Misc. bug: --jinja for llama-cli is not enabled by default as the documentation indicates

[ctataryn]

Name and Version

$ llama-cli --version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce RTX 3070, compute capability 8.6, VMM: yes
version: 7149 (134e6940c)
built with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-cli

Command line

llama-cli -m ./qwen2.5-1.5b-instruct.gguf --lora ./qwen-lora.gguf

Problem description & steps to reproduce

I noticed this issue when I was using the --lora switch for running a fine-tuning LoRA along with a base model. The LoRA didn't seem to take effect, that is, asking the model questions with it enabled rendered the same results as not using the LoRA at all. If I use --jinja however, the fine-tuning LoRA did take effect.

The documentation for llama-cli indicates that --jinja is the default behaviour. However the following pull request suggests that --jinja is only enabled by default when running in server mode, not cli:

#17524

If I can figure out the auto-gen documentation thing I'll also submit a pull request to correct the -cli README

First Bad Commit

No response

Relevant log output

Logs

[CISC]

Your build is just too old, this didn't happen before b7353.

[ctataryn]

@CISC I'm looking at the documentation on master it still says that the default behaviour for llama-cli is that --jinja is enabled by default. But correct me if I'm wrong somehow.

UPDATE: Or perhaps you're saying that now for the cli it is indeed default behaviour?

[CISC]

UPDATE: Or perhaps you're saying that now for the cli it is indeed default behaviour?

Yes, exactly. :)

[ctataryn]

@CISC Thank you for the clarification, sorry for wasting cycles on this!