Open
Description
Occasionally while pulling a model the download may get stuck waiting for a part which experienced an error.
$ ollama run llama2:70b
pulling manifest
pulling 153664158022... 99% |█████████████████████████████████████████████████████████████████████████████████ | (38/39 GB, 619 kB/s) [24m57s:9m15s]
Server log:
2023/11/07 11:36:24 download.go:160: 153664158022 part 75 attempt 0 failed: unexpected EOF, retrying
Workaround is to stop and resume the download.
Activity
killthekitten commentedon Nov 8, 2023
I've accumulated a handful of different connection errors, and EOF sometimes appeared among them. I have two terminals running the pull in parellel, one on my local machine, and another one on GCP, and the failures seem correlated.
One moment both terminals go unresponsive or throw an error, while after some time and several retries the download speed reaches hundreds of mb/s. Could it be related to this issue somehow #850?
cwatt commentedon Dec 6, 2023
I also seem to be experiencing this or a related problem while tring to pull models. Unfortunately, I haven't been able to succesfully pull any models since installing Ollama (0.1.13). Here's an example:
jmorganca commentedon Dec 24, 2023
@cwatt so sorry you hit this error – wondering if this is still something you're hitting on every pull? Thanks for sharing will make sure to take a look at this
cwatt commentedon Jan 2, 2024
@jmorganca On subsequent pull attempts I actually haven't been hitting any more EOF errors, but rather digest mismatch errors like what is described in this issue.
I'm not sure what caused the change in behavior. I hope this helps!
sammcj commentedon May 19, 2024
FYI the model registry seems to have been broken for 3-4 days.
(#1736 (comment))
sammcj commentedon May 24, 2024
This PR fixes the issue for me: #4619
rampageservices commentedon Aug 2, 2024
There is a follow-on PR to #4619 that is yet to be pulled that seems to resolve the remaining issues that people are facing in this issue thread.
#4625
6 remaining items