torch 可以检测到显卡信息，但是运行时提示未检测到CUDA环境及相关错误

[szlf619]

Reminder

• I have read the README and searched the existing issues.

System Info

windows 10
Nvidia GTX 4090 Laptop
llama-factory: 0.9.2.dev0

CUDA版本12.2
torch 2.5.1+cu124
python 3.12
Transformers version: 4.46.1
bitsandbytes 0.44.1.dev0+9315692
Datasets version: 3.1.0
Accelerate version: 1.0.1

Reproduction

torch 可以检测到显卡信息，但是运行时提示未检测到CUDA环境。

python -m bitsandbytes 报以下错误。

训练使用了示例数据中 identity，但是训练时间很久，在以下最后一段代码 Number of trainable parameters = 20,971,520 后停留很久，请问正常吗？

[INFO|2024-12-23 17:41:27] llamafactory.model.model_utils.checkpointing:157 >> Gradient checkpointing enabled.
[INFO|2024-12-23 17:41:27] llamafactory.model.model_utils.attention:157 >> Using torch SDPA for faster training and inference.
[INFO|2024-12-23 17:41:27] llamafactory.model.adapter:157 >> Upcasting trainable params to float32.
[INFO|2024-12-23 17:41:27] llamafactory.model.adapter:157 >> Fine-tuning method: LoRA
[INFO|2024-12-23 17:41:27] llamafactory.model.model_utils.misc:157 >> Found linear modules: k_proj,up_proj,v_proj,o_proj,q_proj,down_proj,gate_proj
C:\Users\PowerPC\AppData\Roaming\Python\Python311\site-packages\bitsandbytes\backends\cpu_xpu_common.py:29: UserWarning: g++ not found, torch.compile disabled for CPU/XPU.
warnings.warn("g++ not found, torch.compile disabled for CPU/XPU.")
[INFO|2024-12-23 17:41:27] llamafactory.model.loader:157 >> trainable params: 20,971,520 || all params: 8,051,232,768 || trainable%: 0.2605
[INFO|trainer.py:698] 2024-12-23 17:41:27,324 >> Using cpu_amp half precision backend
[INFO|trainer.py:2313] 2024-12-23 17:41:27,488 >> ***** Running training *****
[INFO|trainer.py:2314] 2024-12-23 17:41:27,488 >> Num examples = 91
[INFO|trainer.py:2315] 2024-12-23 17:41:27,488 >> Num Epochs = 3
[INFO|trainer.py:2316] 2024-12-23 17:41:27,488 >> Instantaneous batch size per device = 2
[INFO|trainer.py:2319] 2024-12-23 17:41:27,488 >> Total train batch size (w. parallel, distributed & accumulation) = 16
[INFO|trainer.py:2320] 2024-12-23 17:41:27,488 >> Gradient Accumulation steps = 8
[INFO|trainer.py:2321] 2024-12-23 17:41:27,489 >> Total optimization steps = 15
[INFO|trainer.py:2322] 2024-12-23 17:41:27,491 >> Number of trainable parameters = 20,971,520
0%| | 0/15 [00:00<?, ?it/s]C:\Users\PowerPC\AppData\Roaming\Python\Python311\site-packages\transformers\trainer.py:3536: FutureWarning: torch.cpu.amp.autocast(args...) is deprecated. Please use torch.amp.autocast('cpu', args...) instead.
ctx_manager = torch.cpu.amp.autocast(cache_enabled=cache_enabled, dtype=self.amp_dtype)

Expected behavior

想知道问题在哪里，尝试了很多方案都未曾解决。

Others

No response