github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #5791] Ability to pass --predict to llama.cpp server in ollama #29368

New Issue
Open
opened 2026-04-22 08:10:15 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-22 08:10:15 -05:00
Owner
Copy Link

Originally created by @1cekrim on GitHub (Jul 19, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/5791

Due to the structure of the deepseek v2 coder model, there was a problem with gibberish when k shift occurred.
And to solve this, a patch that causes GGML_ASSERT when k shift occurs in the deepseek v2 model has been merged.

Anyway, the easiest way to solve this problem is to pass the --predict -2 option when running the llama.cpp server. This option limits the number of tokens to predict until the context is full.

It would be a good idea to set the n predict value as an environment variable when serving ollama, or set the value in the Modelfile so that it can be passed as the --predict value in NewLlamaServer.
Also, if possible, it would be good to apply it to ollama.com's Deepseek V2 models.

Originally created by @1cekrim on GitHub (Jul 19, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/5791 Due to the structure of the deepseek v2 coder model, there was [a problem with gibberish when k shift occurred.](https://github.com/ggerganov/llama.cpp/issues/8498) And to solve this, [a patch](https://github.com/ggerganov/llama.cpp/pull/8501) that causes GGML_ASSERT when k shift occurs in the deepseek v2 model has been merged. Anyway, the easiest way to solve this problem is to pass the `--predict -2` option when running the llama.cpp server. This option limits the number of tokens to predict until the context is full. It would be a good idea to set the n predict value as an environment variable when serving ollama, or set the value in the Modelfile so that it can be passed as the `--predict` value in `NewLlamaServer`. Also, if possible, it would be good to apply it to ollama.com's Deepseek V2 models. - Related issues - https://github.com/ollama/ollama/issues/5537 - https://github.com/ollama/ollama/issues/5339
GiteaMirror added the feature request label 2026-04-22 08:10:15 -05:00
GiteaMirror commented 2026-04-22 08:10:17 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 19, 2024):

I created a patch to add the ability to set predict in the model file, but it didn't help with the output of https://github.com/ollama/ollama/issues/5339. Maybe it needs a more recent version of llama.cpp.

<!-- gh-comment-id:2240027821 --> @rick-github commented on GitHub (Jul 19, 2024): I created a patch to add the ability to set `predict` in the model file, but it didn't help with the output of https://github.com/ollama/ollama/issues/5339. Maybe it needs a more recent version of llama.cpp.
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 feature request
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#29368
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?