gpt4all icon indicating copy to clipboard operation
gpt4all copied to clipboard

Illegal instruction (core dumped)

Open HappyPony opened this issue 1 year ago • 4 comments

Hi,

I have x86_64 CPU with Ubuntu 22.04.2 LTS, downloaded GPT4All and get this message. `USERNAME@PCNAME:/$ "/opt/gpt4all 0.1.0/bin/chat" QML debugging is enabled. Only use this in a safe environment. gptj_model_load: loading model from 'ggml-gpt4all-j.bin' - please wait ... gptj_model_load: n_vocab = 50400 gptj_model_load: n_ctx = 2048 gptj_model_load: n_embd = 4096 gptj_model_load: n_head = 16 gptj_model_load: n_layer = 28 gptj_model_load: n_rot = 64 gptj_model_load: f16 = 2 Illegal instruction (core dumped)

USERNAME@PCNAME:/$`

Where can I look up what hardware requirements are necessary to run GPT4All?

HappyPony avatar Apr 17 '23 12:04 HappyPony

You're fine. I encountered this exact same issue. The AVX instruction set is not present on budget processors. You'll need to recompile the chat portion locally so that it can determine what instructions are available and utilize those instead. Look for my issue regarding core dumps which I think is still active.

One word of caution. Without AVX instructions on your CPU it will be dog slow. I'm about 1 min per token. For reference, I have a Quad Core processor intel processor from around 2017 with 8GB of ram and everything is pegged while running it but it does run.

Just very, very, slowly.

devlux76 avatar Apr 17 '23 17:04 devlux76

Thanks for the help. I have no experience in compilation.

Look for my issue regarding core dumps which I think is still active.

Have searched here https://github.com/nomic-ai/gpt4all/ and found nothing about it. Would be nice if you could post here the link to chat portion and a ready command line for compilation. Then I will post in Discord this solution. I have seen other requests in Discord, in which exactly this error message is quoted.

I have checked - I have an Intel(R) Pentium(R) CPU G4400T @ 2.90GHz Dualcore with 8 GB RAM. I am curious if GPT4All runs at all with this hardware ;-).

HappyPony avatar Apr 18 '23 04:04 HappyPony

It's sort of an exercise to sort these issues. The process in general is to fork the repo, do a git pull to your computer and run "make" from there you'll have a new binary and you can either run it from there, or move it to your /opt/gpt4all/whatever where the installer put the original.

To answer your other question...

Does the Intel(R) Pentium(R) CPU G4400T @ 2.90GHz support either AVX or AVX2?

Finished browsing Searched: “Intel Pentium CPU G4400T AVX AVX2 support” Clicked on: Intel® Pentium® Processor G4400T Reading content Finished browsing The Intel® Pentium® CPU G4400T @ 2.90GHz supports the instruction set extensions Intel® SSE4.1 and Intel® SSE4.2, but it does not support AVX or AVX2​1 Favicon Intel® Pentium® Processor G4400T

devlux76 avatar Apr 23 '23 09:04 devlux76

OK, I ran Intel processor identification utility with this result - see screenshot:

Esprimo-Q556-CPU

HappyPony avatar Apr 23 '23 10:04 HappyPony

can i download precompiled binary for processor without avx instructions?

gmankab avatar May 16 '23 04:05 gmankab

Stale, please open a new, updated issue if this is still relevant to you.

niansa avatar Aug 11 '23 11:08 niansa

can i download precompiled binary for processor without avx instructions?

No, it'd be unbearably slow.

niansa avatar Aug 11 '23 11:08 niansa