github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 17:49:24 -05:00
Code Issues 14.6k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #6436] Reloading the Same Model During Consecutive Accesses to the ollama API Within OLLAMA_KEEP_ALIVE Duration #29806

New Issue
Closed
opened 2026-04-22 09:03:52 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-22 09:03:52 -05:00
Owner
Copy Link

Originally created by @DavidLetGo on GitHub (Aug 20, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6436

What is the issue?

Issue Summary:

Reloading the same model unnecessarily costs approximately 10 seconds per reload, significantly degrading performance.

Issue Found in Ollama Version 0.3.4:

During debugging, I added debug logging (slog.Debug) to the source code, built, and debugged the application. This led to some useful findings regarding the issue.

Possible Root Cause:

It appears that the normalization process is not functioning correctly. After a previous access to the ollama API, the runner.numParallel value unexpectedly resets to 1, despite the OLLAMA_NUM_PARALLEL environment variable being set to 3.
// Normalize the NumCtx for parallelism
optsExisting.NumCtx = optsExisting.NumCtx / runner.numParallel

Part of log:

  • time=2024-08-20T01:55:48.411-04:00 level=DEBUG source=sched.go:593 msg="Before Normalization" optsExisting.NumCtx=24576 optsExisting.NumCtx=24576 runner.numParallel=1
  • time=2024-08-20T01:55:48.411-04:00 level=DEBUG source=sched.go:596 msg="After Normalization" optsExisting.NumCtx=24576 optsExisting.NumCtx=24576 runner.numParallel=1
    DEBUG [process_single_task] slot data | n_idle_slots=1 n_processing_slots=0 task_id=313 tid="140124911321088" timestamp=1724133348
  • time=2024-08-20T01:55:48.412-04:00 level=DEBUG source=sched.go:616 msg="Debugging paths and options" runner.model.AdapterPaths=[] req.model.AdapterPaths=[] runner.model.ProjectorPaths=[] req.model.ProjectorPaths=[] optsExisting="{NumCtx:24576 NumBatch:512 NumGPU:-1 MainGPU:0 LowVRAM:false F16KV:true LogitsAll:false VocabOnly:false UseMMap: UseMLock:false NumThread:0}" optsNew="{NumCtx:8192 NumBatch:512 NumGPU:-1 MainGPU:0 LowVRAM:false F16KV:true LogitsAll:false VocabOnly:false UseMMap: UseMLock:false NumThread:0}"

The complete log:
debug-ollama-locally.log

Accesses to Ollama API:
$ curl http://localhost:11434/api/generate -d '{
"model": "mixtral-8x7b-instruct",
"prompt": "how is the novel, the wind in the willow?",
"stream": false,
"options": {
"num_ctx": 8192
}
}'
$ curl http://localhost:11434/api/generate -d '{
"model": "mixtral-8x7b-instruct",
"prompt": "How is milk tea?",
"stream": false,
"options": {
"num_ctx": 8192
}
}'

The environment variables used:
"OLLAMA_DEBUG": "true",
"OLLAMA_NUM_PARALLEL": "3"
default "OLLAMA_KEEP_ALIVE "

Devices
Only one GPU: GeForce RTX 4090

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.3.4

Originally created by @DavidLetGo on GitHub (Aug 20, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6436 ### What is the issue? **Issue Summary:** Reloading the same model unnecessarily costs approximately 10 seconds per reload, significantly degrading performance. **Issue Found in Ollama Version 0.3.4:** During debugging, I added debug logging (slog.Debug) to the source code, built, and debugged the application. This led to some useful findings regarding the issue. **Possible Root Cause:** It appears that the normalization process is not functioning correctly. After a previous access to the ollama API, the runner.numParallel value unexpectedly resets to 1, despite the OLLAMA_NUM_PARALLEL environment variable being set to 3. // Normalize the NumCtx for parallelism optsExisting.NumCtx = optsExisting.NumCtx / runner.numParallel **Part of log:** - time=2024-08-20T01:55:48.411-04:00 level=DEBUG source=sched.go:593 msg="Before Normalization" optsExisting.NumCtx=24576 optsExisting.NumCtx=24576 runner.numParallel=1 - time=2024-08-20T01:55:48.411-04:00 level=DEBUG source=sched.go:596 msg="After Normalization" optsExisting.NumCtx=24576 optsExisting.NumCtx=24576 runner.numParallel=1 DEBUG [process_single_task] slot data | n_idle_slots=1 n_processing_slots=0 task_id=313 tid="140124911321088" timestamp=1724133348 - time=2024-08-20T01:55:48.412-04:00 level=DEBUG source=sched.go:616 msg="Debugging paths and options" runner.model.AdapterPaths=[] req.model.AdapterPaths=[] runner.model.ProjectorPaths=[] req.model.ProjectorPaths=[] optsExisting="{NumCtx:24576 NumBatch:512 NumGPU:-1 MainGPU:0 LowVRAM:false F16KV:true LogitsAll:false VocabOnly:false UseMMap:<nil> UseMLock:false NumThread:0}" optsNew="{NumCtx:8192 NumBatch:512 NumGPU:-1 MainGPU:0 LowVRAM:false F16KV:true LogitsAll:false VocabOnly:false UseMMap:<nil> UseMLock:false NumThread:0}" **The complete log:** [debug-ollama-locally.log](https://github.com/user-attachments/files/16670180/debug-ollama-locally.log) **Accesses to Ollama API:** $ curl http://localhost:11434/api/generate -d '{ "model": "mixtral-8x7b-instruct", "prompt": "how is the novel, the wind in the willow?", "stream": false, "options": { "num_ctx": 8192 } }' $ curl http://localhost:11434/api/generate -d '{ "model": "mixtral-8x7b-instruct", "prompt": "How is milk tea?", "stream": false, "options": { "num_ctx": 8192 } }' **The environment variables used:** "OLLAMA_DEBUG": "true", "OLLAMA_NUM_PARALLEL": "3" default "OLLAMA_KEEP_ALIVE " **Devices** Only one GPU: GeForce RTX 4090 ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.3.4
GiteaMirror added the bug label 2026-04-22 09:03:52 -05:00
GiteaMirror commented 2026-04-22 09:03:54 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Aug 20, 2024):

Fixed by https://github.com/ollama/ollama/pull/6402, next release will have this.

<!-- gh-comment-id:2298160194 --> @rick-github commented on GitHub (Aug 20, 2024): Fixed by https://github.com/ollama/ollama/pull/6402, next release will have this.
GiteaMirror commented 2026-04-22 09:03:55 -05:00
Author
Owner
Copy Link

@DavidLetGo commented on GitHub (Aug 20, 2024):

Fixed by #6402, next release will have this.

Thank you.

<!-- gh-comment-id:2298333708 --> @DavidLetGo commented on GitHub (Aug 20, 2024): > Fixed by #6402, next release will have this. Thank you.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#29806
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?