Update app.py

app.py

@@ -1,64 +1,124 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
+# app.py
+import streamlit as st
+from langdetect import detect
+from translator import translate
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import os
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# Config
+LOCAL_MODEL_DIR = "guvi_gpt2_finetuned"  # local path if you fine-tuned locally
+USE_LOCAL = os.path.isdir(LOCAL_MODEL_DIR)
+HF_MODEL_ID = "your-username/guvi_gpt2_finetuned"  # if you use HF InferenceClient instead
 
+# Load tokenizer + model (local)
+@st.cache_resource
+def load_local_model(model_dir):
+    tokenizer = AutoTokenizer.from_pretrained(model_dir)
+    model = AutoModelForCausalLM.from_pretrained(model_dir)
+    model.eval()
+    if torch.cuda.is_available():
+        model.to("cuda")
+    return tokenizer, model
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+if USE_LOCAL:
+    tokenizer, model = load_local_model(LOCAL_MODEL_DIR)
+else:
+    tokenizer = None
+    model = None
+    # if you want use HF Inference API, you'll use streaming_inference.py instead
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+st.set_page_config(page_title="GUVI Multilingual Chatbot", layout="wide")
+st.title("GUVI Multilingual Chatbot")
 
-    messages.append({"role": "user", "content": message})
+# Sidebar controls
+with st.sidebar:
+    st.header("Settings")
+    max_new_tokens = st.slider("Max new tokens", 50, 512, 200)
+    temperature = st.slider("Temperature", 0.1, 1.5, 0.7)
+    top_p = st.slider("Top-p", 0.1, 1.0, 0.95)
+    use_auto_detect = st.checkbox("Auto-detect input language (recommended)", value=True)
+    source_lang_override = st.text_input("Force source lang code (e.g. tam_Taml) — leave empty for auto-detect")
 
-    response = ""
+# Chat UI
+if "history" not in st.session_state:
+    st.session_state.history = []
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+def detect_lang_code(text):
+    # detect returns 'en', 'hi', etc. We need NLLB code mapping
+    try:
+        short = detect(text)
+    except:
+        short = "en"
+    # basic mapping - expand as required
+    mapping = {
+        "en": "eng_Latn",
+        "hi": "hin_Deva",
+        "ta": "tam_Taml",
+        "te": "tel_Telu",
+        "kn": "kan_Knda",
+        "mr": "mar_Deva",
+        "bn": "ben_Beng",
+        # add more mappings
+    }
+    return mapping.get(short, "eng_Latn")
 
-        response += token
-        yield response
+def generate_reply(prompt_en):
+    """
+    Generate reply in English using local model or HF inference API.
+    Returns English text.
+    """
+    if USE_LOCAL and model is not None:
+        input_text = "Q: " + prompt_en + "\nA:"
+        inputs = tokenizer.encode(input_text, return_tensors="pt")
+        if torch.cuda.is_available():
+            inputs = inputs.to("cuda")
+            model.to("cuda")
+        outputs = model.generate(inputs, max_new_tokens=max_new_tokens, temperature=temperature, top_p=top_p, pad_token_id=tokenizer.eos_token_id)
+        reply = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        # extract the answer portion after "A:"
+        if "\nA:" in reply:
+            reply = reply.split("\nA:")[-1].strip()
+        return reply
+    else:
+        # If not local, fallback to instructing user to use HF Inference API
+        return "Model not loaded locally. Push model to Hugging Face Hub and switch to InferenceClient or run locally."
 
+# Input area
+user_input = st.text_area("Enter your message", height=120)
+if st.button("Send"):
+    if not user_input.strip():
+        st.warning("Please enter a message.")
+    else:
+        # language detection / translation
+        if source_lang_override.strip():
+            src_code = source_lang_override.strip()
+        elif use_auto_detect:
+            src_code = detect_lang_code(user_input)
+        else:
+            src_code = "eng_Latn"
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+        needs_translation = src_code != "eng_Latn"
+        if needs_translation:
+            # translate to English
+            st.session_state.history.append(("User (original): " + user_input, ""))
+            prompt_en = translate(user_input, src_lang_code=src_code, tgt_lang_code="eng_Latn")
+        else:
+            prompt_en = user_input
 
+        # get english answer
+        reply_en = generate_reply(prompt_en)
 
-if __name__ == "__main__":
-    demo.launch()
+        # translate back if needed
+        if needs_translation:
+            reply_local = translate(reply_en, src_lang_code="eng_Latn", tgt_lang_code=src_code)
+        else:
+            reply_local = reply_en
+
+        st.session_state.history.append((user_input, reply_local))
+
+# Show chat history
+for user_msg, bot_msg in reversed(st.session_state.history):
+    st.markdown(f"**User:** {user_msg}")
+    st.markdown(f"**Bot:** {bot_msg}")
+    st.write("---")