[GH-ISSUE #8971] download bug? #52334

Open
opened 2026-04-28 23:03:54 -05:00 by GiteaMirror · 5 comments
Owner

Originally created by @brinkqiang on GitHub (Feb 9, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8971

When running ollama run to download a model, the speed is initially fast, but after a while, it drops significantly. When you press ctrl+c to close and reopen it, the speed is fast again, but after some time, it slows down again.

env:
win10
ollama version is 0.5.7

Originally created by @brinkqiang on GitHub (Feb 9, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8971 When running ollama run to download a model, the speed is initially fast, but after a while, it drops significantly. When you press ctrl+c to close and reopen it, the speed is fast again, but after some time, it slows down again. env: win10 ollama version is 0.5.7
GiteaMirror added the feature request label 2026-04-28 23:03:54 -05:00
Author
Owner

@bmizerany commented on GitHub (Mar 4, 2025):

I'm actively working on this. If you have some time, do you mind trying my new branch @ #9485 and letting me know if the problem still persists?

<!-- gh-comment-id:2696473672 --> @bmizerany commented on GitHub (Mar 4, 2025): I'm actively working on this. If you have some time, do you mind trying my new branch @ #9485 and letting me know if the problem still persists?
Author
Owner

@veelasama commented on GitHub (Mar 13, 2025):

The same issue: the pull command downloads at maximum speed, then at around 90%, the speed drops to 100-500 kilobytes per second. Changing DNS servers, IP, or rebooting pc/router doesn't help.

<!-- gh-comment-id:2721325597 --> @veelasama commented on GitHub (Mar 13, 2025): The same issue: the pull command downloads at maximum speed, then at around 90%, the speed drops to 100-500 kilobytes per second. Changing DNS servers, IP, or rebooting pc/router doesn't help.
Author
Owner

@xinggangling commented on GitHub (May 3, 2025):

The same issue: the pull command downloads at maximum speed, then at around 90%, the speed drops to 100-500 kilobytes per second. Changing DNS servers, IP, or rebooting pc/router doesn't help.

same

<!-- gh-comment-id:2848619633 --> @xinggangling commented on GitHub (May 3, 2025): > The same issue: the pull command downloads at maximum speed, then at around 90%, the speed drops to 100-500 kilobytes per second. Changing DNS servers, IP, or rebooting pc/router doesn't help. same
Author
Owner

@SpydirRoot commented on GitHub (Jul 13, 2025):

Any progress on this? Issue persists and is replicated with every pull I do. Same issue as above. Lightning fast until ~90%, then it's time to go make dinner and have a shower.

Using Windows. Ruled out:
Windows 11 vs Windows 10
Local vs Docker Image

PS. Sure, it sucks waiting for that last 30 minutes/10%, but damn Ollama is fantastic. Thanks all =)

<!-- gh-comment-id:3066747229 --> @SpydirRoot commented on GitHub (Jul 13, 2025): Any progress on this? Issue persists and is replicated with every pull I do. Same issue as above. Lightning fast until ~90%, then it's time to go make dinner and have a shower. Using Windows. Ruled out: Windows 11 vs Windows 10 Local vs Docker Image PS. Sure, it sucks waiting for that last 30 minutes/10%, but damn Ollama is fantastic. Thanks all =)
Author
Owner

@perfectecologietool commented on GitHub (Jul 13, 2025):

https://github.com/ollama/ollama/issues/1736#issuecomment-1919592282 we are super lucky to have a free server anyway. Cant even get an ip number like back in the ol days, can only imagine setting up a public mirror for models. Should probably look at a peer to peer like bitTorrent or InterPlanetaryFileSystem for models. The peices are there already, hashfilenames, tcp port server. And there's only a handful of models, just theyre super big.

<!-- gh-comment-id:3067037404 --> @perfectecologietool commented on GitHub (Jul 13, 2025): https://github.com/ollama/ollama/issues/1736#issuecomment-1919592282 we are super lucky to have a free server anyway. Cant even get an ip number like back in the ol days, can only imagine setting up a public mirror for models. Should probably look at a peer to peer like bitTorrent or InterPlanetaryFileSystem for models. The peices are there already, hashfilenames, tcp port server. And there's only a handful of models, just theyre super big.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#52334