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 #4993] AI Models stop working after few user only messages. #65194

New Issue
Open
opened 2026-05-03 19:58:33 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-05-03 19:58:33 -05:00
Owner
Copy Link

Originally created by @TheUntitledGoose on GitHub (Jun 12, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4993

What is the issue?

I've tested this with dolphin-llama3:latest and llama-dolphin:latest. I believe this also might be an issue with other models.

This is less of a Ollama issue, but after supplying the messages field with a few user messages such as:

const history = [
  { role: 'user', content: 'respond' },
  { role: 'user', content: 'a' },
  { role: 'user', content: 'la' },
  { role: 'user', content: 'hi' },
  { role: 'user', content: 'hallo' },
  { role: 'user', content: 'test' }
]

I guess it learns to just not speak? It returns just nothing in the content prompts. I was thinking of maybe combining all these into one user content, but then the AI starts speaking as if it were me.

I am sending this to the /api/chat endpoint.

OS

Windows

GPU

Nvidia

CPU

AMD

Ollama version

0.1.43

Originally created by @TheUntitledGoose on GitHub (Jun 12, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4993 ### What is the issue? I've tested this with `dolphin-llama3:latest` and `llama-dolphin:latest`. I believe this also might be an issue with other models. This is less of a Ollama issue, but after supplying the messages field with a few user messages such as: ```js const history = [ { role: 'user', content: 'respond' }, { role: 'user', content: 'a' }, { role: 'user', content: 'la' }, { role: 'user', content: 'hi' }, { role: 'user', content: 'hallo' }, { role: 'user', content: 'test' } ] ``` I guess it learns to just not speak? It returns just nothing in the content prompts. I was thinking of maybe combining all these into one user content, but then the AI starts speaking as if it were me. I am sending this to the `/api/chat` endpoint. ### OS Windows ### GPU Nvidia ### CPU AMD ### Ollama version 0.1.43
GiteaMirror added the bug label 2026-05-03 19:58:33 -05:00
GiteaMirror commented 2026-05-03 19:58:37 -05:00
Author
Owner
Copy Link

@JerrettDavis commented on GitHub (Jun 13, 2024):

This appears to be a model-specific issue. I'm able to repeat it with llama2, dolphin-llama3, and a couple others, but llama3, mistral, and others work fine.

curl http://localhost:11434/api/chat -d '{
  "model": "llama3",
  "messages": [
  { "role": "user", "content": "respond" },
  { "role": "user", "content": "a" },
  { "role": "user", "content": "la" },
  { "role": "user", "content": "hi" },
  { "role": "user", "content": "hallo" },
  { "role": "user", "content": "test" }
  ]
}'
{"model":"llama3","created_at":"2024-06-13T17:53:24.7736032Z","message":{"role":"assistant","content":"PASS"},"done":false}
{"model":"llama3","created_at":"2024-06-13T17:53:24.7849443Z","message":{"role":"assistant","content":"!"},"done":false}
{"model":"llama3","created_at":"2024-06-13T17:53:24.796801Z","message":{"role":"assistant","content":""},"done_reason":"stop","done":true,"total_duration":6056150600,"load_duration":5976816600,"prompt_eval_count":67,"prompt_eval_duration":53089000,"eval_count":3,"eval_duration":22946000}

curl http://localhost:11434/api/chat -d '{
  "model": "mistral",
  "messages": [
  { "role": "user", "content": "respond" },
  { "role": "user", "content": "a" },
  { "role": "user", "content": "la" },
  { "role": "user", "content": "hi" },
  { "role": "user", "content": "hallo" },
  { "role": "user", "content": "test" }
  ]
}'
{"model":"mistral","created_at":"2024-06-13T17:56:18.617751Z","message":{"role":"assistant","content":" Hello"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.628606Z","message":{"role":"assistant","content":"!"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.6377524Z","message":{"role":"assistant","content":" How"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.6475534Z","message":{"role":"assistant","content":" can"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.6583287Z","message":{"role":"assistant","content":" I"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.6676261Z","message":{"role":"assistant","content":" help"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.677224Z","message":{"role":"assistant","content":" you"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.6876013Z","message":{"role":"assistant","content":" today"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.6966181Z","message":{"role":"assistant","content":"?"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7071325Z","message":{"role":"assistant","content":" Let"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7169096Z","message":{"role":"assistant","content":" me"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7266104Z","message":{"role":"assistant","content":" know"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7365124Z","message":{"role":"assistant","content":" if"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7462925Z","message":{"role":"assistant","content":" there"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7557175Z","message":{"role":"assistant","content":"'"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7661804Z","message":{"role":"assistant","content":"s"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7765488Z","message":{"role":"assistant","content":" something"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.7874269Z","message":{"role":"assistant","content":" specific"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.795822Z","message":{"role":"assistant","content":" you"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.8051151Z","message":{"role":"assistant","content":" need"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.816593Z","message":{"role":"assistant","content":" assistance"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.8272572Z","message":{"role":"assistant","content":" with"},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.8368182Z","message":{"role":"assistant","content":"."},"done":false}
{"model":"mistral","created_at":"2024-06-13T17:56:18.8474615Z","message":{"role":"assistant","content":""},"done_reason":"stop","done":true,"total_duration":3069106500,"load_duration":2815637700,"prompt_eval_count":26,"prompt_eval_duration":21056000,"eval_count":24,"eval_duration":229319000}

curl http://localhost:11434/api/chat -d '{
  "model": "llama2",
  "messages": [
  { "role": "user", "content": "respond" },
  { "role": "user", "content": "a" },
  { "role": "user", "content": "la" },
  { "role": "user", "content": "hi" },
  { "role": "user", "content": "hallo" },
  { "role": "user", "content": "test" }
  ]
}'
{"model":"llama2","created_at":"2024-06-13T18:02:00.2955959Z","message":{"role":"assistant","content":""},"done_reason":"stop","done":true,"total_duration":2741720200,"load_duration":2685596200,"prompt_eval_count":122,"prompt_eval_duration":36680000,"eval_count":3,"eval_duration":16694000}
<!-- gh-comment-id:2166461025 --> @JerrettDavis commented on GitHub (Jun 13, 2024): This appears to be a model-specific issue. I'm able to repeat it with llama2, dolphin-llama3, and a couple others, but llama3, mistral, and others work fine. ```bash curl http://localhost:11434/api/chat -d '{ "model": "llama3", "messages": [ { "role": "user", "content": "respond" }, { "role": "user", "content": "a" }, { "role": "user", "content": "la" }, { "role": "user", "content": "hi" }, { "role": "user", "content": "hallo" }, { "role": "user", "content": "test" } ] }' {"model":"llama3","created_at":"2024-06-13T17:53:24.7736032Z","message":{"role":"assistant","content":"PASS"},"done":false} {"model":"llama3","created_at":"2024-06-13T17:53:24.7849443Z","message":{"role":"assistant","content":"!"},"done":false} {"model":"llama3","created_at":"2024-06-13T17:53:24.796801Z","message":{"role":"assistant","content":""},"done_reason":"stop","done":true,"total_duration":6056150600,"load_duration":5976816600,"prompt_eval_count":67,"prompt_eval_duration":53089000,"eval_count":3,"eval_duration":22946000} ``` ---- ```bash curl http://localhost:11434/api/chat -d '{ "model": "mistral", "messages": [ { "role": "user", "content": "respond" }, { "role": "user", "content": "a" }, { "role": "user", "content": "la" }, { "role": "user", "content": "hi" }, { "role": "user", "content": "hallo" }, { "role": "user", "content": "test" } ] }' {"model":"mistral","created_at":"2024-06-13T17:56:18.617751Z","message":{"role":"assistant","content":" Hello"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.628606Z","message":{"role":"assistant","content":"!"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.6377524Z","message":{"role":"assistant","content":" How"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.6475534Z","message":{"role":"assistant","content":" can"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.6583287Z","message":{"role":"assistant","content":" I"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.6676261Z","message":{"role":"assistant","content":" help"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.677224Z","message":{"role":"assistant","content":" you"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.6876013Z","message":{"role":"assistant","content":" today"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.6966181Z","message":{"role":"assistant","content":"?"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7071325Z","message":{"role":"assistant","content":" Let"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7169096Z","message":{"role":"assistant","content":" me"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7266104Z","message":{"role":"assistant","content":" know"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7365124Z","message":{"role":"assistant","content":" if"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7462925Z","message":{"role":"assistant","content":" there"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7557175Z","message":{"role":"assistant","content":"'"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7661804Z","message":{"role":"assistant","content":"s"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7765488Z","message":{"role":"assistant","content":" something"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.7874269Z","message":{"role":"assistant","content":" specific"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.795822Z","message":{"role":"assistant","content":" you"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.8051151Z","message":{"role":"assistant","content":" need"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.816593Z","message":{"role":"assistant","content":" assistance"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.8272572Z","message":{"role":"assistant","content":" with"},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.8368182Z","message":{"role":"assistant","content":"."},"done":false} {"model":"mistral","created_at":"2024-06-13T17:56:18.8474615Z","message":{"role":"assistant","content":""},"done_reason":"stop","done":true,"total_duration":3069106500,"load_duration":2815637700,"prompt_eval_count":26,"prompt_eval_duration":21056000,"eval_count":24,"eval_duration":229319000} ``` ```bash curl http://localhost:11434/api/chat -d '{ "model": "llama2", "messages": [ { "role": "user", "content": "respond" }, { "role": "user", "content": "a" }, { "role": "user", "content": "la" }, { "role": "user", "content": "hi" }, { "role": "user", "content": "hallo" }, { "role": "user", "content": "test" } ] }' {"model":"llama2","created_at":"2024-06-13T18:02:00.2955959Z","message":{"role":"assistant","content":""},"done_reason":"stop","done":true,"total_duration":2741720200,"load_duration":2685596200,"prompt_eval_count":122,"prompt_eval_duration":36680000,"eval_count":3,"eval_duration":16694000} ```
GiteaMirror commented 2026-05-03 19:58:38 -05:00
Author
Owner
Copy Link

@TheUntitledGoose commented on GitHub (Jun 13, 2024):

I've tested with combining all the context into one user prompt, and then it works. I suppose I can also switch to mistral, since that seems to work best.
Sad that llama3, kinda glitches, and some others don't work. Hoped to use the newer ones for better responses.

<!-- gh-comment-id:2166820712 --> @TheUntitledGoose commented on GitHub (Jun 13, 2024): I've tested with combining all the context into one user prompt, and then it works. I suppose I can also switch to mistral, since that seems to work best. Sad that llama3, kinda glitches, and some others don't work. Hoped to use the newer ones for better responses.
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#65194
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?