common: support negated args

[ngxson]

Add better support for negated args like -no-cnv or --no-jinja, example:

--jinja, --no-jinja                     whether to use jinja template engine for chat (default: disabled)
                                        (env: LLAMA_ARG_JINJA)

For boolean options like --mmap or --kv-offload, the environment variable is handled as shown in this example:

• LLAMA_ARG_MMAP=true means enabled, other accepted values are: 1, on, enabled
• LLAMA_ARG_MMAP=false means disabled, other accepted values are: 0, off, disabled
• If LLAMA_ARG_NO_MMAP is present (no matter the value), it means disabling mmap

---

Note: Before this change, LLAMA_ARG_JINJA=false will simply skip the --jinja arg but does not set --no-jinja. Now, falsey values like 0, false, no, disabled will trigger the negated value.

Same logic also apply to the recently-added preset feature: #17859

[CISC]

Would it make sense to add positive variants to --no-display-prompt, --no-show-timings, --no-warmup, --no-kv-offload, --no-repack, --no-mmproj-offload, --no-mmap, --no-op-offload, --no-ppl, --no-webui, --no-models-autoload, --no-prefill-assistant, (--no-perf?) and give them the same treatment for consistency? --no-host and --no-mmproj are tricky though. :)

Or perhaps just leave the positive variant empty and handle that?

Many of these can benefit from rewording.

[ngxson]

Yes I think we can add negation for most of the boolean variables to establish a pattern. The goal is to encourage future contributors to follow the same path. Wdyt @ggerganov ?

[ggerganov]

Sounds good

[ngxson]

I added the support for most of the args with 2 exceptions:

• --no-mmproj now has 2 aliases: --mmproj-auto and --no-mmproj-auto, avoid clashing with the existing --mmproj option
• --no-host is kept as-is (no positive flag) because I'm not quite sure how this function works. Therefore, I cannot come up with a better flag name or a better help message. Suggestions are welcome

[ngxson]

A small nits is that some args have this in the help message: (default: true|false), while some use enabled|disabled

Probably we can have a string_format that automatically populate the default value to the help message, but it's better an optimization to be done in another PR

[CISC]

Regen server docs again.

[ngxson]

already regen in e73c42b

[ggerganov]

The server is not handling the negated arguments correctly:

./bin/llama-server --no-mmap

0.00.000.443 W main: setting n_parallel = 4 and kv_unified = true (add -kvu to disable this)
0.00.000.450 I build: 7376 (380b4c984) with AppleClang 17.0.0.17000319 for Darwin arm64 (debug)
0.00.000.466 I system info: n_threads = 16, n_threads_batch = 16, total_threads = 24
0.00.000.468 I 
0.00.000.483 I system_info: n_threads = 16 (n_threads_batch = 16) / 24 | Metal : EMBED_LIBRARY = 1 | CPU : NEON = 1 | ARM_FMA = 1 | FP16_VA = 1 | MATMUL_INT8 = 1 | DOTPROD = 1 | ACCELERATE = 1 | REPACK = 1 | 
0.00.000.485 I 
0.00.000.516 I init: using 23 threads for HTTP server
WARNING: Using native backtrace. Set GGML_BACKTRACE_LLDB for more info.
WARNING: GGML_BACKTRACE_LLDB may cause native MacOS Terminal.app to crash.
See: https://github.com/ggml-org/llama.cpp/pull/17869
0   libggml-base.0.9.4.dylib            0x0000000103fad4c8 ggml_print_backtrace_symbols + 48
1   libggml-base.0.9.4.dylib            0x0000000103fad2f8 ggml_print_backtrace + 160
2   libggml-base.0.9.4.dylib            0x0000000103fc43c8 _ZL23ggml_uncaught_exceptionv + 12
3   libc++abi.dylib                     0x00000001817f0c2c _ZSt11__terminatePFvvE + 16
4   libc++abi.dylib                     0x00000001817f4394 __cxa_get_exception_ptr + 0
5   libc++abi.dylib                     0x00000001817f433c _ZN10__cxxabiv1L12failed_throwEPNS_15__cxa_exceptionE + 0
6   llama-server                        0x000000010244ede0 _Z19common_params_parseiPPc13llama_exampleRNSt3__13mapI10common_argNS2_12basic_stringIcNS2_11char_traitsIcEENS2_9allocatorIcEEEENS2_4lessIS4_EENS8_INS2_4pairIKS4_SA_EEEEEE + 860
7   llama-server                        0x00000001022ded9c _ZN14server_presetsC2EiPPcR13common_paramsRKNSt3__112basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEE + 852
8   llama-server                        0x00000001022df450 _ZN14server_presetsC1EiPPcR13common_paramsRKNSt3__112basic_stringIcNS4_11char_traitsIcEENS4_9allocatorIcEEEE + 60
9   llama-server                        0x00000001022e06bc _ZN13server_modelsC2ERK13common_paramsiPPcS4_ + 180
10  llama-server                        0x00000001022e1288 _ZN13server_modelsC1ERK13common_paramsiPPcS4_ + 60
11  llama-server                        0x00000001022ad8d0 _ZN20server_models_routesC2ERK13common_paramsiPPcS4_ + 80
12  llama-server                        0x00000001022ad870 _ZN20server_models_routesC1ERK13common_paramsiPPcS4_ + 60
13  llama-server                        0x00000001022ad824 _ZNSt3__114__construct_atB8ne200100I20server_models_routesJR13common_paramsRiRPPcS7_EPS1_EEPT_SA_DpOT0_ + 72
14  llama-server                        0x00000001022ad7bc _ZNSt3__123__optional_storage_baseI20server_models_routesLb0EE11__constructB8ne200100IJR13common_paramsRiRPPcS9_EEEvDpOT_ + 60
15  llama-server                        0x0000000102291e54 _ZNSt3__18optionalI20server_models_routesE7emplaceB8ne200100IJR13common_paramsRiRPPcS9_EvEERS1_DpOT_ + 68
16  llama-server                        0x000000010228f780 main + 1276
17  dyld                                0x0000000181475d54 start + 7184
libc++abi: terminating due to uncaught exception of type std::invalid_argument: error: invalid argument: --no-mmap
Abort trap: 6

[ngxson]

@ggerganov ah yeah, I forgot the change the function handling presets on server (it only affect router mode), the fix is here: #17991