LLM inference in C/C++
C++ ggml
models: clean up some dead switch branches in old models (#28669)
Iggy Jacksond344123fe2de
diff --git a/src/models/bert.cpp b/src/models/bert.cppindex ca0281d306..9cc03dc56a 100644--- a/src/models/bert.cpp+++ b/src/models/bert.cpp@@ -29,15 +29,13 @@ void llama_model_bert::load_arch_tensors(llama_model_loader &) {tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_token_types}, TENSOR_NOT_REQUIRED);- if (arch == LLM_ARCH_BERT) {- pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0);+ pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0);- cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);- cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {n_embd}, TENSOR_NOT_REQUIRED);+ cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);+ cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {n_embd}, TENSOR_NOT_REQUIRED);- cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- }+ cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);+ cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight", 0), {n_embd}, 0);tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias", 0), {n_embd}, 0);@@ -53,20 +51,10 @@ void llama_model_bert::load_arch_tensors(llama_model_loader &) {layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i), {n_embd}, 0);- if (hparams.moe_every_n_layers > 0 && i % hparams.moe_every_n_layers == 1) {- layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff, n_expert}, 0);- layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0);- layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);- } else {- layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);- layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);- layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);- layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);-- if (arch == LLM_ARCH_NOMIC_BERT) {- layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);- }- }+ layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);+ layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);+ layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);+ layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i), {n_embd}, 0);diff --git a/src/models/jina-bert-v3.cpp b/src/models/jina-bert-v3.cppindex 1c974a6f16..78cf9d8359 100644--- a/src/models/jina-bert-v3.cpp+++ b/src/models/jina-bert-v3.cpp@@ -19,16 +19,6 @@ void llama_model_jina_bert_v3::load_arch_tensors(llama_model_loader &) {tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_token_types}, TENSOR_NOT_REQUIRED);- if (arch == LLM_ARCH_BERT) {- pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0);-- cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);- cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {n_embd}, TENSOR_NOT_REQUIRED);-- cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- }-tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight", 0), {n_embd}, 0);tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias", 0), {n_embd}, 0);@@ -43,20 +33,10 @@ void llama_model_jina_bert_v3::load_arch_tensors(llama_model_loader &) {layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i), {n_embd}, 0);- if (hparams.moe_every_n_layers > 0 && i % hparams.moe_every_n_layers == 1) {- layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff, n_expert}, 0);- layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0);- layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);- } else {- layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);- layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);- layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);- layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);-- if (arch == LLM_ARCH_NOMIC_BERT) {- layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);- }- }+ layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);+ layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);+ layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);+ layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i), {n_embd}, 0);diff --git a/src/models/nomic-bert-moe.cpp b/src/models/nomic-bert-moe.cppindex da4b62919b..924af5e0bc 100644--- a/src/models/nomic-bert-moe.cpp+++ b/src/models/nomic-bert-moe.cpp@@ -4,12 +4,10 @@ void llama_model_nomic_bert_moe::load_arch_hparams(llama_model_loader & ml) {ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);ml.get_key(LLM_KV_MOE_EVERY_N_LAYERS, hparams.moe_every_n_layers, 0);- if (hparams.n_layer() == 12 && hparams.n_embd == 768) {- if (arch == LLM_ARCH_NOMIC_BERT) {- type = LLM_TYPE_137M;- } else if (arch == LLM_ARCH_NOMIC_BERT_MOE && hparams.moe_every_n_layers == 2) {- type = LLM_TYPE_475M;- }+ switch (hparams.n_layer()) {+ case 12:+ type = LLM_TYPE_475M; break;+ default: type = LLM_TYPE_UNKNOWN;}}@@ -22,16 +20,6 @@ void llama_model_nomic_bert_moe::load_arch_tensors(llama_model_loader &) {tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_token_types}, TENSOR_NOT_REQUIRED);- if (arch == LLM_ARCH_BERT) {- pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0);-- cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);- cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {n_embd}, TENSOR_NOT_REQUIRED);-- cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- }-tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight", 0), {n_embd}, 0);tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias", 0), {n_embd}, 0);@@ -55,10 +43,6 @@ void llama_model_nomic_bert_moe::load_arch_tensors(llama_model_loader &) {layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);-- if (arch == LLM_ARCH_NOMIC_BERT) {- layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);- }}layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);diff --git a/src/models/nomic-bert.cpp b/src/models/nomic-bert.cppindex e7fc72286a..509787286e 100644--- a/src/models/nomic-bert.cpp+++ b/src/models/nomic-bert.cpp@@ -1,15 +1,12 @@#include "models.h"void llama_model_nomic_bert::load_arch_hparams(llama_model_loader & ml) {- ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);- ml.get_key(LLM_KV_MOE_EVERY_N_LAYERS, hparams.moe_every_n_layers, 0);+ ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);- if (hparams.n_layer() == 12 && hparams.n_embd == 768) {- if (arch == LLM_ARCH_NOMIC_BERT) {- type = LLM_TYPE_137M;- } else if (arch == LLM_ARCH_NOMIC_BERT_MOE && hparams.moe_every_n_layers == 2) {- type = LLM_TYPE_475M;- }+ switch (hparams.n_layer()) {+ case 12:+ type = LLM_TYPE_137M; break;+ default: type = LLM_TYPE_UNKNOWN;}}@@ -22,16 +19,6 @@ void llama_model_nomic_bert::load_arch_tensors(llama_model_loader &) {tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);type_embd = create_tensor(tn(LLM_TENSOR_TOKEN_TYPES, "weight"), {n_embd, n_token_types}, TENSOR_NOT_REQUIRED);- if (arch == LLM_ARCH_BERT) {- pos_embd = create_tensor(tn(LLM_TENSOR_POS_EMBD, "weight"), {n_embd, n_ctx_train}, 0);-- cls = create_tensor(tn(LLM_TENSOR_CLS, "weight"), {n_embd, n_embd}, TENSOR_NOT_REQUIRED);- cls_b = create_tensor(tn(LLM_TENSOR_CLS, "bias"), {n_embd}, TENSOR_NOT_REQUIRED);-- cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), {n_embd, hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- cls_out_b = create_tensor(tn(LLM_TENSOR_CLS_OUT, "bias"), {hparams.n_cls_out}, TENSOR_NOT_REQUIRED);- }-tok_norm = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "weight", 0), {n_embd}, 0);tok_norm_b = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD_NORM, "bias", 0), {n_embd}, 0);@@ -46,20 +33,12 @@ void llama_model_nomic_bert::load_arch_tensors(llama_model_loader &) {layer.attn_out_norm = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "weight", i), {n_embd}, 0);layer.attn_out_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_OUT_NORM, "bias", i), {n_embd}, 0);- if (hparams.moe_every_n_layers > 0 && i % hparams.moe_every_n_layers == 1) {- layer.ffn_up_exps = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", i), { n_embd, n_ff, n_expert}, 0);- layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0);- layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);- } else {- layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);- layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);- layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);- layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);+ layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);+ layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);+ layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, 0);+ layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);- if (arch == LLM_ARCH_NOMIC_BERT) {- layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);- }- }+ layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);layer.layer_out_norm = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "weight", i), {n_embd}, 0);layer.layer_out_norm_b = create_tensor(tn(LLM_TENSOR_LAYER_OUT_NORM, "bias", i), {n_embd}, 0);