Update app.py

app.py

@@ -21,15 +21,14 @@ if not HF_TOKEN:
     logging.warning("HF_TOKEN environment variable not set. Packaging and uploading will fail.")
 
 api = HfApi()
-OUTPUT_DIR = "optimized_models"
+# Use the /tmp directory which is always writable in a container environment
+OUTPUT_DIR = "/tmp/optimized_models"
 os.makedirs(OUTPUT_DIR, exist_ok=True)
 
-# The Dockerfile guarantees these files exist, so we just define the paths.
 LLAMA_CPP_DIR = Path("llama.cpp")
 LLAMA_CPP_CONVERT_SCRIPT = LLAMA_CPP_DIR / "convert.py"
 LLAMA_CPP_QUANTIZE_SCRIPT = LLAMA_CPP_DIR / "quantize"
 
-# Verify that the build was successful during startup
 if not LLAMA_CPP_QUANTIZE_SCRIPT.exists():
     error_msg = "FATAL ERROR: llama.cpp binaries not found. The Docker build may have failed."
     logging.error(error_msg)
@@ -94,12 +93,10 @@ def stage_3_4_gguf_quantize(model_path_or_id: str, original_model_id: str, quant
     log_stream = "[STAGE 3 & 4] Converting to GGUF using llama.cpp...\n"
     run_id = datetime.now().strftime("%Y%m%d-%H%M%S")
     model_name = original_model_id.replace('/', '_')
-    # Use absolute paths for outputs to avoid issues with changing working directories
     gguf_path = os.path.abspath(os.path.join(OUTPUT_DIR, f"{model_name}-{run_id}-gguf"))
     os.makedirs(gguf_path, exist_ok=True)
     f16_gguf_path = os.path.join(gguf_path, "model-f16.gguf")
     quantized_gguf_path = os.path.join(gguf_path, "model.gguf")
-    # Use absolute path for model input if it's a local directory
     absolute_model_path = os.path.abspath(model_path_or_id) if os.path.exists(model_path_or_id) else model_path_or_id
     try:
         convert_command = ["python3", "convert.py", absolute_model_path, "--outfile", f16_gguf_path, "--outtype", "f16"]
@@ -182,7 +179,7 @@ def run_amop_pipeline(model_id: str, pipeline_type: str, do_prune: bool, prune_p
             raise ValueError("Invalid pipeline type selected.")
         full_log += log
         yield {final_output: "Packaging & Uploading (4/5)", log_output: full_log}
-        final_message, log = stage_5_package_and_upload(model_id, optimized_model_path, full_log, options)
+        final_message, log = stage_5_package_and_upload(model_id, optimized_path, full_log, options)
         full_log += log
         yield {final_output: gr.update(value="SUCCESS", label="Status"), log_output: full_log, success_box: gr.Markdown(f"✅ **Success!** Model available: [{repo_id_for_link}](https://huggingface.co/{repo_id_for_link})", visible=True), run_button: gr.Button(interactive=True, value="Run Optimization Pipeline", variant="primary"), analyze_button: gr.Button(interactive=True, value="Analyze Model")}
     except Exception as e: