Attempt to add more resilience against failure

bench-TriLMs.py

@@ -74,7 +74,7 @@ def quantize(types: Sequence[str] = ALL_TYPES, sizes: Sequence[str] = MODEL_SIZE
                 )
                 logger.info("Running: %s", command)
                 ret = os.system(command)
-                if ret != 0:
+                if ret != 0 or target.is_file() and target.stat().st_size == 0:
                     logger.error("Failed to quantize to %s", target)
                 # Should it still continue?
 
@@ -113,8 +113,8 @@ def llama_bench(
                 command = [str(LLAMA_CPP_PATH / "build" / "bin" / "llama-bench")] + args
                 logger.info("Running: %s", " ".join(command))
                 result = subprocess.run(command, capture_output=True)
-                logger.debug(result.stderr.decode())
-                if result.returncode != 0:
+                logger.debug(result.stderr.decode(errors="ignore"))
+                if result.returncode != 0 or len(result.stdout) == 0:
                     logger.error("Failed to run %s", " ".join(command))
                     break;