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

[PR #14218] tokenizer: merge pre-tok REs to prevent multiple matches #40448

New Issue
Open
opened 2026-04-23 01:20:31 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-23 01:20:31 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/14218
Author: @rick-github
Created: 2/12/2026
Status: 🔄 Open

Base: main ← Head: bpe

📝 Commits (1)

  • 5c927a7 bpe: merge pre-tok REs to prevent multple matches

📊 Changes

1 file changed (+2 additions, -0 deletions)

View changed files

📝 tokenizer/bytepairencoding.go (+2 -0)

📄 Description

NewBytePairEncoding() takes a list of REs for pre-tokenization. If the REs are not mutually exclusive, this causes token duplication. Models affected:

deepseek-v3.1:671b
deepseek-r1:671b
cogito-2.1:671b
r1-1776:671b
deepseek-ocr

If ported to the ollama engine, the following models would also be affected:

granite-code
granite3-dense
granite3-moe
granite3-guardian
granite3.1-dense
granite3.1-moe
granite-3.2
granite3.3

Test:

$ ollama run deepseek-v3.1:671b '"1 2 3 4" repeat the previous string'
Thinking...
First, the user said: "1" then "1 2" then "1 "1 2 3" but that seems messy. Let me read carefully.

The user said: "1" then "1 2" then "1 "1 2 3" but that doesn't make sense. Perhaps it's a pattern.

Perhaps it's a sequence of strings. Let me see the user's message: "1" then "1 2" then "1 "1 2 3"
but that seems off. Maybe it's "1" then "1 2" then "1 2 3" but the user wrote "1 "1 2 3" which might
be a typo.
...

This PR merges the list into a single RE with alternation. An alternative approach would be to modify the call to NewBytePairEncoding() in the model.go file of affected models to pass a pre-concatenated string, or modify the parsing in
split to exit early on a match.

Fixes: #13388

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/14218 **Author:** [@rick-github](https://github.com/rick-github) **Created:** 2/12/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `bpe` --- ### 📝 Commits (1) - [`5c927a7`](https://github.com/ollama/ollama/commit/5c927a742af7023cbe3fe2c003dd6061225084cf) bpe: merge pre-tok REs to prevent multple matches ### 📊 Changes **1 file changed** (+2 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `tokenizer/bytepairencoding.go` (+2 -0) </details> ### 📄 Description NewBytePairEncoding() takes a list of REs for pre-tokenization. If the REs are not mutually exclusive, this causes token duplication. Models affected: deepseek-v3.1:671b deepseek-r1:671b cogito-2.1:671b r1-1776:671b deepseek-ocr If ported to the ollama engine, the following models would also be affected: granite-code granite3-dense granite3-moe granite3-guardian granite3.1-dense granite3.1-moe granite-3.2 granite3.3 Test: ```console $ ollama run deepseek-v3.1:671b '"1 2 3 4" repeat the previous string' Thinking... First, the user said: "1" then "1 2" then "1 "1 2 3" but that seems messy. Let me read carefully. The user said: "1" then "1 2" then "1 "1 2 3" but that doesn't make sense. Perhaps it's a pattern. Perhaps it's a sequence of strings. Let me see the user's message: "1" then "1 2" then "1 "1 2 3" but that seems off. Maybe it's "1" then "1 2" then "1 2 3" but the user wrote "1 "1 2 3" which might be a typo. ... ``` This PR merges the list into a single RE with alternation. An alternative approach would be to modify the call to NewBytePairEncoding() in the model.go file of affected models to pass a pre-concatenated string, or modify the parsing in [split](https://github.com/ollama/ollama/blob/379fd64fa837e51eddeda6c32f5c812d912a6751/tokenizer/bytepairencoding.go#L50) to exit early on a match. Fixes: #13388 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-23 01:20:31 -05:00
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 pull-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#40448
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?