Update app.py

app.py

@@ -21,13 +21,14 @@ if not HF_TOKEN:
     logging.warning("HF_TOKEN environment variable not set. Packaging and uploading will fail.")
 
 api = HfApi()
-# Use the /tmp directory which is always writable in a container environment
 OUTPUT_DIR = "/tmp/optimized_models"
 os.makedirs(OUTPUT_DIR, exist_ok=True)
 
-LLAMA_CPP_DIR = Path("llama.cpp")
+# Use an absolute path to the pre-built location in /opt
+LLAMA_CPP_DIR = Path("/opt/llama.cpp")
+# Binaries are in the 'build/bin' subdirectory from our out-of-source build
+LLAMA_CPP_QUANTIZE_SCRIPT = LLAMA_CPP_DIR / "build" / "bin" / "quantize"
 LLAMA_CPP_CONVERT_SCRIPT = LLAMA_CPP_DIR / "convert.py"
-LLAMA_CPP_QUANTIZE_SCRIPT = LLAMA_CPP_DIR / "quantize"
 
 if not LLAMA_CPP_QUANTIZE_SCRIPT.exists():
     error_msg = "FATAL ERROR: llama.cpp binaries not found. The Docker build may have failed."
@@ -99,8 +100,9 @@ def stage_3_4_gguf_quantize(model_path_or_id: str, original_model_id: str, quant
     quantized_gguf_path = os.path.join(gguf_path, "model.gguf")
     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"]
-        process = subprocess.run(convert_command, check=True, capture_output=True, text=True, cwd=str(LLAMA_CPP_DIR))
+        # The python script can be called directly using its absolute path.
+        convert_command = ["python3", str(LLAMA_CPP_CONVERT_SCRIPT), absolute_model_path, "--outfile", f16_gguf_path, "--outtype", "f16"]
+        process = subprocess.run(convert_command, check=True, capture_output=True, text=True)
         log_stream += f"Executing llama.cpp conversion script...\n{process.stdout}\n"
         if process.stderr: log_stream += f"[STDERR]\n{process.stderr}\n"
         quantize_map = {"q4_k_m": "Q4_K_M", "q5_k_m": "Q5_K_M", "q8_0": "Q8_0", "f16": "F16"}
@@ -110,8 +112,8 @@ def stage_3_4_gguf_quantize(model_path_or_id: str, original_model_id: str, quant
             os.rename(f16_gguf_path, quantized_gguf_path)
         else:
             log_stream += f"Quantizing FP16 GGUF to {target_quant_name}...\n"
-            quantize_command = ["./quantize", f16_gguf_path, quantized_gguf_path, target_quant_name]
-            process = subprocess.run(quantize_command, check=True, capture_output=True, text=True, cwd=str(LLAMA_CPP_DIR))
+            quantize_command = [str(LLAMA_CPP_QUANTIZE_SCRIPT), f16_gguf_path, quantized_gguf_path, target_quant_name]
+            process = subprocess.run(quantize_command, check=True, capture_output=True, text=True)
             log_stream += f"{process.stdout}\n"
             if process.stderr: log_stream += f"[STDERR]\n{process.stderr}\n"
             os.remove(f16_gguf_path)
@@ -179,7 +181,7 @@ 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_path, full_log, options)
+        final_message, log = stage_5_package_and_upload(model_id, optimized_model_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: