github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4797] Stop token behavior changes when specifying list of stop tokens #28786

New Issue
Closed
opened 2026-04-22 07:19:09 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-04-22 07:19:09 -05:00
Owner
Copy Link

Originally created by @ccreutzi on GitHub (Jun 3, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4797

Originally assigned to: @jmorganca on GitHub.

What is the issue?

Setting a stop token of "k=1 was" for this call has no effect, as expected:

$ curl -s http://localhost:11434/api/chat -d '{
>               "model": "mistral","options": {"top_k":1,"stop":["k=1 was"]},
>               "stream": false,
>               "messages":[{"role":"user","content":"Top-k sampling with k=1 returns a definite answer."}]
>             }' | jq .message.content
" Yes, that's correct! Top-k sampling with k=1 is also known as \"taking the top element\" or \"greedy selection.\" It simply selects the highest value (or any other sortable attribute) from a dataset without replacement. Since it only returns one item, it guarantees a definite answer. This method is often used in various machine learning and data analysis applications where you want to focus on the most significant or representative element of your dataset."

Setting a top token of "1" also has the expected effect, of ending after k= in the (constant) output:

$ curl -s http://localhost:11434/api/chat -d '{
>               "model": "mistral","options": {"top_k":1,"stop":["1"]},
>               "stream": false,
>               "messages":[{"role":"user","content":"Top-k sampling with k=1 returns a definite answer."}]
>             }' | jq .message.content
" Yes, that's correct! Top-k sampling with k="

But combining the two results in the generated text ending before the k= tokens:

$ curl -s http://localhost:11434/api/chat -d '{
>               "model": "mistral","options": {"top_k":1,"stop":["k=1 was","1"]},
>               "stream": false,
>               "messages":[{"role":"user","content":"Top-k sampling with k=1 returns a definite answer."}]
>             }' | jq .message.content
" Yes, that's correct! Top-k sampling with"

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.1.38, installed with Homebrew

Originally created by @ccreutzi on GitHub (Jun 3, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4797 Originally assigned to: @jmorganca on GitHub. ### What is the issue? Setting a stop token of `"k=1 was"` for this call has no effect, as expected: ``` $ curl -s http://localhost:11434/api/chat -d '{ > "model": "mistral","options": {"top_k":1,"stop":["k=1 was"]}, > "stream": false, > "messages":[{"role":"user","content":"Top-k sampling with k=1 returns a definite answer."}] > }' | jq .message.content " Yes, that's correct! Top-k sampling with k=1 is also known as \"taking the top element\" or \"greedy selection.\" It simply selects the highest value (or any other sortable attribute) from a dataset without replacement. Since it only returns one item, it guarantees a definite answer. This method is often used in various machine learning and data analysis applications where you want to focus on the most significant or representative element of your dataset." ``` Setting a top token of `"1"` also has the expected effect, of ending after `k=` in the (constant) output: ``` $ curl -s http://localhost:11434/api/chat -d '{ > "model": "mistral","options": {"top_k":1,"stop":["1"]}, > "stream": false, > "messages":[{"role":"user","content":"Top-k sampling with k=1 returns a definite answer."}] > }' | jq .message.content " Yes, that's correct! Top-k sampling with k=" ``` But combining the two results in the generated text ending before the `k=` tokens: ``` $ curl -s http://localhost:11434/api/chat -d '{ > "model": "mistral","options": {"top_k":1,"stop":["k=1 was","1"]}, > "stream": false, > "messages":[{"role":"user","content":"Top-k sampling with k=1 returns a definite answer."}] > }' | jq .message.content " Yes, that's correct! Top-k sampling with" ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.1.38, installed with Homebrew
GiteaMirror added the bugapi labels 2026-04-22 07:19:09 -05:00
GiteaMirror commented 2026-04-22 07:19:13 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Nov 12, 2024):

Hi there, this should be fixed now. Thanks for the issue!

<!-- gh-comment-id:2469414298 --> @jmorganca commented on GitHub (Nov 12, 2024): Hi there, this should be fixed now. Thanks for the issue!
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 api 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#28786
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?