Fehler beim Konvertieren von Huggingface -Datensatz in den TensorFlow -Datensatz während der Verwendung von CUDA ⇐ Python

[Anonymous]

Ich versuche, Fonetuning mit einem Datensatz von Suggingface zu erledigen. Ich bekomme immer wieder InternalError.

Code: Select all

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 575.64.01              Driver Version: 576.88         CUDA Version: 12.9     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M.  |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 5070 Ti     On  |   00000000:01:00.0  On |                  N/A |
|  0%   55C    P3             30W /  300W |    4355MiB /  16303MiB |      1%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA GeForce RTX 3070        On  |   00000000:04:00.0 Off |                  N/A |
|  0%   41C    P8              8W /  240W |    1147MiB /   8192MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
< /code>
Ich habe 96 GB RAM.!pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

!pip3 install tensorflow[and-cuda] transformers datasets ipywidgets tf_keras huggingface_hub

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-base")
model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-base")

from datasets import load_dataset

# Load the Helsinki-NLP/opus-100 dataset
dataset = load_dataset('Helsinki-NLP/opus-100', 'en-es')
print(dataset['train'][0])
# Access column names for a specific split (e.g., 'train')
column_names = dataset["train"].column_names
print(column_names)

# Preprocess the dataset for input into the model
def preprocess_data(examples):
inputs = ['Translate from English to Spanish: {example["en"]}' for example in examples['translation']]
targets = [example['es'] for example in examples['translation']]

# Tokenize inputs
model_inputs = tokenizer(inputs, max_length=128, truncation=True, padding="max_length")

# Tokenize targets
with tokenizer.as_target_tokenizer():
labels = tokenizer(targets, max_length=128, truncation=True, padding="max_length")

model_inputs["labels"] = labels["input_ids"]

# For decoder inputs
decoder_inputs = tokenizer(targets, max_length=128, truncation=True, padding="max_length")
model_inputs["decoder_input_ids"] = decoder_inputs["input_ids"]

return model_inputs

# Apply preprocessing to the dataset
train_dataset = dataset['train'].select(range(5000)).map(preprocess_data, batched=True)
test_dataset = dataset['test'].select(range(1000)).map(preprocess_data, batched=True)

print(train_dataset[0])

# Convert Hugging Face datasets to TensorFlow datasets
train_dataset = train_dataset.to_tf_dataset(
columns=['input_ids', 'attention_mask', 'decoder_input_ids'],
label_cols=['labels'],
shuffle=True,
batch_size=16,
collate_fn=None
)
test_dataset = test_dataset.to_tf_dataset(
columns=['input_ids', 'attention_mask', 'decoder_input_ids'],
label_cols=['labels'],
shuffle=False,
batch_size=16,
collate_fn=None
)
< /code>
Der Fehler, den ich erhalte, ist: < /p>
2025-07-10 16:03:54.511675: W tensorflow/compiler/mlir/tools/kernel_gen/tf_gpu_runtime_wrappers.cc:40] 'cuModuleLoadData(&module, data)' failed with 'CUDA_ERROR_INVALID_PTX'

2025-07-10 16:03:54.511738: W tensorflow/compiler/mlir/tools/kernel_gen/tf_gpu_runtime_wrappers.cc:40] 'cuModuleGetFunction(&function, module, kernel_name)' failed with 'CUDA_ERROR_INVALID_HANDLE'

2025-07-10 16:03:54.511750: W tensorflow/core/framework/op_kernel.cc:1844] INTERNAL:  'cuLaunchKernel(function, gridX, gridY, gridZ, blockX, blockY, blockZ, 0, reinterpret_cast(stream), params, nullptr)' failed with 'CUDA_ERROR_INVALID_HANDLE'

Gibt es sowieso, um dieses Problem zu beheben?