github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-16 13:19:40 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #7256] Last character being truncated by stop sequence #51123

New Issue
Closed
opened 2026-04-28 18:29:28 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-28 18:29:28 -05:00
Owner
Copy Link

Originally created by @someone13574 on GitHub (Oct 18, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7256

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

When running inference in raw mode with '\n\n' as a stop sequence, it seems like punctuation is being removed with the stop sequence. I assume this is because of a bug with how partial stop sequences are handled.

OS

Linux

GPU

Other

CPU

AMD

Ollama version

0.3.11

Originally created by @someone13574 on GitHub (Oct 18, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7256 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? When running inference in raw mode with '\n\n' as a stop sequence, it seems like punctuation is being removed with the stop sequence. I assume this is because of a bug with how partial stop sequences are handled. ### OS Linux ### GPU Other ### CPU AMD ### Ollama version 0.3.11
GiteaMirror added the bugneeds more info labels 2026-04-28 18:29:29 -05:00
GiteaMirror commented 2026-04-28 18:29:34 -05:00
Author
Owner
Copy Link

@someone13574 commented on GitHub (Oct 18, 2024):

Found a PR in llama.cpp fixing the issue. https://github.com/ggerganov/llama.cpp/pull/9867

<!-- gh-comment-id:2423207565 --> @someone13574 commented on GitHub (Oct 18, 2024): Found a PR in llama.cpp fixing the issue. https://github.com/ggerganov/llama.cpp/pull/9867
GiteaMirror commented 2026-04-28 18:29:35 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Oct 23, 2024):

@someone13574 we're just about to remove server.cpp and switch to the Go based runner, which has different parsing logic. Could you try out the 0.4.0 RC and see if it exhibits the same problem?

https://github.com/ollama/ollama/releases

<!-- gh-comment-id:2432729939 --> @dhiltgen commented on GitHub (Oct 23, 2024): @someone13574 we're just about to [remove server.cpp](https://github.com/ollama/ollama/pull/7157) and switch to the [Go based runner](https://github.com/ollama/ollama/blob/main/docs/development.md#transition-to-go-runner), which has different parsing logic. Could you try out the 0.4.0 RC and see if it exhibits the same problem? https://github.com/ollama/ollama/releases
GiteaMirror commented 2026-04-28 18:29:36 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Oct 3, 2025):

I think this issue is pretty old, but I did test it out just to make certain it was working. I used ChatGPT as the stop parameter just to be obvious. Here's the full test using raw mode:

% curl localhost:11434/api/generate -d '{"raw": true, "model": "gpt-oss:20b", "prompt": "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cutoff: 2024-06\nCurrent date: 2025-10-03\n\nReasoning: medium\n\n# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>user<|message|>Hello friend!<|end|><|start|>assistant<|channel|>final<|message|>Hello human!<|end|><|start|>user<|message|>What is your name?<|end|><|start|>assistant", "options": {"stop": ["ChatGPT"]}, "stream": false}'
{"model":"gpt-oss:20b","created_at":"2025-10-03T22:16:20.34754Z","response":"\u003c|channel|\u003eanalysis\u003c|message|\u003eWe need to answer politely.\u003c|end|\u003e\u003c|start|\u003eassistant\u003c|channel|\u003efinal\u003c|message|\u003eI’m ","done":true,"done_reason":"stop","total_duration":612033125,"load_duration":91495417,"prompt_eval_count":88,"prompt_eval_duration":266511417,"eval_count":19,"eval_duration":253525791}
<!-- gh-comment-id:3367379250 --> @pdevine commented on GitHub (Oct 3, 2025): I think this issue is pretty old, but I did test it out just to make certain it was working. I used `ChatGPT` as the stop parameter just to be obvious. Here's the full test using raw mode: ``` % curl localhost:11434/api/generate -d '{"raw": true, "model": "gpt-oss:20b", "prompt": "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cutoff: 2024-06\nCurrent date: 2025-10-03\n\nReasoning: medium\n\n# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>user<|message|>Hello friend!<|end|><|start|>assistant<|channel|>final<|message|>Hello human!<|end|><|start|>user<|message|>What is your name?<|end|><|start|>assistant", "options": {"stop": ["ChatGPT"]}, "stream": false}' {"model":"gpt-oss:20b","created_at":"2025-10-03T22:16:20.34754Z","response":"\u003c|channel|\u003eanalysis\u003c|message|\u003eWe need to answer politely.\u003c|end|\u003e\u003c|start|\u003eassistant\u003c|channel|\u003efinal\u003c|message|\u003eI’m ","done":true,"done_reason":"stop","total_duration":612033125,"load_duration":91495417,"prompt_eval_count":88,"prompt_eval_duration":266511417,"eval_count":19,"eval_duration":253525791} ```
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 needs more info
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#51123
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?