Open CoCoNuTeK opened 3 weeks ago
SO after i run this
# Preload the pretrained TimesFM model tfm = timesfm.TimesFm( context_len=SEQ_LEN, horizon_len=PRED_LEN, input_patch_len=32, output_patch_len=128, num_layers=20, model_dims=1280, backend="gpu" if torch.cuda.is_available() else "cpu", )
I get The NVIDIA driver's CUDA version is 12.2 which is older than the ptxas CUDA version (12.5.40). Because the driver is older than the ptxas version, XLA is disabling parallel compilation, which may slow down compilation. You should update your NVIDIA driver or use the NVIDIA-provided CUDA forward compatibility packages.
meaning that the jax which is compatible with 12.2 needs to be 0.4.23 however the paxml and praxis need it to be 0.4.26
How can i solve this so the inference or the model overall runs faster??
Try downgrading paxml and praxis to version 1.2.0 to see if it is available
SO after i run this
I get The NVIDIA driver's CUDA version is 12.2 which is older than the ptxas CUDA version (12.5.40). Because the driver is older than the ptxas version, XLA is disabling parallel compilation, which may slow down compilation. You should update your NVIDIA driver or use the NVIDIA-provided CUDA forward compatibility packages.
meaning that the jax which is compatible with 12.2 needs to be 0.4.23 however the paxml and praxis need it to be 0.4.26
How can i solve this so the inference or the model overall runs faster??