github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-05 23:53:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #831] Context modification #46908

New Issue
Closed
opened 2026-04-28 01:58:42 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-28 01:58:42 -05:00
Owner
Copy Link

Originally created by @VladimirKras on GitHub (Oct 18, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/831

Sometimes I would like to steer a dialogue in a certain direction by adding a fake message on behalf of the LLM. How to achieve that with Ollama seems quite opaque:

  1. The context that is sent is just an array of token ids, which is hard to manipulate.
  2. The tokenizer and de-tokenizer aren't exposed.
Originally created by @VladimirKras on GitHub (Oct 18, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/831 Sometimes I would like to steer a dialogue in a certain direction by adding a fake message on behalf of the LLM. How to achieve that with Ollama seems quite opaque: 1. The context that is sent is just an array of token ids, which is hard to manipulate. 2. The tokenizer and de-tokenizer aren't exposed.
GiteaMirror added the questionfeedback wanted labels 2026-04-28 01:58:49 -05:00
GiteaMirror commented 2026-04-28 01:58:52 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Oct 25, 2023):

The context is intended to be opaque to the user. In that sense, you're expected to pass into the context what was returned by the previous response.

If your use case requires changing history, it'll probably be better to not use the context at all but track the conversation yourself.

<!-- gh-comment-id:1779903701 --> @mxyng commented on GitHub (Oct 25, 2023): The context is intended to be opaque to the user. In that sense, you're expected to pass into the context what was returned by the previous response. If your use case requires changing history, it'll probably be better to not use the context at all but track the conversation yourself.
GiteaMirror commented 2026-04-28 01:58:54 -05:00
Author
Owner
Copy Link

@MelihOzb commented on GitHub (Oct 28, 2023):

The context shouldn't be opaque to the user, assume I want to load a past conversation with AI (just like you can do with ChatGPT and other LLMs). If we can insert context (i.e. access to tokenizer) then we can read the whole conversation from a database, tokenize them and insert them into context and continue where we left. But now, I have to choose another method to be able to continue past conversation.

<!-- gh-comment-id:1783679216 --> @MelihOzb commented on GitHub (Oct 28, 2023): The context shouldn't be opaque to the user, assume I want to load a past conversation with AI (just like you can do with ChatGPT and other LLMs). If we can insert context (i.e. access to tokenizer) then we can read the whole conversation from a database, tokenize them and insert them into context and continue where we left. But now, I have to choose another method to be able to continue past conversation.
GiteaMirror commented 2026-04-28 01:58:56 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Oct 28, 2023):

@MelihOzb the current implementation supports that without needing to tokenize a previous conversation. You can see an example of exactly what you've described in https://github.com/mxyng/discollama. It saves the context in redis which can be loaded at a later point to continue the conversation.

<!-- gh-comment-id:1783708598 --> @mxyng commented on GitHub (Oct 28, 2023): @MelihOzb the current implementation supports that without needing to tokenize a previous conversation. You can see an example of exactly what you've described in https://github.com/mxyng/discollama. It [saves](https://github.com/mxyng/discollama/blob/main/discollama.py#L61) the context in redis which can be [loaded](https://github.com/mxyng/discollama/blob/main/discollama.py#L69) at a later point to continue the conversation.
GiteaMirror commented 2026-04-28 01:58:57 -05:00
Author
Owner
Copy Link

@emsi commented on GitHub (Dec 1, 2023):

OK, so regardless the context modification, how I can fake previous LLM response with ollama? This technique is used to better guide models without system prompt and acts as an example. Is it possible at all with ollama?

<!-- gh-comment-id:1836754854 --> @emsi commented on GitHub (Dec 1, 2023): OK, so regardless the context modification, how I can fake previous LLM response with ollama? This technique is used to better guide models without system prompt and acts as an example. Is it possible at all with ollama?
GiteaMirror commented 2026-04-28 01:58:57 -05:00
Author
Owner
Copy Link

@emsi commented on GitHub (Dec 14, 2023):

It's possible that chat api #1392 addresses that but I didn't test it yet.

<!-- gh-comment-id:1855810201 --> @emsi commented on GitHub (Dec 14, 2023): It's possible that chat api #1392 addresses that but I didn't test it yet.
GiteaMirror commented 2026-04-28 01:58:58 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Jan 16, 2024):

While it might not the intent of the issue creator, you can create and modify chat history as well as use multi-shot inputs using the chat API.

<!-- gh-comment-id:1894621930 --> @mxyng commented on GitHub (Jan 16, 2024): While it might not the intent of the issue creator, you can create and modify chat history as well as use multi-shot inputs using the chat API.
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 feedback wanted question
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#46908
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?