Set ggml log level?

[atroyn]

log_disable() from log.h does not seem to suppress logging from ggml:

https://github.com/chroma-core/chroma/blob/3218c966a3c0975fd5b8bbb3f8ef495e6597130b/cpp/llama.cpp/ggml/include/ggml.h#L555-L561

How can I set the ggml loglevel at either compile-time or runtime, when using llama.cpp ?

[ggerganov]

The log.h will soon be completely overhauled. Follow #8566 for updates

[ggerganov]

Which logs from ggml are you looking to suppress? I think there is nothing coming from within the ggml library when you build in Release.

[atroyn]

The log output when loading a model for example:

llama_model_loader: loaded meta data with 22 key-value pairs and 197 tensors from m-model-f16.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv   0:                       general.architecture str              = bert
llama_model_loader: - kv   1:                               general.name str              = snowflake-arctic-embed-m
llama_model_loader: - kv   2:                           bert.block_count u32              = 12
llama_model_loader: - kv   3:                        bert.context_length u32              = 512
llama_model_loader: - kv   4:                      bert.embedding_length u32              = 768
llama_model_loader: - kv   5:                   bert.feed_forward_length u32              = 3072
llama_model_loader: - kv   6:                  bert.attention.head_count u32              = 12
llama_model_loader: - kv   7:          bert.attention.layer_norm_epsilon f32              = 0.000000
llama_model_loader: - kv   8:                          general.file_type u32              = 1
llama_model_loader: - kv   9:                      bert.attention.causal bool             = false
llama_model_loader: - kv  10:                          bert.pooling_type u32              = 2
llama_model_loader: - kv  11:            tokenizer.ggml.token_type_count u32              = 2
llama_model_loader: - kv  12:                       tokenizer.ggml.model str              = bert
llama_model_loader: - kv  13:                         tokenizer.ggml.pre str              = jina-v2-en
llama_model_loader: - kv  14:                      tokenizer.ggml.tokens arr[str,30522]   = ["[PAD]", "[unused0]", "[unused1]", "...
llama_model_loader: - kv  15:                  tokenizer.ggml.token_type arr[i32,30522]   = [3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  16:            tokenizer.ggml.unknown_token_id u32              = 100
llama_model_loader: - kv  17:          tokenizer.ggml.seperator_token_id u32              = 102
llama_model_loader: - kv  18:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  19:                tokenizer.ggml.cls_token_id u32              = 101
llama_model_loader: - kv  20:               tokenizer.ggml.mask_token_id u32              = 103
llama_model_loader: - kv  21:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  124 tensors
llama_model_loader: - type  f16:   73 tensors
llm_load_vocab: special tokens cache size = 5
llm_load_vocab: token to piece cache size = 0.2032 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = bert
llm_load_print_meta: vocab type       = WPM
llm_load_print_meta: n_vocab          = 30522
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 512
llm_load_print_meta: n_embd           = 768
llm_load_print_meta: n_layer          = 12
llm_load_print_meta: n_head           = 12
llm_load_print_meta: n_head_kv        = 12
llm_load_print_meta: n_rot            = 64
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 64
llm_load_print_meta: n_embd_head_v    = 64
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 768
llm_load_print_meta: n_embd_v_gqa     = 768
llm_load_print_meta: f_norm_eps       = 1.0e-12
llm_load_print_meta: f_norm_rms_eps   = 0.0e+00
llm_load_print_meta: f_clamp_kqv      = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 3072
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0

...

[atroyn]

Ah I see, building in Release should clear this - will try.

[ggerganov]

Nah, my bad. To suppress those logs, you can set an empty callback at the start of the program:

int main(int argc, char ** argv) {
    llama_log_set([](ggml_log_level /*level*/, const char * /*text*/, void * /*user_data*/) {
        // noop
    }, NULL);

...

[atroyn]

In my case, I am calling into llama with python bindings so I don't have a main, but will call this on instantiation of the class and see.