Enhance ECHO_MODE functionality and logging for improved testing and environment configuration

app.py

@@ -20,7 +20,23 @@ HEALTH_PORT = int(os.getenv("HEALTH_PORT", "8080"))
 GRADIO_HOST = os.getenv("GRADIO_HOST", "0.0.0.0")
 GRADIO_PORT = int(os.getenv("GRADIO_PORT", "7860"))
 DEFAULT_MAX_NEW_TOKENS = int(os.getenv("DEFAULT_MAX_NEW_TOKENS", "128"))
-ECHO_MODE = os.getenv("ECHO_MODE", "off").lower()  # 'off'|'echo'|'useless'
+ECHO_MODE = os.getenv("ECHO_MODE", "off").lower()  # default env; UI can override at runtime
+OFFLINE_FLAG = os.getenv("OFFLINE", "0").lower() in {"1", "true", "yes"}
+if OFFLINE_FLAG:
+    os.environ["HF_HUB_OFFLINE"] = "1"
+
+def _log_cache_env():
+    try:
+        import huggingface_hub as _hub
+        hub_cache = getattr(_hub.constants, 'HF_HUB_CACHE', None)
+    except Exception:
+        hub_cache = None
+    logging.info(
+        "Cache config: HF_HOME=%s TRANSFORMERS_CACHE=%s HF_HUB_OFFLINE=%s hub_cache=%s",
+        os.getenv("HF_HOME"), os.getenv("TRANSFORMERS_CACHE"), os.getenv("HF_HUB_OFFLINE"), hub_cache
+    )
+
+_log_cache_env()
 
 
 # ---------------- Logging ----------------
@@ -127,21 +143,22 @@ def predict(instruction: str,
             do_sample: bool,
             temperature: float,
             top_p: float,
-            num_return_sequences: int):
+            num_return_sequences: int,
+            mode: str):
     """Generate text using the cached pipeline and return output or error message."""
     if not instruction or not instruction.strip():
         return "⚠️ مهرباني وکړئ یوه لارښوونه ولیکئ."  # please provide an instruction
 
     # Fast path: echo/useless mode avoids loading large models during testing.
-    if ECHO_MODE in ("echo", "useless"):
-        # If echo mode, return exactly the combined prompt
+    active_mode = (mode or "").strip().lower() or ECHO_MODE
+    if active_mode in ("echo", "useless"):
         prompt = instruction.strip()
         if input_text and input_text.strip():
             prompt += "\n" + input_text.strip()
-        if ECHO_MODE == "echo":
+        if active_mode == "echo":
             return prompt
-        # If useless mode, return a simple placeholder text
-        return "This is a useless placeholder response."
+        else:
+            return "This is a useless placeholder response."
 
     # Build a simple prompt: instruction (+ input if provided)
     prompt = instruction.strip()
@@ -189,11 +206,13 @@ def predict(instruction: str,
 
 def build_ui():
     with gr.Blocks() as demo:
+        device_label = "GPU" if _detect_device() != -1 else "CPU"
         gr.Markdown(
-            """
+            f"""
             # ZamAI mT5 Pashto Demo
             اپلیکیشن  **ZamAI-mT5-Pashto** د پښتو لارښوونو لپاره.  
-            لاندې تنظیمات بدل کړئ او لارښوونه ولیکئ ترڅو ځواب ترلاسه کړئ.
+            **Device:** {device_label}  |  **Env Mode:** {ECHO_MODE}  |  **Offline:** {os.getenv('HF_HUB_OFFLINE','0')}  
+            که د موډ بدلول غواړئ لاندې د Mode selector څخه استفاده وکړئ.
             """
         )
 
@@ -213,6 +232,12 @@ def build_ui():
                 input_text = gr.Textbox(lines=2, placeholder="اختیاري متن...", label="متن")
                 output = gr.Textbox(label="ځواب", interactive=False, lines=8)
                 generate_btn = gr.Button("جوړول", variant="primary")
+                mode_selector = gr.Dropdown(
+                    choices=["off", "echo", "useless"],
+                    value=ECHO_MODE,
+                    label="Mode (off=real, echo=return prompt, useless=fixed)",
+                    interactive=True,
+                )
 
             with gr.Column(scale=1):
                 gr.Markdown("### د تولید تنظیمات")
@@ -227,7 +252,7 @@ def build_ui():
 
         generate_btn.click(
             fn=predict,
-            inputs=[instruction_textbox, input_text, max_new_tokens, num_beams, do_sample, temperature, top_p, num_return_sequences],
+            inputs=[instruction_textbox, input_text, max_new_tokens, num_beams, do_sample, temperature, top_p, num_return_sequences, mode_selector],
             outputs=output,
         )
 
@@ -242,4 +267,6 @@ if __name__ == "__main__":
         logger.exception("Failed to start health server")
 
     demo = build_ui()
-    demo.launch(server_name=GRADIO_HOST, server_port=GRADIO_PORT)
+    demo.launch(server_name=GRADIO_HOST, server_port=GRADIO_PORT)
+
+    logging.info("HF_HOME=%s TRANSFORMERS_CACHE=%s", os.getenv("HF_HOME"), os.getenv("TRANSFORMERS_CACHE"))