jan
jan copied to clipboard
janhq
bug: macOS Sonoma 4.1.2 Intel cannot download models
Describe the bug When trying to download any given model on the models page from Jan, the download starts but hangs at 0% and after a certain period of time just stops. No error is given.
Steps to reproduce Steps to reproduce the behavior:
- Go to Hub
- Click on any "Download" button from any given model
- Watch the download-bar come up and disappear again.
- No error is presented, the progress-bar just disappears.
Expected behavior Expect the download to start, or present an error with what the issue might be, that it doesn't start.
Screenshots
Environment details
- Operating System: MacOS Sonoma 14.2.1
- Jan Version: 0.4.3 stable
- Processor: Intel 12-Core Xeon W 3.3Ghz
- RAM: 96GB
- Any additional relevant hardware specifics: AMD Radeon Pro W5700X 16GB, 1TB SSD.
Additional context Also tried downloading with Jan having full disk-access in the macOS settings (which I think it shouldn't need), but that didn't change described the behaviour.
I just downloaded the Apple Silicon version of Jan 0.4.3 Stable on my M1 Max MacBook Pro. The issue is not present there; it runs on macOS Sonoma 14.0. Furthermore it runs out of the box as expected on AS, so without full disk access I can download and run any given model. Hence, the issue seems to manifest on the Intel version for macOS.
@vlbosch Are you getting GPU acceleration with the AMD eGPU? If so, how? I'm also running Mac Intel build.
I have 64gb RAM and Jan is only running CPU inference with token speeds ~0.5/s on hub-downloaded Mistral 7b. Really desperate to speed things up.
@vlbosch can you reproduce this on the nightly build? https://jan.ai/install/nightly/
This issue is not happening on my Mac M2, and since our OfficeLab is lacking Mac Intel, I'm assigning this to @namchuai who is using Mac Intel to help reproduce this. Thank you.
@vlbosch thank you for reporting the problem. Unfortunately, I'm not facing that issue. I can download normally.
@vlbosch could you please try with the nightly build which @0xSage specified above? And if you still got that error, please help us to:
- Ensure that you have privilege to write to ~/jan/models
- Ensure that your network are good to go (not under any proxy, etc.)
- Upload the log file at
/Users/$(whoami)/jan/logs
Thank you!
@vlbosch can you reproduce this on the nightly build? https://jan.ai/install/nightly/
I tried it with the nightly. The issue persists, but now I am presented with the error "TIMEDOUT" with something that seems like an IPv6-address.
@namchuai I tried it with both full disk-access and without, but that doesn't make a difference. I will retrieve the log for you and upload it here.
@vlbosch Are you getting GPU acceleration with the AMD eGPU? If so, how? I'm also running Mac Intel build.
I have 64gb RAM and Jan is only running CPU inference with token speeds ~0.5/s on hub-downloaded Mistral 7b. Really desperate to speed things up.
I don't have an eGPU, on the Mac Pro it is directly via PCIe. Unfortunately Jan doesn't seem to accelerate with GPU support in Intel Macs. If it should and I just need to configure something, please let me know so I can try.
sorry for the delay. I don't see anything suspicious in log. I think we will need to do some work:
- From Jan's side: incoporate more log in the next release to troubleshoot the issue.
- Help from @vlbosch : could you re-validate below:
- You have permission to write to
~/Users/$(whoami)/jan - You have enough space on
/Users/$(whoami)by checking Disk Utility - Your internet are ready, you are not behind a proxy
Thank you for your help!
Closing this given we are a few release past this. Please open up a new issue if this happens in any later releases, and pls include logs (which are hopefully more verbose now :) )