Misc. bug: Inconsistent handling of boolean command line arguments in environment

[sawntoe]

Name and Version

version: 6517 (69ffd89)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu

Operating systems

Linux

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

llama-server

Command line

llama-server

Problem description & steps to reproduce

LLAMA_ARG_JINJA env variable only accepts true and false does not accept truthy arguments like enabled or disabled unlike other variables like LLAMA_FLASH_ATTN. This discrepancy is made worse by the fact that the terminology used to describe boolean variables in environment in the doc is enabled/disabled, and there is nothing in the log to indicate whether jinja has been enabled or not. Needless to say this has and will cause much headaches to anyone attempting to configure llama-server via environment variables.

First Bad Commit

No response

Relevant log output

[ngxson]

LLAMA_FLASH_ATTN is a tri-state: enabled, disabled or auto. It is not a boolean

[sawntoe]

I would humbly argue that doesn't matter.

add_opt(common_arg({ "-fa", "--flash-attn" }, "[on|off|auto]",
                       string_format("set Flash Attention use ('on', 'off', or 'auto', default: '%s')",
                                     llama_flash_attn_type_name(params.flash_attn_type)),
                       [](common_params & params, const std::string & value) {
                           if (is_truthy(value)) {
                               params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_ENABLED;
                           } else if (is_falsey(value)) {
                               params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_DISABLED;
                           } else if (is_autoy(value)) {
                               params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_AUTO;
                           } else {
                               throw std::runtime_error(
                                   string_format("error: unkown value for --flash-attn: '%s'\n", value.c_str()));
                           }
                       }).set_env("LLAMA_ARG_FLASH_ATTN"));

There is very clearly truthy/falsey checking that is not being implemented globally. With no relevant documentation, its impossible to tell what config values to supply without delving into the source code.