Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 01:05:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '0fac87b157305eb82a70902327abffbbce25bd3e' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/workflows/build-android.yml
#	.github/workflows/hip-quality-check.yml
#	docs/multimodal.md
#	scripts/hip/gcn-cdna-vgpr-check.py
#	scripts/snapdragon/windows/run-bench.ps1
#	scripts/snapdragon/windows/run-cli.ps1
#	scripts/snapdragon/windows/run-tool.ps1
#	tests/test-backend-ops.cpp
#	tests/test-llama-archs.cpp
#	tools/imatrix/imatrix.cpp
#	tools/mtmd/CMakeLists.txt
This commit is contained in:
Concedo
2026-03-29 01:14:33 +08:00
parent 674b7f5eee 0fac87b157
commit aac220f7e3
32 changed files with 1371 additions and 109 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-arch.cpp
+4
View File
@@ -73,6 +73,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
{ LLM_ARCH_ARCTIC, "arctic" },
{ LLM_ARCH_DEEPSEEK, "deepseek" },
{ LLM_ARCH_DEEPSEEK2, "deepseek2" },
{ LLM_ARCH_DEEPSEEK2OCR, "deepseek2-ocr" },
{ LLM_ARCH_CHATGLM, "chatglm" },
{ LLM_ARCH_GLM4, "glm4" },
{ LLM_ARCH_GLM4_MOE, "glm4moe" },
@@ -1571,6 +1572,7 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
LLM_TENSOR_FFN_UP_SHEXP,
};
case LLM_ARCH_DEEPSEEK2:
case LLM_ARCH_DEEPSEEK2OCR:
case LLM_ARCH_MISTRAL4:
return {
LLM_TENSOR_TOKEN_EMBD,
@@ -1579,6 +1581,8 @@ static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
LLM_TENSOR_ATTN_NORM,
LLM_TENSOR_ATTN_Q_A_NORM,
LLM_TENSOR_ATTN_KV_A_NORM,
LLM_TENSOR_ATTN_K, // deepseek-ocr
LLM_TENSOR_ATTN_V, // deepseek-ocr
LLM_TENSOR_ATTN_Q,
LLM_TENSOR_ATTN_Q_A,
LLM_TENSOR_ATTN_Q_B,
src/llama-arch.h
+1
View File
@@ -77,6 +77,7 @@ enum llm_arch {
LLM_ARCH_ARCTIC,
LLM_ARCH_DEEPSEEK,
LLM_ARCH_DEEPSEEK2,
LLM_ARCH_DEEPSEEK2OCR,
LLM_ARCH_CHATGLM,
LLM_ARCH_GLM4,
LLM_ARCH_GLM4_MOE,
src/llama-chat.cpp
+6
View File
@@ -49,6 +49,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
{ "deepseek", LLM_CHAT_TEMPLATE_DEEPSEEK },
{ "deepseek2", LLM_CHAT_TEMPLATE_DEEPSEEK_2 },
{ "deepseek3", LLM_CHAT_TEMPLATE_DEEPSEEK_3 },
{ "deepseek-ocr", LLM_CHAT_TEMPLATE_DEEPSEEK_OCR },
{ "command-r", LLM_CHAT_TEMPLATE_COMMAND_R },
{ "llama3", LLM_CHAT_TEMPLATE_LLAMA_3 },
{ "chatglm3", LLM_CHAT_TEMPLATE_CHATGLM_3 },
@@ -548,6 +549,11 @@ int32_t llm_chat_apply_template(
if (add_ass) {
ss << LU8("<｜Assistant｜>");
}
} else if (tmpl == LLM_CHAT_TEMPLATE_DEEPSEEK_OCR) {
for (auto message : chat) {
// no template
ss << message->content;
}
} else if (tmpl == LLM_CHAT_TEMPLATE_EXAONE_3) {
// ref: https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct/discussions/8#66bae61b1893d14ee8ed85bb
// EXAONE-3.0-7.8B-Instruct
src/llama-chat.h
+1
View File
@@ -28,6 +28,7 @@ enum llm_chat_template {
LLM_CHAT_TEMPLATE_DEEPSEEK,
LLM_CHAT_TEMPLATE_DEEPSEEK_2,
LLM_CHAT_TEMPLATE_DEEPSEEK_3,
LLM_CHAT_TEMPLATE_DEEPSEEK_OCR,
LLM_CHAT_TEMPLATE_COMMAND_R,
LLM_CHAT_TEMPLATE_LLAMA_3,
LLM_CHAT_TEMPLATE_CHATGLM_3,
src/llama-graph.cpp
+1 -1
View File
@@ -1516,7 +1516,7 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
if (!weight_before_ffn) {
experts = ggml_mul(ctx0, experts, weights);
cb(cur, "ffn_moe_weighted", il);
cb(experts, "ffn_moe_weighted", il);
}
ggml_tensor * cur_experts[LLAMA_MAX_EXPERTS] = { nullptr };
src/llama-kv-cache.cpp
-1
View File
@@ -1566,7 +1566,6 @@ ggml_tensor * llama_kv_cache::build_rope_shift(
// ref: https://github.com/ggml-org/llama.cpp/pull/13870
? LLAMA_ROPE_TYPE_NEOX
: hparams.rope_type;
ggml_tensor * tmp;
if (ggml_is_quantized(cur->type)) {
src/llama-model.cpp
+79 -19
View File
@@ -484,6 +484,8 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_CONTEXT_LENGTH, hparams.n_ctx_train);
ml.get_key(LLM_KV_EMBEDDING_LENGTH, hparams.n_embd);
ml.get_key(LLM_KV_EMBEDDING_LENGTH_OUT, hparams.n_embd_out_impl, false);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
ml.get_key(LLM_KV_BLOCK_COUNT, hparams.n_layer);
ml.get_key(LLM_KV_EXPERT_COUNT, hparams.n_expert, false);
ml.get_key(LLM_KV_EXPERT_USED_COUNT, hparams.n_expert_used, false);
@@ -862,8 +864,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
case LLM_ARCH_BERT:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
switch (hparams.n_layer) {
case 3:
@@ -895,8 +895,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
}
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
switch (hparams.n_layer) {
case 12:
@@ -911,8 +909,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
case LLM_ARCH_JINA_BERT_V2:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
hparams.f_max_alibi_bias = 8.0f;
switch (hparams.n_layer) {
@@ -924,8 +920,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
case LLM_ARCH_JINA_BERT_V3:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
switch (hparams.n_layer) {
case 24:
@@ -937,8 +931,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
case LLM_ARCH_NOMIC_BERT_MOE:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
ml.get_key(LLM_KV_MOE_EVERY_N_LAYERS, hparams.moe_every_n_layers, 0);
if (hparams.n_layer == 12 && hparams.n_embd == 768) {
@@ -952,8 +944,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
case LLM_ARCH_NEO_BERT:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
if (hparams.n_layer == 28) {
type = LLM_TYPE_250M;
@@ -962,8 +952,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
case LLM_ARCH_EUROBERT:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
if (hparams.n_layer == 12) {
type = LLM_TYPE_SMALL; // 0.2B
@@ -1027,7 +1015,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
// fall through
case LLM_ARCH_QWEN2:
{
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) {
case 24: type = hparams.n_embd == 1024 ? LLM_TYPE_0_5B : LLM_TYPE_1B; break;
@@ -1109,7 +1096,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
} break;
case LLM_ARCH_QWEN3:
{
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
switch (hparams.n_layer) {
case 28: type = hparams.n_embd == 1024 ? LLM_TYPE_0_6B : LLM_TYPE_1_7B; break;
@@ -1401,7 +1387,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA, hparams.rope_freq_base_train_swa, false);
ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW, hparams.n_swa);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_POOLING_TYPE, hparams.pooling_type, false);
//applied only if model converted with --sentence-transformers-dense-modules
ml.get_key(LLM_KV_DENSE_2_FEAT_IN, hparams.dense_2_feat_in, false);
@@ -1750,6 +1735,26 @@ void llama_model::load_hparams(llama_model_loader & ml) {
default: type = LLM_TYPE_UNKNOWN;
}
} break;
case LLM_ARCH_DEEPSEEK2OCR:
{
// similar to deepseek2, but without MLA
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT, hparams.n_layer_dense_lead, false);
ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp);
ml.get_key(LLM_KV_EXPERT_SHARED_COUNT, hparams.n_expert_shared);
ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE, hparams.expert_weights_scale, false);
ml.get_key(LLM_KV_EXPERT_WEIGHTS_NORM, hparams.expert_weights_norm, false);
ml.get_key(LLM_KV_EXPERT_GATING_FUNC, hparams.expert_gating_func, false);
if (hparams.expert_gating_func == LLAMA_EXPERT_GATING_FUNC_TYPE_NONE) {
hparams.expert_gating_func = LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX;
}
switch (hparams.n_layer) {
case 12: type = LLM_TYPE_3B; break;
default: type = LLM_TYPE_UNKNOWN;
}
} break;
case LLM_ARCH_PLM:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
@@ -2198,7 +2203,6 @@ void llama_model::load_hparams(llama_model_loader & ml) {
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_GROUPNORM_EPS, hparams.f_norm_group_eps);
ml.get_key(LLM_KV_ATTENTION_GROUPNORM_GROUPS, hparams.n_norm_groups);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn, false);
} break;
case LLM_ARCH_BAILINGMOE:
{
@@ -5125,6 +5129,60 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0);
create_tensor_gate_up_exps(layer, i, n_embd, n_ff_exp, n_expert, 0);
// Shared expert branch
layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), { n_ff_exp * n_expert_shared, n_embd}, 0);
layer.ffn_up_shexp = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
}
}
} break;
case LLM_ARCH_DEEPSEEK2OCR:
{
// similar to deepseek2, but without MLA
const int64_t n_ff_exp = hparams.n_ff_exp;
const int64_t n_expert_shared = hparams.n_expert_shared;
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
// output
output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
// try to load output.weight, if not found, use token_embd (tied embeddings)
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
if (!output) {
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
}
for (int i = 0; i < n_layer; ++i) {
auto & layer = layers[i];
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0);
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd}, 0);
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd}, 0);
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
// norm
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
if (i < (int) hparams.n_layer_dense_lead) {
layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0);
layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);
layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
} else {
layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
layer.ffn_exp_probs_b = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, TENSOR_NOT_REQUIRED);
if (n_expert == 0) {
throw std::runtime_error("n_expert must be > 0");
}
if (n_expert_used == 0) {
throw std::runtime_error("n_expert_used must be > 0");
}
// MoE branch
layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp, n_embd, n_expert}, 0);
create_tensor_gate_up_exps(layer, i, n_embd, n_ff_exp, n_expert, 0);
// Shared expert branch
layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, n_ff_exp * n_expert_shared}, 0);
layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), { n_ff_exp * n_expert_shared, n_embd}, 0);
@@ -8016,7 +8074,7 @@ void llama_model::print_info() const {
LLAMA_LOG_INFO("%s: expert_weights_scale = %.1f\n", __func__, hparams.expert_weights_scale);
}
if (arch == LLM_ARCH_DEEPSEEK2 || arch == LLM_ARCH_GLM_DSA || arch == LLM_ARCH_MISTRAL4) {
if (arch == LLM_ARCH_DEEPSEEK2 || arch == LLM_ARCH_DEEPSEEK2OCR || arch == LLM_ARCH_GLM_DSA || arch == LLM_ARCH_MISTRAL4) {
LLAMA_LOG_INFO("%s: n_layer_dense_lead = %d\n", __func__, hparams.n_layer_dense_lead);
LLAMA_LOG_INFO("%s: n_lora_q = %d\n", __func__, hparams.n_lora_q);
LLAMA_LOG_INFO("%s: n_lora_kv = %d\n", __func__, hparams.n_lora_kv);
@@ -8593,6 +8651,7 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
llm = std::make_unique<llm_build_deepseek>(*this, params);
} break;
case LLM_ARCH_DEEPSEEK2:
case LLM_ARCH_DEEPSEEK2OCR:
case LLM_ARCH_GLM_DSA:
case LLM_ARCH_MISTRAL4:
{
@@ -8993,6 +9052,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
case LLM_ARCH_ARCTIC:
case LLM_ARCH_DEEPSEEK:
case LLM_ARCH_DEEPSEEK2:
case LLM_ARCH_DEEPSEEK2OCR:
case LLM_ARCH_PLM:
case LLM_ARCH_CHATGLM:
case LLM_ARCH_GRANITE:
src/llama-quant.cpp
+4 -1
View File
@@ -344,7 +344,10 @@ static bool tensor_allows_quantization(const llama_model_quantize_params * param
quantize &= name.find("attn_rel_b.weight") == std::string::npos;
// do not quantize specific multimodal tensors
quantize &= name.find(".position_embd.") == std::string::npos;
quantize &= name.find(".position_embd") == std::string::npos;
quantize &= name.find("sam.patch_embd") == std::string::npos;
quantize &= name.find("sam.pos_embd") == std::string::npos;
quantize &= name.find(".rel_pos") == std::string::npos;
return quantize;
}
src/llama-vocab.cpp
+3 -1
View File
@@ -2188,7 +2188,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
} else if (
tokenizer_pre == "qwen2" ||
tokenizer_pre == "deepseek-r1-qwen" ||
tokenizer_pre == "kormo") {
tokenizer_pre == "kormo" ||
tokenizer_pre == "f2llmv2") {
pre_type = LLAMA_VOCAB_PRE_TYPE_QWEN2;
clean_spaces = false;
} else if (
@@ -2728,6 +2729,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
|| t.first == "[EOS]" // Kimi-K2
|| t.first == "<|end_of_text|>"
|| t.first == "<end_of_utterance>" // smoldocling
|| t.first == "<｜end▁of▁sentence｜>" // deepseek-ocr
) {
special_eog_ids.insert(t.second);
if ((attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
src/models/deepseek2.cpp
+35 -1
View File
@@ -2,6 +2,9 @@
llm_build_deepseek2::llm_build_deepseek2(const llama_model & model, const llm_graph_params & params) :
llm_graph_context(params) {
// lite variants include DeepSeek-V2-Lite, GigaChat3-10B-A1.8B
bool is_ocr = model.arch == LLM_ARCH_DEEPSEEK2OCR;
const bool is_mla = hparams.is_mla();
// note: these are the actual head sizes you get when treating as MHA or after "decompression" using wv_b for MLA
@@ -54,7 +57,38 @@ llm_build_deepseek2::llm_build_deepseek2(const llama_model & model, const llm_gr
cb(cur, "attn_norm", il);
// self_attention
{
if (is_ocr) {
const int n_embed_head = hparams.n_embd / hparams.n_head();
const int ocr_rope_type = GGML_ROPE_TYPE_NEOX;
GGML_ASSERT(n_embed_head == n_embd_head_k && n_embed_head == n_embd_head_v);
ggml_tensor * Qcur = NULL;
ggml_tensor * Kcur = NULL;
ggml_tensor * Vcur = NULL;
Qcur = ggml_mul_mat(ctx0, model.layers[il].wq, cur);
Kcur = ggml_mul_mat(ctx0, model.layers[il].wk, cur);
Vcur = ggml_mul_mat(ctx0, model.layers[il].wv, cur);
cb(Qcur, "q", il);
cb(Kcur, "k", il);
cb(Vcur, "v", il);
Qcur = ggml_reshape_3d(ctx0, Qcur, n_embed_head, n_head, n_tokens);
Kcur = ggml_reshape_3d(ctx0, Kcur, n_embed_head, n_head, n_tokens);
Vcur = ggml_reshape_3d(ctx0, Vcur, n_embed_head, n_head, n_tokens);
GGML_ASSERT(fabs(freq_base - 10000.0) < 1e-4);
Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_embed_head, ocr_rope_type, 0, freq_base, 1, 0, 1, 0, 0);
Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_embed_head, ocr_rope_type, 0, freq_base, 1, 0, 1, 0, 0);
cb(Qcur, "q_pe", il);
cb(Kcur, "k_pe", il);
cur = build_attn(inp_attn_kv,
model.layers[il].wo, NULL,
Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
cb(cur, "attn_out", il);
}
else {
ggml_tensor * q = NULL;
const bool is_lite = model.layers[il].wq;