Update app.py

app.py

@@ -24,41 +24,16 @@ api = HfApi()
 OUTPUT_DIR = "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"
 
-def setup_llama_cpp():
-    if not LLAMA_CPP_DIR.exists():
-        logging.info("Cloning llama.cpp repository...")
-        try:
-            subprocess.run(["git", "clone", "https://github.com/ggerganov/llama.cpp.git"], check=True, capture_output=True, text=True)
-            logging.info("llama.cpp cloned successfully.")
-        except subprocess.CalledProcessError as e:
-            raise RuntimeError(f"Failed to clone llama.cpp. Error: {e.stderr}")
-
-    requirements_path = LLAMA_CPP_DIR / "requirements.txt"
-    if requirements_path.exists():
-        logging.info("Installing llama.cpp Python dependencies...")
-        try:
-            subprocess.run(["pip", "install", "-r", str(requirements_path)], check=True, capture_output=True, text=True)
-            logging.info("llama.cpp Python dependencies installed successfully.")
-        except subprocess.CalledProcessError as e:
-            raise RuntimeError(f"Failed to install llama.cpp requirements. Error: {e.stderr}")
-
-    if not LLAMA_CPP_QUANTIZE_SCRIPT.exists():
-        logging.info("llama.cpp binaries not found. Building with CMake...")
-        try:
-            subprocess.run(["cmake", "."], cwd=str(LLAMA_CPP_DIR), check=True, capture_output=True, text=True)
-            subprocess.run(["cmake", "--build", "."], cwd=str(LLAMA_CPP_DIR), check=True, capture_output=True, text=True)
-            logging.info("llama.cpp binaries built successfully with CMake.")
-        except subprocess.CalledProcessError as e:
-            raise RuntimeError(f"Failed to build llama.cpp with CMake. Error: {e.stderr}")
-
-try:
-    setup_llama_cpp()
-except Exception as e:
-    logging.error(f"FATAL ERROR during llama.cpp setup: {e}", exc_info=True)
+# 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)
+    raise RuntimeError(error_msg)
 
 def stage_1_analyze_model(model_id: str):
     log_stream = "[STAGE 1] Analyzing model...\n"
@@ -119,10 +94,12 @@ 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('/', '_')
-    gguf_path = os.path.join(OUTPUT_DIR, f"{model_name}-{run_id}-gguf")
+    # 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.abspath(os.path.join(gguf_path, "model-f16.gguf"))
-    quantized_gguf_path = os.path.abspath(os.path.join(gguf_path, "model.gguf"))
+    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"]