github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #12702] deepseek-v3.1:671b-cloud not parsing tool calls correctly #54943

New Issue
Closed
opened 2026-04-29 08:03:44 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-29 08:03:44 -05:00
Owner
Copy Link

Originally created by @jonigl on GitHub (Oct 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12702

Originally assigned to: @ParthSareen on GitHub.

What is the issue?

I noticed that when using the model deepseek-v3.1:671b-cloud, tool calls are not being parsed correctly. The tool responses appears in the thinking content output but aren’t properly recognized or executed.

I’ve been testing with ollmcp (mcp-client-for-ollama) and a simple MCP server that provides weather data.

Here’s a asciinema showing the issue:

Image

Steps to Reproduce:

  1. Run ollmcp (mcp-client-for-ollama) with a simple MCP server (e.g., a weather data provider).
  2. Use the model deepseek-v3.1:671b-cloud.
  3. Ask the model to use the weather tool (e.g., “What’s the weather in Paris right now?”).
  4. Observe that the model generates a tool call response, but it isn’t parsed or executed.

Expected Behavior:

The model’s tool call output should be parsed correctly and executed by the MCP client, returning the weather data from the tool.

Actual Behavior:

Tool call responses appear in the output but aren’t parsed — they’re treated as plain text instead of structured tool invocations.
See the attached asciinema for an example.

Environment:

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

No response

Originally created by @jonigl on GitHub (Oct 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12702 Originally assigned to: @ParthSareen on GitHub. ### What is the issue? I noticed that when using the model `deepseek-v3.1:671b-cloud`, tool calls are not being parsed correctly. The tool responses appears in the thinking content output but aren’t properly recognized or executed. I’ve been testing with [ollmcp (mcp-client-for-ollama)](https://github.com/jonigl/mcp-client-for-ollama) and a simple MCP server that provides weather data. Here’s a asciinema showing the issue: [![Image](https://github.com/user-attachments/assets/3fed6f16-f060-4e0d-8841-6bec562fcc4b)](https://asciinema.org/a/dlrYn2fthSIgNKlvKL8wIHTZR) --- ### **Steps to Reproduce:** 1. Run `ollmcp` (mcp-client-for-ollama) with a simple MCP server (e.g., a weather data provider). 2. Use the model `deepseek-v3.1:671b-cloud`. 3. Ask the model to use the weather tool (e.g., “What’s the weather in Paris right now?”). 4. Observe that the model generates a tool call response, but it isn’t parsed or executed. --- ### **Expected Behavior:** The model’s tool call output should be parsed correctly and executed by the MCP client, returning the weather data from the tool. --- ### **Actual Behavior:** Tool call responses appear in the output but aren’t parsed — they’re treated as plain text instead of structured tool invocations. See the attached asciinema for an example. --- ### **Environment:** * **Client:** [ollmcp v0.20.0 (mcp-client-for-ollama)](https://github.com/jonigl/mcp-client-for-ollama) * **Ollama version:** 0.12.6 * **Model:** `deepseek-v3.1:671b-cloud` * **MCP Server:** Simple weather tool (custom) * **OS:** macOS 15.7 (Apple M1) ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version _No response_
GiteaMirror added the bug label 2026-04-29 08:03:44 -05:00
GiteaMirror commented 2026-04-29 08:03:49 -05:00
Author
Owner
Copy Link

@ParthSareen commented on GitHub (Oct 21, 2025):

Hey! DeepseekV3.1 should be used without thinking if you want to use tools. Sorry for the confusion!

<!-- gh-comment-id:3428898972 --> @ParthSareen commented on GitHub (Oct 21, 2025): Hey! DeepseekV3.1 should be used without thinking if you want to use tools. Sorry for the confusion!
GiteaMirror commented 2026-04-29 08:03:51 -05:00
Author
Owner
Copy Link

@unixmonk commented on GitHub (Oct 21, 2025):

You need the deepseek-chat model for tool-calling.

<!-- gh-comment-id:3429404339 --> @unixmonk commented on GitHub (Oct 21, 2025): You need the deepseek-chat model for tool-calling.
GiteaMirror commented 2026-04-29 08:03:52 -05:00
Author
Owner
Copy Link

@jonigl commented on GitHub (Oct 21, 2025):

Hey @ParthSareen!

I didn’t realize that. I had assumed both features worked together since they are mentioned on the Ollama page: https://ollama.com/library/deepseek-v3.1.

After your clarification, I checked the Hugging Face page and found this line 🤦‍♂️:

“Toolcall is supported in non-thinking mode.”
(https://huggingface.co/deepseek-ai/DeepSeek-V3.1#toolcall)

I think it might be helpful to include this note on the Ollama site as well for clarity.

Does the API indicate anywhere that tool use is disabled when thinking mode is enabled? Maybe through an attribute or a response field? That could help confirm when this limitation applies to the cloud model.

Thanks again for pointing this out! 🙂

<!-- gh-comment-id:3429840365 --> @jonigl commented on GitHub (Oct 21, 2025): Hey @ParthSareen! I didn’t realize that. I had assumed both features worked together since they are mentioned on the Ollama page: [https://ollama.com/library/deepseek-v3.1](https://ollama.com/library/deepseek-v3.1). After your clarification, I checked the Hugging Face page and found this line 🤦‍♂️: > “Toolcall is supported in non-thinking mode.” > ([https://huggingface.co/deepseek-ai/DeepSeek-V3.1#toolcall](https://huggingface.co/deepseek-ai/DeepSeek-V3.1#toolcall)) I think it might be helpful to include this note on the Ollama site as well for clarity. Does the API indicate anywhere that tool use is disabled when thinking mode is enabled? Maybe through an attribute or a response field? That could help confirm when this limitation applies to the cloud model. Thanks again for pointing this out! 🙂
GiteaMirror commented 2026-04-29 08:03:53 -05:00
Author
Owner
Copy Link

@ParthSareen commented on GitHub (Oct 21, 2025):

Will probably add a check somewhere as it's difficult to find + not documented on our end. Will fix!

<!-- gh-comment-id:3429883747 --> @ParthSareen commented on GitHub (Oct 21, 2025): Will probably add a check somewhere as it's difficult to find + not documented on our end. Will fix!
GiteaMirror commented 2026-04-29 08:03:53 -05:00
Author
Owner
Copy Link

@jonigl commented on GitHub (Oct 22, 2025):

Thanks, that sounds great!

<!-- gh-comment-id:3429971610 --> @jonigl commented on GitHub (Oct 22, 2025): Thanks, that sounds great!
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#54943
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?