Does not work on Mac? Causing System Crashes building and running

[kuro337]

Is Ollama not meant to be run on ARM macs?

I followed these steps

git clone [email protected]:ollama/ollama.git
cd ollama
go generate ./...
go build .

./ollama

# First time running
[1]    1651 killed     ./ollama

# After running again
 ./ollama

# hangs indefinitely

Then it hands indefinitely - I am not able to Terminate it and even using kill does not work

./ollama   

^C^C^C^C

# or any combination of cancels/sigterms

Deleting it for now, will try to run on my Ubuntu with some clarification

Is this the way to run and serve a Model over HTTP?

# steps to run the REST API?

./ollama serve

./ollama run mixtral:8x7b-instruct-v0.1-q5_1

curl http://localhost:11434/api/generate -d '{
  "model": "mixtral",
  "messages": [
    { "role": "system", "content": "Explain using Async in Scala?" }
  ]
}'

Thank you , would appreciate any pointers

I have the latest version of Go , running on a Macbook with 128gb memory

[kuro337]

Also for reference I have llama.cpp and it works fine for running .gguf models - so doesn't seem to be an issue related to system deps

[dhiltgen]

Is it possible you're running under Rosetta?

% sysctl -n sysctl.proc_translated

If that says "1" you're emulating x86, not running on native ARM.

[kuro337]

Running on Native ARM

sysctl -n sysctl.proc_translated
0

I ran this natively not in a container so should be ARM, so the steps I followed were fine?

I can try again

[dhiltgen]

In that case, perhaps some build dependency isn't satisfied. Have you follow the developer guide instructions for installing the required minimum tools? https://github.com/ollama/ollama/blob/main/docs/development.md#development

If those are satisfied, and the compiled binary is still crashing, maybe there's some AV monitor on your system that is triggering? All the maintainers use ARM macs, and I've never seen this failure mode.

[hoyyeva]

Hi @kuro337 are you still experiencing the issue? Let us know if there is anything else that we can help you to resolve the issue.