XLA Kompelirung fehlgeschlagen

[666Teufel]

=== TensorFlow Diagnose Tool ===

TensorFlow Version: 2.20.0-dev0+selfbuilt

Verfügbare GPUs:
✅ PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')

Teste XLA-Kompilierung:
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1748720218.610720 27 gpu_device.cc:2019] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 13272 MB memory: -> device: 0, name: NVIDIA GeForce RTX 5060 Ti, pci bus id: 0000:2b:00.0, compute capability: 12.0
2025-05-31 19:36:59.180751: I external/local_xla/xla/service/service.cc:153] XLA service 0x173594d0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2025-05-31 19:36:59.180792: I external/local_xla/xla/service/service.cc:161] StreamExecutor device (0): NVIDIA GeForce RTX 5060 Ti, Compute Capability 12.0
2025-05-31 19:36:59.219742: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:269] disabling MLIR crash reproducer, set env var MLIR_CRASH_REPRODUCER_DIRECTORY to enable.
I0000 00:00:1748720220.035048 27 cuda_dnn.cc:529] Loaded cuDNN version 90800
2025-05-31 19:37:00.138606: W tensorflow/core/framework/op_kernel.cc:1855] OP_REQUIRES failed at xla_ops.cc:590 : UNAVAILABLE: No PTX compilation provider is available. Neither ptxas/nvlink nor nvjtlink is available. As a fallback you can enable JIT compilation in the CUDA driver via the flag --xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found. Details:

• Has NvJitLink support: LibNvJitLink is not supported (disabled during compilation).
• Has NvPtxCompiler support: LibNvPtxCompiler is not supported (disabled during compilation).
• Parallel compilation support is desired: 0
• ptxas_path: Couldn't find a suitable version of ptxas. The following locations were considered: ./cuda_sdk_lib/bin/ptxas, /workspace/venv/bin/ptxas, /workspace/venv/bin/ptxas, /usr/local/cuda/bin/ptxas, /usr/local/sbin/ptxas, /usr/local/bin/ptxas, /usr/sbin/ptxas, /usr/bin/ptxas, /sbin/ptxas, /bin/ptxas, diagnose.py.runfiles/cuda_nvcc/bin/ptxas, diagnos/cuda_nvcc/bin/ptxas, bin/ptxas, /usr/local/cuda/bin/ptxas, /opt/cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/ptxas
• ptxas_version: Couldn't find a suitable version of ptxas. The following locations were considered: ./cuda_sdk_lib/bin/ptxas, /workspace/venv/bin/ptxas, /workspace/venv/bin/ptxas, /usr/local/cuda/bin/ptxas, /usr/local/sbin/ptxas, /usr/local/bin/ptxas, /usr/sbin/ptxas, /usr/bin/ptxas, /sbin/ptxas, /bin/ptxas, diagnose.py.runfiles/cuda_nvcc/bin/ptxas, diagnos/cuda_nvcc/bin/ptxas, bin/ptxas, /usr/local/cuda/bin/ptxas, /opt/cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/ptxas
• nvlink_path: Couldn't find a suitable version of nvlink. The following locations were considered: ./cuda_sdk_lib/bin/nvlink, /workspace/venv/bin/nvlink, /workspace/venv/bin/nvlink, /usr/local/cuda/bin/nvlink, /usr/local/sbin/nvlink, /usr/local/bin/nvlink, /usr/sbin/nvlink, /usr/bin/nvlink, /sbin/nvlink, /bin/nvlink, diagnose.py.runfiles/cuda_nvcc/bin/nvlink, diagnos/cuda_nvcc/bin/nvlink, bin/nvlink, /usr/local/cuda/bin/nvlink, /opt/cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/nvlink
• nvlink_version: Couldn't find a suitable version of nvlink. The following locations were considered: ./cuda_sdk_lib/bin/nvlink, /workspace/venv/bin/nvlink, /workspace/venv/bin/nvlink, /usr/local/cuda/bin/nvlink, /usr/local/sbin/nvlink, /usr/local/bin/nvlink, /usr/sbin/nvlink, /usr/bin/nvlink, /sbin/nvlink, /bin/nvlink, diagnose.py.runfiles/cuda_nvcc/bin/nvlink, diagnos/cuda_nvcc/bin/nvlink, bin/nvlink, /usr/local/cuda/bin/nvlink, /opt/cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/nvlink
• Driver compilation is enabled: 0

❌ XLA-Kompilierung fehlgeschlagen!
Fehler:
No PTX compilation provider is available. Neither ptxas/nvlink nor nvjtlink is available. As a fallback you can enable JIT compilation in the CUDA driver via the flag --xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found. Details:

• Has NvJitLink support: LibNvJitLink is not supported (disabled during compilation).
• Has NvPtxCompiler support: LibNvPtxCompiler is not supported (disabled during compilation).
• Parallel compilation support is desired: 0
• ptxas_path: Couldn't find a suitable version of ptxas. The following locations were considered: ./cuda_sdk_lib/bin/ptxas, /workspace/venv/bin/ptxas, /workspace/venv/bin/ptxas, /usr/local/cuda/bin/ptxas, /usr/local/sbin/ptxas, /usr/local/bin/ptxas, /usr/sbin/ptxas, /usr/bin/ptxas, /sbin/ptxas, /bin/ptxas, diagnose.py.runfiles/cuda_nvcc/bin/ptxas, diagnos/cuda_nvcc/bin/ptxas, bin/ptxas, /usr/local/cuda/bin/ptxas, /opt/cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/ptxas
• ptxas_version: Couldn't find a suitable version of ptxas. The following locations were considered: ./cuda_sdk_lib/bin/ptxas, /workspace/venv/bin/ptxas, /workspace/venv/bin/ptxas, /usr/local/cuda/bin/ptxas, /usr/local/sbin/ptxas, /usr/local/bin/ptxas, /usr/sbin/ptxas, /usr/bin/ptxas, /sbin/ptxas, /bin/ptxas, diagnose.py.runfiles/cuda_nvcc/bin/ptxas, diagnos/cuda_nvcc/bin/ptxas, bin/ptxas, /usr/local/cuda/bin/ptxas, /opt/cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/ptxas, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/ptxas
• nvlink_path: Couldn't find a suitable version of nvlink. The following locations were considered: ./cuda_sdk_lib/bin/nvlink, /workspace/venv/bin/nvlink, /workspace/venv/bin/nvlink, /usr/local/cuda/bin/nvlink, /usr/local/sbin/nvlink, /usr/local/bin/nvlink, /usr/sbin/nvlink, /usr/bin/nvlink, /sbin/nvlink, /bin/nvlink, diagnose.py.runfiles/cuda_nvcc/bin/nvlink, diagnos/cuda_nvcc/bin/nvlink, bin/nvlink, /usr/local/cuda/bin/nvlink, /opt/cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/nvlink
• nvlink_version: Couldn't find a suitable version of nvlink. The following locations were considered: ./cuda_sdk_lib/bin/nvlink, /workspace/venv/bin/nvlink, /workspace/venv/bin/nvlink, /usr/local/cuda/bin/nvlink, /usr/local/sbin/nvlink, /usr/local/bin/nvlink, /usr/sbin/nvlink, /usr/bin/nvlink, /sbin/nvlink, /bin/nvlink, diagnose.py.runfiles/cuda_nvcc/bin/nvlink, diagnos/cuda_nvcc/bin/nvlink, bin/nvlink, /usr/local/cuda/bin/nvlink, /opt/cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../nvidia/cuda_nvcc/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../cuda/bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../bin/nvlink, /workspace/venv/lib/python3.12/site-packages/tensorflow/python/platform/../../../../../../../bin/nvlink
• Driver compilation is enabled: 0

Executing non-communication op <__inference_xla_test_fn_8> originally returned UnavailableError, and was replaced by InternalError to avoid invoking TF network error handling logic. [Op:__inference_Executing non-communication op <__inference_xla_test_fn_8> originally returned UnavailableError, and was replaced by InternalError to avoid invoking TF network error handling logic. [Op:__inference_xla_test_fn_8]

Teste einfache GPU-Mathematik:
✅ MatMul erfolgreich auf GPU ausgeführt.
Ergebnis:
[[ 7. 10.]
[15. 22.]]

CUDA-Umgebungsvariablen:
CUDA_HOME/PATH: None
XLA_FLAGS: (nicht gesetzt)

=== Diagnose abgeschlossen ===

[666Teufel]

test Programm: import tensorflow as tf
from tensorflow.python.framework import test_util
import os

print("=== TensorFlow Diagnose Tool ===\n")

TensorFlow-Version

print(f"TensorFlow Version: {tf.version}")

Prüfe GPU-Erkennung

print("\n> Verfügbare GPUs:")
gpus = tf.config.list_physical_devices('GPU')
if not gpus:
print("❌ Keine GPU gefunden!")
else:
for gpu in gpus:
print(f"✅ {gpu}")

XLA aktivieren

print("\n> Teste XLA-Kompilierung:")
@tf.function(jit_compile=True)
def xla_test_fn(x):
return tf.math.sin(x) + tf.math.cos(x)

try:
result = xla_test_fn(tf.constant([1.0, 2.0, 3.0]))
print("✅ XLA-Kompilierung erfolgreich!")
except Exception as e:
print("❌ XLA-Kompilierung fehlgeschlagen!")
print(f"Fehler:\n{e}")

CUDA-spezifischer Test

print("\n> Teste einfache GPU-Mathematik:")
try:
with tf.device('/GPU:0'):
a = tf.constant([[1.0, 2.0], [3.0, 4.0]])
b = tf.matmul(a, a)
print("✅ MatMul erfolgreich auf GPU ausgeführt.")
print("Ergebnis:\n", b.numpy())
except RuntimeError as e:
print("❌ Fehler bei GPU-Nutzung:")
print(e)
except Exception as e:
print("❌ Unerwarteter Fehler:")
print(e)

Optional: CUDA-Umgebungsvariablen prüfen

print("\n> CUDA-Umgebungsvariablen:")
cuda_dir = os.environ.get("CUDA_HOME") or os.environ.get("CUDA_PATH")
xla_flag = os.environ.get("XLA_FLAGS")
print(f"CUDA_HOME/PATH: {cuda_dir}")
print(f"XLA_FLAGS: {xla_flag if xla_flag else '(nicht gesetzt)'}")

print("\n=== Diagnose abgeschlossen ===")