Fixes for rwkv-world template and the missing inputs.use_jinja in llama-cli

[MollySophia]

• Handle the rwkv-world chat template better
• Fix a bug where inputs.use_jinja is not set according to g_params->use_jinja in llama-cli

[CISC]

I have no idea if this is the correct way to set the system prompt for RWKV7, but I'm sure you do, so LGTM. :)

Nice catch on use_jinja!

[MollySophia]

I have no idea if this is the correct way to set the system prompt for RWKV7, but I'm sure you do, so LGTM. :)

Thanks for the reminder! I just checked it again with @BlinkDL and it should indeed be another one.

[zhouwg]

@MollySophia, nice to meet you and sorry to bother you.

today I found your excellent project https://github.com/MollySophia/rwkv-qualcomm
and read the following file carefully.
https://github.com/MollySophia/rwkv-mobile/blob/master/src/backends/qnn/qnn_backend.cpp

as you said in your rwkv-qualcomm project:
MollySophia/rwkv-qualcomm#4 (comment)

I totally agree with what you said: the approach in your project is exactly similar to the approach in the official QNN solution, or it's also an End-to-End inference solution.

obviously, you are an AI expert and also good at/familiar with Qualcomm's QNN SDK.

I know very very very very little knowledge about hardcore AI tech although I'm good at Android sys dev and dev in the embedded system.

I have a candidate PR in the llama.cpp project:#12326

could you help to take a look if you have time?

thanks so much!