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

[GH-ISSUE #11878] Ollama Turbo doesn't support "tool" role in messages #69942

New Issue
Closed
opened 2026-05-04 19:50:21 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-05-04 19:50:21 -05:00
Owner
Copy Link

Originally created by @mgs-ckopsa on GitHub (Aug 13, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11878

Originally assigned to: @ParthSareen on GitHub.

What is the issue?

Steps to reproduce:

This works when the role is set to 'user':

curl -X POST 'https://ollama.com/api/chat' \
  -H 'Host: ollama.com' \
  -H 'Authorization: Bearer abc123' \
  --data '{"model":"gpt-oss:120b","stream":false,"options":{},"messages":[{"role":"user","content": "Hi"}]}'

This yields an upstream error when the role is switched to "tool":

curl -X POST 'https://ollama.com/api/chat' \
  -H 'Host: ollama.com' \
  -H 'Authorization: Bearer abc123' \
  --data '{"model":"gpt-oss:120b","stream":false,"options":{},"messages":[{"role":"tool","content": "Hi"}]}'

This works as expected when running locally, but when using ollama turbo it shows upstream error

Relevant log output

curl -X POST 'https://ollama.com/api/chat' \
  -H 'Authorization: Bearer abc123' \
  --data '{"model":"gpt-oss:120b","stream":false,"options":{},"messages":[{"role":"tool","user": "Hi"}]}'
{"error": "upstream error"}%

OS

No response

GPU

No response

CPU

No response

Ollama version

No response

Originally created by @mgs-ckopsa on GitHub (Aug 13, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11878 Originally assigned to: @ParthSareen on GitHub. ### What is the issue? Steps to reproduce: This works when the role is set to 'user': ```bash curl -X POST 'https://ollama.com/api/chat' \ -H 'Host: ollama.com' \ -H 'Authorization: Bearer abc123' \ --data '{"model":"gpt-oss:120b","stream":false,"options":{},"messages":[{"role":"user","content": "Hi"}]}' ``` This yields an upstream error when the role is switched to "tool": ```bash curl -X POST 'https://ollama.com/api/chat' \ -H 'Host: ollama.com' \ -H 'Authorization: Bearer abc123' \ --data '{"model":"gpt-oss:120b","stream":false,"options":{},"messages":[{"role":"tool","content": "Hi"}]}' ``` This works as expected when running locally, but when using ollama turbo it shows `upstream error` ### Relevant log output ```shell curl -X POST 'https://ollama.com/api/chat' \ -H 'Authorization: Bearer abc123' \ --data '{"model":"gpt-oss:120b","stream":false,"options":{},"messages":[{"role":"tool","user": "Hi"}]}' {"error": "upstream error"}% ``` ### OS _No response_ ### GPU _No response_ ### CPU _No response_ ### Ollama version _No response_
GiteaMirror added the bug label 2026-05-04 19:50:21 -05:00
GiteaMirror commented 2026-05-04 19:50:25 -05:00
Author
Owner
Copy Link

@ParthSareen commented on GitHub (Aug 13, 2025):

Sorry about this! Taking a look now

<!-- gh-comment-id:3185060209 --> @ParthSareen commented on GitHub (Aug 13, 2025): Sorry about this! Taking a look now
GiteaMirror commented 2026-05-04 19:50:26 -05:00
Author
Owner
Copy Link

@ParthSareen commented on GitHub (Aug 13, 2025):

@mgs-ckopsa are you passing in the tool call history and tools? That's most likely causing the issue right now. Here's an example you can try:

curl https://ollama.com/api/chat \
  -X POST \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OLLAMA_TURBO_KEY" \
  -d '{
        "model": "gpt-oss:20b",
        "stream": false,
        "messages": [
          {
            "role": "user",
            "content": "What is the weather like in London? What are the conditions in Toronto?"
          },
          {
            "role": "assistant",
            "thinking": "We have two requests: weather in London, conditions in Toronto. Use functions: get_weather, get_weather_conditions. For London: likely current temperature? get_weather. For Toronto: get_weather_conditions. We need to use tool calls.",
            "tool_calls": [
              {
                "function": {
                  "name": "get_weather",
                  "arguments": {"city": "London"}
                }
              }
            ]
          },
          {
            "role": "tool",
            "content": "The temperature in London is 7°C",
            "tool_name": "get_weather"
          },
          {
            "role": "assistant",
            "thinking": "Now get conditions Toronto.",
            "tool_calls": [
              {
                "function": {
                  "name": "get_weather_conditions",
                  "arguments": {"city": "Toronto"}
                }
              }
            ]
          },
          {
            "role": "tool",
            "content": "rainy",
            "tool_name": "get_weather_conditions"
          }
        ],
        "tools": [
          {
            "type": "function",
            "function": {
              "name": "get_weather",
              "description": "Get the current temperature for a city",
              "parameters": {
                "type": "object",
                "required": ["city"],
                "properties": {
                  "city": {
                    "type": "string",
                    "description": "The name of the city"
                  }
                }
              }
            }
          },
          {
            "type": "function",
            "function": {
              "name": "get_weather_conditions",
              "description": "Get the weather conditions for a city",
              "parameters": {
                "type": "object",
                "required": ["city"],
                "properties": {
                  "city": {
                    "type": "string",
                    "description": "The name of the city"
                  }
                }
              }
            }
          }
        ]
      }'

I will be consolidating the behavior between turbo and local, and return some better errors!

<!-- gh-comment-id:3185922901 --> @ParthSareen commented on GitHub (Aug 13, 2025): @mgs-ckopsa are you passing in the tool call history and tools? That's most likely causing the issue right now. Here's an example you can try: ``` curl https://ollama.com/api/chat \ -X POST \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $OLLAMA_TURBO_KEY" \ -d '{ "model": "gpt-oss:20b", "stream": false, "messages": [ { "role": "user", "content": "What is the weather like in London? What are the conditions in Toronto?" }, { "role": "assistant", "thinking": "We have two requests: weather in London, conditions in Toronto. Use functions: get_weather, get_weather_conditions. For London: likely current temperature? get_weather. For Toronto: get_weather_conditions. We need to use tool calls.", "tool_calls": [ { "function": { "name": "get_weather", "arguments": {"city": "London"} } } ] }, { "role": "tool", "content": "The temperature in London is 7°C", "tool_name": "get_weather" }, { "role": "assistant", "thinking": "Now get conditions Toronto.", "tool_calls": [ { "function": { "name": "get_weather_conditions", "arguments": {"city": "Toronto"} } } ] }, { "role": "tool", "content": "rainy", "tool_name": "get_weather_conditions" } ], "tools": [ { "type": "function", "function": { "name": "get_weather", "description": "Get the current temperature for a city", "parameters": { "type": "object", "required": ["city"], "properties": { "city": { "type": "string", "description": "The name of the city" } } } } }, { "type": "function", "function": { "name": "get_weather_conditions", "description": "Get the weather conditions for a city", "parameters": { "type": "object", "required": ["city"], "properties": { "city": { "type": "string", "description": "The name of the city" } } } } } ] }' ``` I will be consolidating the behavior between turbo and local, and return some better errors!
GiteaMirror commented 2026-05-04 19:50:27 -05:00
Author
Owner
Copy Link

@mgs-ckopsa commented on GitHub (Aug 13, 2025):

The original request included tools and tool call history, but I was just looking for the smallest example that would give the error.

The example you gave here works as expected.

I will see if I can put together a better example that shows my issue., but yes returning better errors would be helpful along those lines.

Just to leave somewhat of a paper trail, I was using the "http_request" tool within Amazon's strand agent python sdk and was consistently getting an upstream error whenever using ollama turbo and no error when using ollama local.

<!-- gh-comment-id:3186017596 --> @mgs-ckopsa commented on GitHub (Aug 13, 2025): The original request included tools and tool call history, but I was just looking for the smallest example that would give the error. The example you gave here works as expected. I will see if I can put together a better example that shows my issue., but yes returning better errors would be helpful along those lines. Just to leave somewhat of a paper trail, I was using the "http_request" tool within Amazon's strand agent python sdk and was consistently getting an upstream error whenever using ollama turbo and no error when using ollama local.
GiteaMirror commented 2026-05-04 19:50:28 -05:00
Author
Owner
Copy Link

@ParthSareen commented on GitHub (Aug 13, 2025):

@mgs-ckopsa yeah apologies for the obscure errors. Will get to that ASAP. And I haven't tried the strand agent but we don't support openai compatibility on ollama.com directly so that could be another cause of the issue. Again once I add some better error messages this should be a bit easier to deal with. We're also going to be adding some new ways of using turbo which hopefully will also improve this entire experience!

<!-- gh-comment-id:3186049111 --> @ParthSareen commented on GitHub (Aug 13, 2025): @mgs-ckopsa yeah apologies for the obscure errors. Will get to that ASAP. And I haven't tried the strand agent but we don't support openai compatibility on ollama.com directly so that could be another cause of the issue. Again once I add some better error messages this should be a bit easier to deal with. We're also going to be adding some new ways of using turbo which hopefully will also improve this entire experience!
GiteaMirror commented 2026-05-04 19:50:28 -05:00
Author
Owner
Copy Link

@ashworthpayne commented on GitHub (Aug 14, 2025):

I think I speak for all the worlds' GPU-poor when I say "Thank you".

If anyone cares, I'd love to be able to push my LLM to Turbo and use it with tools etc. I could pay a bit more for the feature, but keep in mind that I am GPU poor for real reasons 👈

<!-- gh-comment-id:3188790578 --> @ashworthpayne commented on GitHub (Aug 14, 2025): I think I speak for all the worlds' GPU-poor when I say "Thank you". If anyone cares, I'd love to be able to push my LLM to Turbo and use it with tools etc. I could pay a bit more for the feature, but keep in mind that I am GPU poor for real reasons 👈
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#69942
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?