Jitting fails with cpu, gpu & cuda

google-research / timesfm

TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.

https://research.google/blog/a-decoder-only-foundation-model-for-time-series-forecasting/

Apache License 2.0

3.02k stars 228 forks source link

Jitting fails with cpu, gpu & cuda #51

Open Blochergroup opened 1 month ago

Blochergroup commented 1 month ago

Tried running this library: https://github.com/gabaid971/ts-forecaster

Doesn't matter If I try with my CPU, GPU (radeon 780 integrated), or cuda (RTX 4060). When i run it in a jupyter notebook it crashes during "jitting decoding" when I load from checkpoint

My parameters: tfm = timesfm.TimesFm( context_len=128, horizon_len=14, input_patch_len=32, output_patch_len=128, num_layers=20, model_dims=1280, backend="cpu", )

mariomoonetedsfsd commented 1 month ago

I also encountered this problem at first. I switched to the server. This problem will not occur when using Ubuntu system.

Blochergroup commented 1 month ago

Unfortunately not working on my Ubuntu which I run with WSL

sjc981231 commented 2 weeks ago

Unfortunately not working on my Ubuntu which I run with WSL

Hey bro. I use WSL and then succeed. Maybe you encountered the problem that the process is killed when jitting decoding. In this way, you need to open C:\Users{user_name}.wslconfig. Change the memory to >=16GB. Then open the windows PowerShell/cmd and input the command 'wsl --shutdown'. That's how I solve the question. Hope my suggestion could help you.