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 #7075] Hallucination fix? #4491

New Issue
Closed
opened 2026-04-12 15:25:02 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-12 15:25:02 -05:00
Owner
Copy Link

Originally created by @Lu-Yi-Fan on GitHub (Oct 2, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7075

Hi, when i use the models(llama3:70b/llama3:latest) via ollama, it seems to keep track of all the conversations and query. This causes hallucination and information to appear across different channels which shouldnt be the case. What could be a possible remedy for this? Would it be possible to instaniate the model without keeping track of the history. Thank you in advance

Originally created by @Lu-Yi-Fan on GitHub (Oct 2, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7075 Hi, when i use the models(llama3:70b/llama3:latest) via ollama, it seems to keep track of all the conversations and query. This causes hallucination and information to appear across different channels which shouldnt be the case. What could be a possible remedy for this? Would it be possible to instaniate the model without keeping track of the history. Thank you in advance
GiteaMirror added the feature request label 2026-04-12 15:25:02 -05:00
GiteaMirror commented 2026-04-12 15:25:03 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 2, 2024):

How are you querying the model? ollama run or some other client?

<!-- gh-comment-id:2387853601 --> @rick-github commented on GitHub (Oct 2, 2024): How are you querying the model? `ollama run` or some other client?
GiteaMirror commented 2026-04-12 15:25:04 -05:00
Author
Owner
Copy Link

@Lu-Yi-Fan commented on GitHub (Oct 2, 2024):

Hi I am currently running via server side, the ollama model via api calls, using langchain_community.llms import Ollama.
To share more context perhaps: I am currently working on a RAG system that ingests PDFs. And I have a seperate conversation chain and seperate vectorstore for each PDF to "Talk to the PDF". However, the issue is that after many conversations with lets say PDF A and PDF B, information from PDF B would get leaked into the conversation with PDF A. I have checked through my code and made sure the conversation chains have no spillovers and it is not due to the wrong assignment of variables. When I tested it using ollama run to test my hypothesis "Conversations are stored", it is true as the model is able to repeat what I had said earlier on, hence my suspicion to the hallucination. Thank you

<!-- gh-comment-id:2387862719 --> @Lu-Yi-Fan commented on GitHub (Oct 2, 2024): Hi I am currently running via server side, the ollama model via api calls, using langchain_community.llms import Ollama. To share more context perhaps: I am currently working on a RAG system that ingests PDFs. And I have a seperate conversation chain and seperate vectorstore for each PDF to "Talk to the PDF". However, the issue is that after many conversations with lets say PDF A and PDF B, information from PDF B would get leaked into the conversation with PDF A. I have checked through my code and made sure the conversation chains have no spillovers and it is not due to the wrong assignment of variables. When I tested it using ollama run to test my hypothesis "Conversations are stored", it is true as the model is able to repeat what I had said earlier on, hence my suspicion to the hallucination. Thank you
GiteaMirror commented 2026-04-12 15:25:04 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 2, 2024):

The ollama server doesn't store any state. If you are finding that the server is responding as if it is, then that would seem to be a problem with the client. See https://github.com/ollama/ollama/issues/6992 for the exact opposite of your problem, where a user was wondering why the server wasn't using earlier conversations in responses.

ollama run is a client that does preserve conversational history. You can reset the history with the /clear command.

<!-- gh-comment-id:2387975246 --> @rick-github commented on GitHub (Oct 2, 2024): The ollama server doesn't store any state. If you are finding that the server is responding as if it is, then that would seem to be a problem with the client. See https://github.com/ollama/ollama/issues/6992 for the exact opposite of your problem, where a user was wondering why the server wasn't using earlier conversations in responses. `ollama run` is a client that does preserve conversational history. You can reset the history with the `/clear` command.
GiteaMirror commented 2026-04-12 15:25:04 -05:00
Author
Owner
Copy Link

@Lu-Yi-Fan commented on GitHub (Oct 2, 2024):

ok thank you for the findings, I will check and test again and get back to you. Thank you!

<!-- gh-comment-id:2388021176 --> @Lu-Yi-Fan commented on GitHub (Oct 2, 2024): ok thank you for the findings, I will check and test again and get back to you. Thank you!
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 feature 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#4491
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?