[GH-ISSUE #13034] Partial download handling #34396

Open
opened 2026-04-22 17:55:04 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @nealrauhauser on GitHub (Nov 10, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13034

The handling of partially complete downloads leaves a LOT to be desired.

I'm using Ollama from the command line, how do I see what partly finished downloads I have?

I've partly completed a download and decide against finishing, how do I remove the partial?

I learned that a server restart will clean up the partials, but I had to Google, when it really should be a clearly visible option in the interface.

Originally created by @nealrauhauser on GitHub (Nov 10, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13034 The handling of partially complete downloads leaves a **LOT** to be desired. I'm using Ollama from the command line, how do I see what partly finished downloads I have? I've partly completed a download and decide against finishing, how do I remove the partial? I learned that a server restart will clean up the partials, but I had to Google, when it really should be a clearly visible option in the interface.
GiteaMirror added the feature request label 2026-04-22 17:55:04 -05:00
Author
Owner

@webdev23 commented on GitHub (Dec 26, 2025):

Similarly, the downloader is able to restart in some conditions, but in many cases it does simply start again from scratch.

Devs should not try to reinvent the wheel, use robust libraries such as wget or curl, a computer that is switched off MUST be able to continue an aborted download from its cache, transparently, and users MUST be able to throttle a download easily.

We are talking about 10's of GBs here, many users doesn't have cloud level internet connections speed.
The current download mechanism is killing the planet and your servers with terabytes of unnecessary downloads.

Examples:
ollama run deepseek-r1:671b (404GB, require 1 week at 1MB/S)
CTRL+C
ollama run deepseek-r1:671b (again, a bit later)
=> the download continues [OK]

ollama run deepseek-r1:671b (404GB)
CTRL+C
... reboot
ollama run deepseek-r1:671b (404GB)
Starts again from zero [NOT OK]

<!-- gh-comment-id:3692243225 --> @webdev23 commented on GitHub (Dec 26, 2025): Similarly, the downloader is able to restart in some conditions, but in many cases it does simply start again from scratch. **Devs should not try to reinvent the wheel**, use robust libraries such as wget or curl, a computer that is switched off **MUST** be able to continue an aborted download from its cache, transparently, and users **MUST** be able to throttle a download easily. We are talking about 10's of GBs here, many users doesn't have cloud level internet connections speed. The current download mechanism is killing the planet and your servers with terabytes of unnecessary downloads. **Examples:** `ollama run deepseek-r1:671b` (404GB, require 1 week at 1MB/S) CTRL+C `ollama run deepseek-r1:671b` (again, a bit later) => the download continues [OK] `ollama run deepseek-r1:671b` (404GB) CTRL+C ... reboot `ollama run deepseek-r1:671b` (404GB) Starts again from zero [**NOT OK**]
Author
Owner

@rick-github commented on GitHub (Jan 14, 2026):

OLLAMA_NOPRUNE

<!-- gh-comment-id:3748645841 --> @rick-github commented on GitHub (Jan 14, 2026): [`OLLAMA_NOPRUNE`](https://github.com/ollama/ollama/blob/875cecba74a3722cc8d490d3c65682f8ed82d122/envconfig/config.go#L288)
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#34396