[BUG] ollama deepseek-r1-tool-calling
Bug Description
</think>
{"name": "mcp__postgres_home_assistant__get_object_details", "parameters": {"object_name": "grid_power", "object_type": "table", "schema_name": "public"}}
tokens used: 3798, model: Ollama (MFDoom/deepseek-r1-tool-calling:32b)
Steps to Reproduce Please provide the steps to reproduce the bug:
- I am using postgresql mcp server with ollama deepseek-r1-tool calling
- Ask something to query the db, which fails; with other models it works.
Expected Results
For example:
I'll check the database for humidity data related to the areca palm. Let me look for sensor data that might contain this information.
mcp__postgres_home_assistant__list_schemas
mcp__postgres_home_assistant__list_objects
mcp__postgres_home_assistant__get_object_details
Let me query for humidity data related to the areca palm by searching for relevant entity IDs in the database:
mcp__postgres_home_assistant__execute_sql
Let me query for the most recent humidity reading from the areca palm sensor:
mcp__postgres_home_assistant__execute_sql
47
tokens used: 3647, model: Ollama (qwen3-coder:30b)
But I am unsure whether bug is in chatbox or ollama is improperly format the tool request?
Desktop (please complete the following information):
- Operating System: Linux
- Application Version: 1.16.4
Is there any error? I don't see any in the log you posted. BTW deepseek-r1's tool use ability is very weak, you can try some other model like qwen or gemma
I checked the main.log and there are no errors.
Yes qwen works, but I easily get this qwen3-coder to "cycle forever" ... and wanted to try with deepseek, which has better reasoning.
If no error, what do you mean by it fails? What behavior is unexpected
If you look at this output, seems like ollama requested a tool, but then everything stopped?
If you see "expected results" above from qwen, it works fine.
` {"name": "mcp__postgres_home_assistant__get_object_details", "parameters": {"object_name": "grid_power", "object_type": "table", "schema_name": "public"}}
tokens used: 3798, model: Ollama (MFDoom/deepseek-r1-tool-calling:32b) `