github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4990] First value different on CUDA/ROCM when setting seed #65192

New Issue
Open
opened 2026-05-03 19:58:23 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-05-03 19:58:23 -05:00
Owner
Copy Link

Originally created by @jmorganca on GitHub (Jun 12, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4990

What is the issue?

This seems to be an issue with the kv cache on Nvidia/AMD GPUs. See https://github.com/ggerganov/llama.cpp/issues/2838

Originally created by @jmorganca on GitHub (Jun 12, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4990 ### What is the issue? This seems to be an issue with the kv cache on Nvidia/AMD GPUs. See https://github.com/ggerganov/llama.cpp/issues/2838
GiteaMirror added the bugamdnvidia labels 2026-05-03 19:58:23 -05:00
GiteaMirror commented 2026-05-03 19:58:26 -05:00
Author
Owner
Copy Link

@ScreamingHawk commented on GitHub (Jun 16, 2024):

Worth noting, setting temperature: 0 didn't seem to give consistent results either. Despite what is said in that linked ticket. Not sure if that is a related issue or something separate. See #5012 for reproducible error

<!-- gh-comment-id:2171825292 --> @ScreamingHawk commented on GitHub (Jun 16, 2024): Worth noting, setting `temperature: 0` didn't seem to give consistent results either. Despite what is said in that linked ticket. Not sure if that is a related issue or something separate. See #5012 for reproducible error
GiteaMirror commented 2026-05-03 19:58:28 -05:00
Author
Owner
Copy Link

@pulinagrawal commented on GitHub (Feb 13, 2025):

I teach a class of 40 students who are using seed with Ollama. I am fairly certain that the issue is with prompt caching. Seed stops working when they run their code multiple times. If Ollama is restarted they get the first result they got when they ran it the first time.

https://github.com/ggerganov/llama.cpp/issues/2838#issuecomment-1817601850
also references this issue with cache_prompt.

In ollama cache_prompt seems to be permanently set based on this 8cf16063a5/llm/server.go (L701)

<!-- gh-comment-id:2657680228 --> @pulinagrawal commented on GitHub (Feb 13, 2025): I teach a class of 40 students who are using seed with Ollama. I am fairly certain that the issue is with prompt caching. Seed stops working when they run their code multiple times. If Ollama is restarted they get the first result they got when they ran it the first time. https://github.com/ggerganov/llama.cpp/issues/2838#issuecomment-1817601850 also references this issue with cache_prompt. In ollama cache_prompt seems to be permanently set based on this https://github.com/ollama/ollama/blob/8cf16063a52deb416e16039c73264e26f7e9a43a/llm/server.go#L701
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 amd bug nvidia
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#65192
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?