github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #5877] Ollama API not seeing messages provided in conversation_history #3666

New Issue
Closed
opened 2026-04-12 14:27:45 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-12 14:27:45 -05:00
Owner
Copy Link

Originally created by @barclaybrown on GitHub (Jul 23, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/5877

What is the issue?

When I pass a list of dictionaries (messages) to ollama.chat it seems that the model does not see anything other than the latest message. For example, I want the model to get a bunch of text, and then answer a question about it. I send something like:

role : system content: You are a helpful assistant
role: user content: a bunch of reference text
role: user content: a question related to the reference text

then I get back

role: assistant content: an answer unrelated to the reference text, as if it doesn't see it

Is this a bug, or maybe I'm doing something wrong?

OS

Windows

GPU

Nvidia

CPU

No response

Ollama version

0.2.7

Originally created by @barclaybrown on GitHub (Jul 23, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/5877 ### What is the issue? When I pass a list of dictionaries (messages) to ollama.chat it seems that the model does not see anything other than the latest message. For example, I want the model to get a bunch of text, and then answer a question about it. I send something like: role : system content: You are a helpful assistant role: user content: a bunch of reference text role: user content: a question related to the reference text then I get back role: assistant content: an answer unrelated to the reference text, as if it doesn't see it Is this a bug, or maybe I'm doing something wrong? ### OS Windows ### GPU Nvidia ### CPU _No response_ ### Ollama version 0.2.7
GiteaMirror added the bug label 2026-04-12 14:27:45 -05:00
GiteaMirror commented 2026-04-12 14:27:48 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 23, 2024):

Can you provide an example? It looks to work here:

$ curl -s localhost:11434/api/chat -d '{
  "model": "qwen2:0.5b",
  "messages": [
    {
      "role": "user",
      "content": "an apple is $1.50"
    },
    {
      "role": "user",
      "content": "how much is an apple?"
    }
  ],
  "stream": false
}' | jq .message

{
  "role": "assistant",
  "content": "An apple costs $1.50."
}
<!-- gh-comment-id:2245487070 --> @rick-github commented on GitHub (Jul 23, 2024): Can you provide an example? It looks to work here: ```sh $ curl -s localhost:11434/api/chat -d '{ "model": "qwen2:0.5b", "messages": [ { "role": "user", "content": "an apple is $1.50" }, { "role": "user", "content": "how much is an apple?" } ], "stream": false }' | jq .message ``` ```json { "role": "assistant", "content": "An apple costs $1.50." } ```
GiteaMirror commented 2026-04-12 14:27:49 -05:00
Author
Owner
Copy Link

@barclaybrown commented on GitHub (Jul 23, 2024):

Thanks very much This might have given me the hint I needed. The text I'm supplying in the first user message is 45000 tokens, which I thought Phi3 128k should be fine with. Is there a chance that Ollama is throwing it out because it's too long? My same example worked, like yours, when I chopped the text down to about 4k tokens. If there's a limit to user message length, I could split up my text?

<!-- gh-comment-id:2245604681 --> @barclaybrown commented on GitHub (Jul 23, 2024): Thanks very much This might have given me the hint I needed. The text I'm supplying in the first user message is 45000 tokens, which I thought Phi3 128k should be fine with. Is there a chance that Ollama is throwing it out because it's too long? My same example worked, like yours, when I chopped the text down to about 4k tokens. If there's a limit to user message length, I could split up my text?
GiteaMirror commented 2026-04-12 14:27:50 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jul 23, 2024):

The default context length for ollama is 2048, so it's likely the beginning text is being discarded (check the server logs for "input truncated"). You can pass an option in the API call to increase the window:

$ curl -s localhost:11434/api/chat -d '{
  "model": "qwen2:0.5b",
  "messages": [
    {
      "role": "user",
      "content": "an apple is $1.50"
    },
    {
      "role": "user",
      "content": "how much is an apple?"
    }
  ],
  "options": { "num_ctx": 45000 },
  "stream": false
}' | jq .message

Be aware that this can significantly increase the VRAM/RAM requirements for the model, memory usage is quadratic on the length of the context window.

<!-- gh-comment-id:2245621867 --> @rick-github commented on GitHub (Jul 23, 2024): The default context length for ollama is 2048, so it's likely the beginning text is being discarded (check the server logs for "input truncated"). You can pass an option in the API call to increase the window: ``` $ curl -s localhost:11434/api/chat -d '{ "model": "qwen2:0.5b", "messages": [ { "role": "user", "content": "an apple is $1.50" }, { "role": "user", "content": "how much is an apple?" } ], "options": { "num_ctx": 45000 }, "stream": false }' | jq .message ``` Be aware that this can significantly increase the VRAM/RAM requirements for the model, memory usage is quadratic on the length of the context window.
GiteaMirror commented 2026-04-12 14:27:51 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Sep 12, 2024):

@rick-github 's answer is correct. I'll go ahead and close this out.

<!-- gh-comment-id:2347322435 --> @pdevine commented on GitHub (Sep 12, 2024): @rick-github 's answer is correct. I'll go ahead and close this out.
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
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#3666
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?