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 #409] Custom model based on codellama just outputs blank lines #25947

New Issue
Closed
opened 2026-04-22 01:48:49 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-22 01:48:49 -05:00
Owner
Copy Link

Originally created by @tomduncalf on GitHub (Aug 25, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/409

Hi, super cool project, impressed how easy it was to get started!

I created a custom model based on codellama with a system message explaining its task, but when I run it with my input, the model just infinitely outputs blank lines.

I saw on a comment on Hacker News that this was a more general problem with codellama until you fixed it, so I wondered if you had any specific thoughts or advice on what might be causing this?

Thanks!

Originally created by @tomduncalf on GitHub (Aug 25, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/409 Hi, super cool project, impressed how easy it was to get started! I created a custom model based on codellama with a system message explaining its task, but when I run it with my input, the model just infinitely outputs blank lines. I saw on a [comment on Hacker News](https://news.ycombinator.com/item?id=37252690) that this was a more general problem with codellama until you fixed it, so I wondered if you had any specific thoughts or advice on what might be causing this? Thanks!
GiteaMirror commented 2026-04-22 01:48:52 -05:00
Author
Owner
Copy Link

@BruceMacD commented on GitHub (Aug 25, 2023):

Hey @tomduncalf, we tried a few things. The model seems to be pretty sensitive to configuration.

What seemed to help the most with stability was setting the rope frequency base. Here is what that looks like in a Modelfile:

PARAMETER rope_frequency_base 1000000

It also seemed pretty sensitive to white-space at the end of the prompt. Here's an example Modelfile you can try working from that will get our configuration by default:

FROM codellama
TEMPLATE """
[INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>>

{{ end }}{{ .Prompt }} [/INST]
"""
SYSTEM """
Provide answers in JavaScript
"""
<!-- gh-comment-id:1693795444 --> @BruceMacD commented on GitHub (Aug 25, 2023): Hey @tomduncalf, we tried a few things. The model seems to be pretty sensitive to configuration. What seemed to help the most with stability was setting the rope frequency base. Here is what that looks like in a Modelfile: ``` PARAMETER rope_frequency_base 1000000 ``` It also seemed pretty sensitive to white-space at the end of the prompt. Here's an example Modelfile you can try working from that will get our configuration by default: ``` FROM codellama TEMPLATE """ [INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>> {{ end }}{{ .Prompt }} [/INST] """ SYSTEM """ Provide answers in JavaScript """ ```
GiteaMirror commented 2026-04-22 01:48:52 -05:00
Author
Owner
Copy Link

@tomduncalf commented on GitHub (Aug 25, 2023):

Hey @BruceMacD, that seems to have helped! Thanks very much :)

Out of interest, where can I find the templates you use internally?

Thanks,
Tom

<!-- gh-comment-id:1693844825 --> @tomduncalf commented on GitHub (Aug 25, 2023): Hey @BruceMacD, that seems to have helped! Thanks very much :) Out of interest, where can I find the templates you use internally? Thanks, Tom
GiteaMirror commented 2026-04-22 01:48:53 -05:00
Author
Owner
Copy Link

@BruceMacD commented on GitHub (Aug 30, 2023):

@tomduncalf

It will be in each library's description soon at https://ollama.ai/library

The other option (which works right now) is to download the model, and output the model info during an interactive session.

$ ollama run codellama
>>> /help
commands:
  /help
  /list
  /set
  ├── history
  ├── nohistory
  ├── verbose
  ├── quiet
  ├── mode
  ├────── vim
  ├────── emacs
  ├────── default
  /show
  ├── license
  ├── system
  ├── template
  /exit
  /bye

>>> /show template
[INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>>

Hope that helps.

<!-- gh-comment-id:1699511208 --> @BruceMacD commented on GitHub (Aug 30, 2023): @tomduncalf It will be in each library's description soon at https://ollama.ai/library The other option (which works right now) is to download the model, and output the model info during an interactive session. ``` $ ollama run codellama >>> /help commands: /help /list /set ├── history ├── nohistory ├── verbose ├── quiet ├── mode ├────── vim ├────── emacs ├────── default /show ├── license ├── system ├── template /exit /bye >>> /show template [INST] {{ if and .First .System }}<<SYS>>{{ .System }}<</SYS>> ``` Hope that helps.
GiteaMirror commented 2026-04-22 01:48:53 -05:00
Author
Owner
Copy Link

@tomduncalf commented on GitHub (Aug 30, 2023):

Ooh, nice. It does help! Thanks :)

<!-- gh-comment-id:1699518489 --> @tomduncalf commented on GitHub (Aug 30, 2023): Ooh, nice. It does help! Thanks :)
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
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#25947
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?