Skip to content

Commit 3ff2279

Browse files
ssam18ggerganov
authored and
arthw
committed
fix: GLM-DSA crash in llama-tokenize when using vocab_only (ggml-org#22102)
* llama: fix crash in print_info for GLM-DSA when vocab_only is set * addressed code review comments * cont : simplify --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
1 parent 1a43b5e commit 3ff2279

1 file changed

Lines changed: 93 additions & 93 deletions

File tree

‎src/llama-model.cpp‎

Lines changed: 93 additions & 93 deletions
Original file line numberDiff line numberDiff line change
@@ -8180,114 +8180,114 @@ void llama_model::print_info() const {
81808180
LLAMA_LOG_INFO("%s: n_cls_out = %u\n", __func__, hparams.n_cls_out);
81818181

81828182
size_t i = 0;
8183-
for (auto label : classifier_labels) {
8183+
for (const auto & label : classifier_labels) {
81848184
LLAMA_LOG_INFO("%s: cls_label[%2zu] = %s\n", __func__, i++, label.c_str());
81858185
}
81868186
}
8187-
}
81888187

8189-
if (arch == LLM_ARCH_MAMBA ||
8190-
arch == LLM_ARCH_MAMBA2 ||
8191-
arch == LLM_ARCH_JAMBA ||
8192-
arch == LLM_ARCH_FALCON_H1 ||
8193-
arch == LLM_ARCH_PLAMO2 ||
8194-
arch == LLM_ARCH_GRANITE_HYBRID ||
8195-
arch == LLM_ARCH_QWEN3NEXT ||
8196-
arch == LLM_ARCH_QWEN35 ||
8197-
arch == LLM_ARCH_QWEN35MOE ||
8198-
arch == LLM_ARCH_NEMOTRON_H ||
8199-
arch == LLM_ARCH_NEMOTRON_H_MOE) {
8200-
LLAMA_LOG_INFO("%s: ssm_d_conv = %u\n", __func__, hparams.ssm_d_conv);
8201-
LLAMA_LOG_INFO("%s: ssm_d_inner = %u\n", __func__, hparams.ssm_d_inner);
8202-
LLAMA_LOG_INFO("%s: ssm_d_state = %u\n", __func__, hparams.ssm_d_state);
8203-
LLAMA_LOG_INFO("%s: ssm_dt_rank = %u\n", __func__, hparams.ssm_dt_rank);
8204-
LLAMA_LOG_INFO("%s: ssm_n_group = %u\n", __func__, hparams.ssm_n_group);
8205-
LLAMA_LOG_INFO("%s: ssm_dt_b_c_rms = %d\n", __func__, hparams.ssm_dt_b_c_rms);
8206-
}
8188+
if (arch == LLM_ARCH_MAMBA ||
8189+
arch == LLM_ARCH_MAMBA2 ||
8190+
arch == LLM_ARCH_JAMBA ||
8191+
arch == LLM_ARCH_FALCON_H1 ||
8192+
arch == LLM_ARCH_PLAMO2 ||
8193+
arch == LLM_ARCH_GRANITE_HYBRID ||
8194+
arch == LLM_ARCH_QWEN3NEXT ||
8195+
arch == LLM_ARCH_QWEN35 ||
8196+
arch == LLM_ARCH_QWEN35MOE ||
8197+
arch == LLM_ARCH_NEMOTRON_H ||
8198+
arch == LLM_ARCH_NEMOTRON_H_MOE) {
8199+
LLAMA_LOG_INFO("%s: ssm_d_conv = %u\n", __func__, hparams.ssm_d_conv);
8200+
LLAMA_LOG_INFO("%s: ssm_d_inner = %u\n", __func__, hparams.ssm_d_inner);
8201+
LLAMA_LOG_INFO("%s: ssm_d_state = %u\n", __func__, hparams.ssm_d_state);
8202+
LLAMA_LOG_INFO("%s: ssm_dt_rank = %u\n", __func__, hparams.ssm_dt_rank);
8203+
LLAMA_LOG_INFO("%s: ssm_n_group = %u\n", __func__, hparams.ssm_n_group);
8204+
LLAMA_LOG_INFO("%s: ssm_dt_b_c_rms = %d\n", __func__, hparams.ssm_dt_b_c_rms);
8205+
}
82078206

8208-
LLAMA_LOG_INFO("%s: model type = %s\n", __func__, type_name().c_str());
8209-
if (pimpl->n_elements >= 1e12) {
8210-
LLAMA_LOG_INFO("%s: model params = %.2f T\n", __func__, pimpl->n_elements*1e-12);
8211-
} else if (pimpl->n_elements >= 1e9) {
8212-
LLAMA_LOG_INFO("%s: model params = %.2f B\n", __func__, pimpl->n_elements*1e-9);
8213-
} else if (pimpl->n_elements >= 1e6) {
8214-
LLAMA_LOG_INFO("%s: model params = %.2f M\n", __func__, pimpl->n_elements*1e-6);
8215-
} else {
8216-
LLAMA_LOG_INFO("%s: model params = %.2f K\n", __func__, pimpl->n_elements*1e-3);
8217-
}
8207+
LLAMA_LOG_INFO("%s: model type = %s\n", __func__, type_name().c_str());
8208+
if (pimpl->n_elements >= 1e12) {
8209+
LLAMA_LOG_INFO("%s: model params = %.2f T\n", __func__, pimpl->n_elements*1e-12);
8210+
} else if (pimpl->n_elements >= 1e9) {
8211+
LLAMA_LOG_INFO("%s: model params = %.2f B\n", __func__, pimpl->n_elements*1e-9);
8212+
} else if (pimpl->n_elements >= 1e6) {
8213+
LLAMA_LOG_INFO("%s: model params = %.2f M\n", __func__, pimpl->n_elements*1e-6);
8214+
} else {
8215+
LLAMA_LOG_INFO("%s: model params = %.2f K\n", __func__, pimpl->n_elements*1e-3);
8216+
}
82188217

8219-
// general kv
8220-
LLAMA_LOG_INFO("%s: general.name = %s\n", __func__, name.c_str());
8218+
// general kv
8219+
LLAMA_LOG_INFO("%s: general.name = %s\n", __func__, name.c_str());
82218220

8222-
if (arch == LLM_ARCH_DEEPSEEK) {
8223-
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8224-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8225-
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8226-
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8227-
}
8221+
if (arch == LLM_ARCH_DEEPSEEK) {
8222+
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8223+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8224+
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8225+
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8226+
}
82288227

8229-
if (arch == LLM_ARCH_DEEPSEEK2 || arch == LLM_ARCH_DEEPSEEK2OCR || arch == LLM_ARCH_GLM_DSA || arch == LLM_ARCH_MISTRAL4) {
8230-
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8231-
LLAMA_LOG_INFO("%s: n_lora_q = %d\n", __func__, hparams.n_lora_q);
8232-
LLAMA_LOG_INFO("%s: n_lora_kv = %d\n", __func__, hparams.n_lora_kv);
8233-
LLAMA_LOG_INFO("%s: n_embd_head_k_mla = %d\n", __func__, hparams.n_embd_head_k_mla());
8234-
LLAMA_LOG_INFO("%s: n_embd_head_v_mla = %d\n", __func__, hparams.n_embd_head_v_mla());
8235-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8236-
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8237-
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8238-
LLAMA_LOG_INFO("%s: expert_weights_norm = %d\n", __func__, hparams.expert_weights_norm);
8239-
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
8240-
}
8228+
if (arch == LLM_ARCH_DEEPSEEK2 || arch == LLM_ARCH_DEEPSEEK2OCR || arch == LLM_ARCH_GLM_DSA || arch == LLM_ARCH_MISTRAL4) {
8229+
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8230+
LLAMA_LOG_INFO("%s: n_lora_q = %d\n", __func__, hparams.n_lora_q);
8231+
LLAMA_LOG_INFO("%s: n_lora_kv = %d\n", __func__, hparams.n_lora_kv);
8232+
LLAMA_LOG_INFO("%s: n_embd_head_k_mla = %d\n", __func__, hparams.n_embd_head_k_mla());
8233+
LLAMA_LOG_INFO("%s: n_embd_head_v_mla = %d\n", __func__, hparams.n_embd_head_v_mla());
8234+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8235+
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8236+
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8237+
LLAMA_LOG_INFO("%s: expert_weights_norm = %d\n", __func__, hparams.expert_weights_norm);
8238+
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
8239+
}
82418240

8242-
if (arch == LLM_ARCH_QWEN2MOE) {
8243-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8244-
LLAMA_LOG_INFO("%s: n_ff_shexp = %d\n", __func__, hparams.n_ff_shexp);
8245-
}
8241+
if (arch == LLM_ARCH_QWEN2MOE) {
8242+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8243+
LLAMA_LOG_INFO("%s: n_ff_shexp = %d\n", __func__, hparams.n_ff_shexp);
8244+
}
82468245

8247-
if (arch == LLM_ARCH_QWEN3MOE || arch == LLM_ARCH_OPENAI_MOE || arch == LLM_ARCH_QWEN3VLMOE || arch == LLM_ARCH_RND1) {
8248-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8249-
}
8246+
if (arch == LLM_ARCH_QWEN3MOE || arch == LLM_ARCH_OPENAI_MOE || arch == LLM_ARCH_QWEN3VLMOE || arch == LLM_ARCH_RND1) {
8247+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8248+
}
82508249

8251-
if (arch == LLM_ARCH_MINICPM ||
8252-
arch == LLM_ARCH_GRANITE ||
8253-
arch == LLM_ARCH_GRANITE_MOE ||
8254-
arch == LLM_ARCH_GRANITE_HYBRID ||
8255-
arch == LLM_ARCH_NEMOTRON_H_MOE) {
8256-
LLAMA_LOG_INFO("%s: f_embedding_scale = %f\n", __func__, hparams.f_embedding_scale);
8257-
LLAMA_LOG_INFO("%s: f_residual_scale = %f\n", __func__, hparams.f_residual_scale);
8258-
LLAMA_LOG_INFO("%s: f_attention_scale = %f\n", __func__, hparams.f_attention_scale);
8259-
LLAMA_LOG_INFO("%s: n_ff_shexp = %d\n", __func__, hparams.n_ff_shexp);
8260-
}
8250+
if (arch == LLM_ARCH_MINICPM ||
8251+
arch == LLM_ARCH_GRANITE ||
8252+
arch == LLM_ARCH_GRANITE_MOE ||
8253+
arch == LLM_ARCH_GRANITE_HYBRID ||
8254+
arch == LLM_ARCH_NEMOTRON_H_MOE) {
8255+
LLAMA_LOG_INFO("%s: f_embedding_scale = %f\n", __func__, hparams.f_embedding_scale);
8256+
LLAMA_LOG_INFO("%s: f_residual_scale = %f\n", __func__, hparams.f_residual_scale);
8257+
LLAMA_LOG_INFO("%s: f_attention_scale = %f\n", __func__, hparams.f_attention_scale);
8258+
LLAMA_LOG_INFO("%s: n_ff_shexp = %d\n", __func__, hparams.n_ff_shexp);
8259+
}
82618260

8262-
if (arch == LLM_ARCH_BAILINGMOE) {
8263-
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8264-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8265-
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8266-
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8267-
LLAMA_LOG_INFO("%s: expert_weights_norm = %d\n", __func__, hparams.expert_weights_norm);
8268-
}
8261+
if (arch == LLM_ARCH_BAILINGMOE) {
8262+
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8263+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8264+
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8265+
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8266+
LLAMA_LOG_INFO("%s: expert_weights_norm = %d\n", __func__, hparams.expert_weights_norm);
8267+
}
82698268

8270-
if (arch == LLM_ARCH_BAILINGMOE2) {
8271-
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8272-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8273-
LLAMA_LOG_INFO("%s: n_ff_shexp = %d\n", __func__, hparams.n_ff_shexp);
8274-
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8275-
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8276-
LLAMA_LOG_INFO("%s: expert_weights_norm = %d\n", __func__, hparams.expert_weights_norm);
8277-
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
8278-
LLAMA_LOG_INFO("%s: nextn_predict_layers = %d\n", __func__, hparams.nextn_predict_layers);
8279-
}
8269+
if (arch == LLM_ARCH_BAILINGMOE2) {
8270+
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
8271+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8272+
LLAMA_LOG_INFO("%s: n_ff_shexp = %d\n", __func__, hparams.n_ff_shexp);
8273+
LLAMA_LOG_INFO("%s: n_expert_shared = %d\n", __func__, hparams.n_expert_shared);
8274+
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
8275+
LLAMA_LOG_INFO("%s: expert_weights_norm = %d\n", __func__, hparams.expert_weights_norm);
8276+
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
8277+
LLAMA_LOG_INFO("%s: nextn_predict_layers = %d\n", __func__, hparams.nextn_predict_layers);
8278+
}
82808279

8281-
if (arch == LLM_ARCH_SMALLTHINKER || arch == LLM_ARCH_LFM2MOE) {
8282-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8283-
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
8284-
}
8280+
if (arch == LLM_ARCH_SMALLTHINKER || arch == LLM_ARCH_LFM2MOE) {
8281+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8282+
LLAMA_LOG_INFO("%s: expert_gating_func = %s\n", __func__, llama_expert_gating_func_name((llama_expert_gating_func_type) hparams.expert_gating_func));
8283+
}
82858284

8286-
if (arch == LLM_ARCH_GROVEMOE) {
8287-
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8288-
LLAMA_LOG_INFO("%s: n_ff_chexp = %d\n", __func__, hparams.n_ff_chexp);
8289-
LLAMA_LOG_INFO("%s: n_group_experts = %d\n", __func__, hparams.n_group_experts);
8290-
LLAMA_LOG_INFO("%s: expert_group_scale = %.2f\n", __func__, hparams.expert_group_scale);
8285+
if (arch == LLM_ARCH_GROVEMOE) {
8286+
LLAMA_LOG_INFO("%s: n_ff_exp = %d\n", __func__, hparams.n_ff_exp);
8287+
LLAMA_LOG_INFO("%s: n_ff_chexp = %d\n", __func__, hparams.n_ff_chexp);
8288+
LLAMA_LOG_INFO("%s: n_group_experts = %d\n", __func__, hparams.n_group_experts);
8289+
LLAMA_LOG_INFO("%s: expert_group_scale = %.2f\n", __func__, hparams.expert_group_scale);
8290+
}
82918291
}
82928292

82938293
vocab.print_info();

0 commit comments

Comments
 (0)