Bug on Ubuntu 22: software doesnt work

[lelapin123]

Installation smooth, no problem

So i do a python ingest.py and everything is fine, but then later:

load INSTRUCTOR_Transformer max_seq_length 512 Using embedded DuckDB with persistence: data will be stored in: /mnt/6903a017-f604-4f90-9652-324e10b3e675/work/h2oai/localgpt/localGPT Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s] Killed

(note: i am not doing that in my home folder)

[TitleOS]

Having the exact same issue with Ubuntu 22 LTS. 32GB of ram, Nvidia Tesla m40 20gb.

[akitkumar24]

I have a 32GB system running Ubuntu 20.04 on which am running the "python3 run_localGPT.py". After the execution, the RAM went to 0 in some sec and then I got the same issue. I am uncertain how much memory is needed to run the model.

[mindwellsolutions]

I am getting the same error on Ubuntu 22.04.2 .

load INSTRUCTOR_Transformer max_seq_length 512 Using embedded DuckDB with persistence: data will be stored in: /home/ailocal/localGPT/DB Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s] Killed

[OGoodness]

+1

[duramaxlb7]

I am having the same issue. Just installed on Ubuntu 22.04.

16GB of Memory; 25GB of Swap. the process fails when Swap reaches 100% utilization.

Running on: cuda load INSTRUCTOR_Transformer max_seq_length 512 Using embedded DuckDB with persistence: data will be stored in: /home/MYNAME/snap/localgpt/localGPT/DB Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]Killed

[jocastrocUnal]

same here

[TitleOS]

While not an ideal solution, with having 32gb of physical ram, I expanded my swap file to be an additional 32gb, for a total of 64gb of combined physical and digital ram. With that, I am able to run the program, enter queries, and receive expected answers. It should be noted that this solution is incredibly slow, even with a PCI4 nvme ssd being used for the swap file.