Bug : Thinking mode cannot be disabled
There is no way to suppress the <think> block output in lfm2.5-thinking:1.2b.
Expected: When Ollama parameter think is set to false, the model should not output <think> blocks.
Actual: The model always emits a verbose <think> block on every response. think: 'false' isn't working.
Request: Please add support to disable thinking mode via a model parameter (e.g. think: false in Ollama OPTIONS), or provide a non-thinking variant of the model on Ollama.
I also reached out to the ollama github they have given the following statement:
The model doesn't have a non-thinking variant, and the template supplied by the model authors shows no mechanism for controlling thinking. It could probably be faked by adding a template that pre-fills the output with an empty block, but that's not always reliable.
Bug : Thinking mode cannot be disabled
There is no way to suppress the
<think>block output inlfm2.5-thinking:1.2b.Expected: When Ollama parameter
thinkis set tofalse, the model should not output<think>blocks.Actual: The model always emits a verbose
<think>block on every response. think: 'false' isn't working.Request: Please add support to disable thinking mode via a model parameter (e.g.
think: falsein Ollama OPTIONS), or provide a non-thinking variant of the model on Ollama.I also reached out to the ollama github they have given the following statement:
The model doesn't have a non-thinking variant, and the template supplied by the model authors shows no mechanism for controlling thinking. It could probably be faked by adding a template that pre-fills the output with an empty block, but that's not always reliable.