Update app.py

app.py

@@ -1,7 +1,9 @@
-import streamlit as st
+import os
 import requests
 import torch
 import scipy.io.wavfile
+import streamlit as st
+from io import BytesIO
 from transformers import (
     AutoTokenizer, 
     AutoModelForCausalLM, 
@@ -9,26 +11,8 @@ from transformers import (
     AutoProcessor, 
     MusicgenForConditionalGeneration
 )
-from io import BytesIO
 from streamlit_lottie import st_lottie
 
-
-import os
-from transformers import AutoTokenizer, AutoModelForCausalLM
-
-my_token = os.getenv("HF_TOKEN")
-
-tokenizer = AutoTokenizer.from_pretrained(
-    "meta-llama/Llama-3-70B-Instruct", 
-    use_auth_token=my_token
-)
-model = AutoModelForCausalLM.from_pretrained(
-    "meta-llama/Llama-3-70B-Instruct",
-    use_auth_token=my_token,
-    torch_dtype=torch.float16,
-    device_map="auto"
-)
-
 # ---------------------------------------------------------------------
 # 1) PAGE CONFIG
 # ---------------------------------------------------------------------
@@ -131,17 +115,20 @@ lottie_animation = load_lottie_url(LOTTIE_URL)
 # 4) LOAD LLAMA 3 (GATED MODEL) - WITH use_auth_token
 # ---------------------------------------------------------------------
 @st.cache_resource
-def load_llama_pipeline(model_id: str, device: str):
+def load_llama_pipeline(model_id: str, device: str, token: str):
     """
-    Load the Llama 3 model from Hugging Face.
-    Requires huggingface-cli login if model is gated.
+    Load the Llama 3 model from Hugging Face with a user token.
+    token: The HF access token from environment or secrets.
     """
-    tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=True)
+    tokenizer = AutoTokenizer.from_pretrained(
+        model_id, 
+        use_auth_token=token
+    )
     model = AutoModelForCausalLM.from_pretrained(
         model_id,
+        use_auth_token=token,
         torch_dtype=torch.float16 if device == "auto" else torch.float32,
-        device_map=device,
-        use_auth_token=True
+        device_map=device
     )
     text_gen_pipeline = pipeline(
         "text-generation",
@@ -208,8 +195,9 @@ with col1:
     st.markdown(
         """
         Create **radio imaging promos** and **jingles** with Llama 3 + MusicGen.  
-        **Note**: You must have access to `"meta-llama/Llama-3-70B-Instruct"` on Hugging Face, 
-        and be logged in via `huggingface-cli login`.
+        **Note**: 
+        - You must have access to `"meta-llama/Llama-3-70B-Instruct"` on Hugging Face.
+        - You must provide your HF token in the environment (e.g., HF_TOKEN).
         """
     )
 with col2:
@@ -245,13 +233,19 @@ with col_device:
         help="If you have GPU, 'auto' tries to use it; CPU might be slow."
     )
 
+# Grab your token from environment
+my_token = os.getenv("HF_TOKEN")
+if not my_token:
+    st.error("No HF_TOKEN found. Please set it in your HF Space secrets or environment variables.")
+    st.stop()
+
 if st.button("📝 Generate Promo Script"):
     if not prompt.strip():
         st.error("Please type some concept first.")
     else:
         with st.spinner("Generating script with Llama 3..."):
             try:
-                llm_pipeline = load_llama_pipeline(llama_model_id, device_option)
+                llm_pipeline = load_llama_pipeline(llama_model_id, device_option, my_token)
                 final_script = generate_radio_script(prompt, llm_pipeline)
                 st.session_state["final_script"] = final_script
                 st.success("Promo script generated!")
@@ -301,7 +295,7 @@ st.markdown(
     """
     <div class="footer-note">
     © 2025 Radio Imaging with Llama 3 – Built using Hugging Face & Streamlit. <br>
-    Log in via <code>huggingface-cli</code> and ensure access to <strong>meta-llama/Llama-3-70B-Instruct</strong>.
+    Log in or provide <code>HF_TOKEN</code> and ensure access to <strong>meta-llama/Llama-3-70B-Instruct</strong>.
     </div>
     """,
     unsafe_allow_html=True