
+            gr.Dropdown(["English", "Hindi", "Telugu", "Spanish", "French"], label="Translate Blog To", value="English"),
+        ],
+        outputs=[
+            gr.Textbox(label="Suggested Blog Titles"),
+            gr.Textbox(label="Selected Blog Title"),
+            gr.Textbox(label="Generated Blog Content"),
+            gr.Textbox(label="Blog Summary"),
+            gr.Textbox(label="Translated Blog Content"),
+        ],
+        title="🚀 AI-Powered Blog Generator",
+    )
+
+# ✅ Launch the Gradio App
 app.launch(share=True)