search

KimMeen / Time-LLM

[ICLR 2024] Official implementation of " šŸ¦™ Time-LLM: Time Series Forecasting by Reprogramming Large Language Models"
https://arxiv.org/abs/2310.01728
Apache License 2.0
1.02k stars 179 forks source link

About system requirements to train the time LLM model #76

Closed shreyaskorde16 closed 1 month ago

shreyaskorde16 commented 1 month ago

Hello, I noticed that timeĀ LLM is a cutting-edge technique for time series prediction, and I'd like to learn more about it. As a Master student, I wish to do my master thesis on this issue, but I am unable to execute run_main.py. I'm getting a system-kill error. Because there is less ram. So, could you please inform me how much RAM and GPU will be required to train these models? I have 32GB RAM and a 6GB GPU.

kwuking commented 1 month ago

Hello, I noticed that timeĀ LLM is a cutting-edge technique for time series prediction, and I'd like to learn more about it. As a Master student, I wish to do my master thesis on this issue, but I am unable to execute run_main.py. I'm getting a system-kill error. Because there is less ram. So, could you please inform me how much RAM and GPU will be required to train these models? I have 32GB RAM and a 6GB GPU.

Hello, currently if the base model uses LLaMA, it's best to have an 80GB A100; it's recommended to switch to GPT-2, which should be able to meet your requirements.