ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-27 00:24:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Move the Qwen-3.5 models to the standard attention mechanism (#1329)

Browse Source
This commit is contained in:
Kawrakow
2026-02-26 15:50:51 +01:00
committed by GitHub
parent 757bee6238
commit 62a7dcac5a
1 changed files with 47 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files

191
src/llama-build-context.cpp
View File

@@ -1827,9 +1827,9 @@ std::tuple<ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*> llm_build_con
cb(Kcur, "Kcur_normed", il);
ggml_build_forward_expand(gf, Kcur);
}
gate = ggml_sigmoid(ctx0, gate);
//gate = ggml_sigmoid(ctx0, gate);
//gate = ggml_reshape_2d(ctx0, gate, gate->ne[0]*gate->ne[1], gate->ne[2]);
cb(gate, "gate", il);
//cb(gate, "gate", il);
return {Qcur, Kcur, Vcur, gate};
}
@@ -4536,62 +4536,6 @@ ggml_cgraph * llm_build_context::build_qwen35moe() {
const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
int sections[4];
std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);
auto build_layer_attn = [&](ggml_tensor * cur, ggml_tensor * inp_pos, ggml_tensor * KQ_mask, int il) -> ggml_tensor * {
auto Qaux = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
auto Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
auto Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
cb(Qaux, "Qaux", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);
ggml_build_forward_expand(gf, Qaux);
ggml_build_forward_expand(gf, Kcur);
ggml_build_forward_expand(gf, Vcur);
Qaux = ggml_reshape_3d(ctx0, Qaux, n_embd_head * 2, n_head, n_tokens);
auto Qcur = ggml_cont(ctx0, ggml_view_3d(ctx0, Qaux, n_embd_head, n_head, n_tokens, Qaux->nb[1], Qaux->nb[2], 0));
auto gate = ggml_cont_2d(ctx0, ggml_view_3d(ctx0, Qaux, n_embd_head, n_head, n_tokens, Qaux->nb[1], Qaux->nb[2], n_embd_head*ggml_element_size(Qaux)), n_embd_head*n_head, n_tokens);
cb(Qcur, "Qcur", il);
cb(gate, "gate", il);
Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
Qcur = llm_build_norm(ctx0, Qcur, hparams, model.layers[il].attn_q_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(Qcur, "Qcur_normed", il);
Kcur = llm_build_norm(ctx0, Kcur, hparams, model.layers[il].attn_k_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(Kcur, "Kcur_normed", il);
Qcur = ggml_rope_multi(ctx0, Qcur, inp_pos, nullptr,
n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
Kcur = ggml_rope_multi(ctx0, Kcur, inp_pos, nullptr,
n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
cb(Qcur, "Qcur_roped", il);
cb(Kcur, "Kcur_roped", il);
ggml_tensor * attn = llm_build_kv(ctx0, lctx, kv_self, gf, nullptr, nullptr,
Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv,
hparams.f_attention_scale == 0.0f ? 1.0f / sqrtf(float(n_embd_head)) : hparams.f_attention_scale, cb, il);
cb(attn, "attn_pregate", il);
gate = ggml_sigmoid(ctx0, gate);
cb(gate, "gate_sigmoid", il);
attn = ggml_mul(ctx0, attn, gate);
cb(attn, "attn_gated", il);
attn = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, attn);
cb(attn, "attn_output", il);
return attn;
};
ggml_tensor * inpL = llm_build_inp_embd(ctx0, lctx, hparams, batch, model.tok_embd, cb);
ggml_tensor * inp_pos = build_inp_pos();
ggml_tensor * inp_out_ids = n_tokens > 1 ? build_inp_out_ids() : nullptr;
@@ -4601,6 +4545,8 @@ ggml_cgraph * llm_build_context::build_qwen35moe() {
cb(lctx.inp_s_seq_qnext, "inp_s_seq_qnext", -1);
ggml_set_input(lctx.inp_s_seq_qnext);
float KQ_scale = hparams.f_attention_scale == 0.0f ? 1.0f / sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
ggml_tensor * causal_mask = nullptr;
ggml_tensor * identity = nullptr;
ggml_tensor * diag_mask = nullptr;
@@ -4616,25 +4562,26 @@ ggml_cgraph * llm_build_context::build_qwen35moe() {
ggml_tensor * cur = nullptr;
for (int il = 0; il < n_layer; ++il) {
ggml_tensor * inpSA = inpL;
cur = llm_build_norm(ctx0, inpL, hparams, model.layers[il].attn_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(cur, "attn_norm", il);
if (hparams.is_recurrent(il)) {
ggml_tensor * inpSA = inpL;
cur = llm_build_norm(ctx0, inpL, hparams, model.layers[il].attn_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(cur, "attn_norm", il);
cur = delta.build_layer_attn_linear(ctx0, gf, cur, causal_mask, identity, diag_mask, il, cb);
if (il == n_layer - 1 && inp_out_ids) {
cur = ggml_get_rows(ctx0, cur, inp_out_ids);
inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
}
cur = ggml_add(ctx0, cur, inpSA);
cb(cur, "attn_residual", il);
} else {
cur = build_layer_attn(cur, inp_pos, KQ_mask, il);
cur = build_std_attention(gf, model.layers[il].attn_norm, inpL, inp_pos, il == n_layer - 1 ? inp_out_ids : nullptr, nullptr,
KQ_mask, nullptr, nullptr, KQ_scale, 0.0f, 0, il, true, false, true, false, true);
}
if (il == n_layer - 1 && inp_out_ids) {
cur = ggml_get_rows(ctx0, cur, inp_out_ids);
inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
}
cur = ggml_add(ctx0, cur, inpSA);
cb(cur, "attn_residual", il);
cur = llm_build_std_moe_ffn(ctx0, lctx, model.layers[il].ffn_norm, cur,
model.layers[il].ffn_gate_inp, nullptr,
model.layers[il].ffn_up_exps, nullptr,
@@ -4673,62 +4620,6 @@ ggml_cgraph * llm_build_context::build_qwen35() {
const int64_t n_embd_head = hparams.n_embd_head_v;
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
int sections[4];
std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);
auto build_layer_attn = [&](ggml_tensor * cur, ggml_tensor * inp_pos, ggml_tensor * KQ_mask, int il) -> ggml_tensor * {
auto Qaux = llm_build_lora_mm(lctx, ctx0, model.layers[il].wq, cur);
auto Kcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wk, cur);
auto Vcur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wv, cur);
cb(Qaux, "Qaux", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);
ggml_build_forward_expand(gf, Qaux);
ggml_build_forward_expand(gf, Kcur);
ggml_build_forward_expand(gf, Vcur);
Qaux = ggml_reshape_3d(ctx0, Qaux, n_embd_head * 2, n_head, n_tokens);
auto Qcur = ggml_cont(ctx0, ggml_view_3d(ctx0, Qaux, n_embd_head, n_head, n_tokens, Qaux->nb[1], Qaux->nb[2], 0));
auto gate = ggml_cont_2d(ctx0, ggml_view_3d(ctx0, Qaux, n_embd_head, n_head, n_tokens, Qaux->nb[1], Qaux->nb[2], n_embd_head*ggml_element_size(Qaux)), n_embd_head*n_head, n_tokens);
cb(Qcur, "Qcur", il);
cb(gate, "gate", il);
Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
Qcur = llm_build_norm(ctx0, Qcur, hparams, model.layers[il].attn_q_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(Qcur, "Qcur_normed", il);
Kcur = llm_build_norm(ctx0, Kcur, hparams, model.layers[il].attn_k_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(Kcur, "Kcur_normed", il);
Qcur = ggml_rope_multi(ctx0, Qcur, inp_pos, nullptr,
n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
Kcur = ggml_rope_multi(ctx0, Kcur, inp_pos, nullptr,
n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
cb(Qcur, "Qcur_roped", il);
cb(Kcur, "Kcur_roped", il);
ggml_tensor * attn = llm_build_kv(ctx0, lctx, kv_self, gf, nullptr, nullptr,
Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv,
hparams.f_attention_scale == 0.0f ? 1.0f / sqrtf(float(n_embd_head)) : hparams.f_attention_scale, cb, il);
cb(attn, "attn_pregate", il);
gate = ggml_sigmoid(ctx0, gate);
cb(gate, "gate_sigmoid", il);
attn = ggml_mul(ctx0, attn, gate);
cb(attn, "attn_gated", il);
attn = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, attn);
cb(attn, "attn_output", il);
return attn;
};
ggml_tensor * inpL = llm_build_inp_embd(ctx0, lctx, hparams, batch, model.tok_embd, cb);
ggml_tensor * inp_pos = build_inp_pos();
ggml_tensor * inp_out_ids = n_tokens > 1 ? build_inp_out_ids() : nullptr;
@@ -4738,6 +4629,8 @@ ggml_cgraph * llm_build_context::build_qwen35() {
cb(lctx.inp_s_seq_qnext, "inp_s_seq_qnext", -1);
ggml_set_input(lctx.inp_s_seq_qnext);
float KQ_scale = hparams.f_attention_scale == 0.0f ? 1.0f / sqrtf(float(n_embd_head)) : hparams.f_attention_scale;
ggml_tensor * causal_mask = nullptr;
ggml_tensor * identity = nullptr;
ggml_tensor * diag_mask = nullptr;
@@ -4753,25 +4646,23 @@ ggml_cgraph * llm_build_context::build_qwen35() {
ggml_tensor * cur = nullptr;
for (int il = 0; il < n_layer; ++il) {
ggml_tensor * inpSA = inpL;
cur = llm_build_norm(ctx0, inpL, hparams, model.layers[il].attn_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(cur, "attn_norm", il);
if (hparams.is_recurrent(il)) {
ggml_tensor * inpSA = inpL;
cur = llm_build_norm(ctx0, inpL, hparams, model.layers[il].attn_norm, nullptr, LLM_NORM_RMS, cb, il);
cb(cur, "attn_norm", il);
cur = delta.build_layer_attn_linear(ctx0, gf, cur, causal_mask, identity, diag_mask, il, cb);
if (il == n_layer - 1 && inp_out_ids) {
cur = ggml_get_rows(ctx0, cur, inp_out_ids);
inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
}
cur = ggml_add(ctx0, cur, inpSA);
cb(cur, "attn_residual", il);
} else {
cur = build_layer_attn(cur, inp_pos, KQ_mask, il);
cur = build_std_attention(gf, model.layers[il].attn_norm, inpL, inp_pos, il == n_layer - 1 ? inp_out_ids : nullptr, nullptr,
KQ_mask, nullptr, nullptr, KQ_scale, 0.0f, 0, il, true, false, true, false, true);
}
if (il == n_layer - 1 && inp_out_ids) {
cur = ggml_get_rows(ctx0, cur, inp_out_ids);
inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
}
cur = ggml_add(ctx0, cur, inpSA);
cb(cur, "attn_residual", il);
cur = llm_build_ffn(ctx0, lctx, model.layers[il].ffn_norm, cur,
model.layers[il].ffn_up, NULL, NULL,
model.layers[il].ffn_gate, NULL, NULL,
@@ -10254,7 +10145,7 @@ ggml_tensor * llm_build_context::build_std_attention(ggml_cgraph * gf, ggml_tens
auto the_k_norm = model.layers[il].attn_k_norm ? model.layers[il].attn_k_norm->extra ?
((ggml_split_tensor_t *)model.layers[il].attn_k_norm->extra)->splits[id] : model.layers[il].attn_k_norm : nullptr;
ggml_tensor *Qcur, *Kcur, *Vcur, *gate = nullptr;
if (model.arch == LLM_ARCH_QWEN3NEXT) {
if (model.arch == LLM_ARCH_QWEN3NEXT || model.arch == LLM_ARCH_QWEN35 || model.arch == LLM_ARCH_QWEN35MOE) {
auto [Q, K, V, G] = llm_build_mul_mat_qkv_gated(gf, cur, split_wq, split_wk, split_wv,
the_q_norm, the_k_norm, il);
Qcur = Q; Kcur = K; Vcur = V; gate = G;
@@ -10393,7 +10284,13 @@ ggml_tensor * llm_build_context::build_std_attention(ggml_cgraph * gf, ggml_tens
cur = ggml_reshape_2d(ctx0, cur, split_wo->ne[0], n_tokens);
cb(cur, "flash_attn_reshaped", il_cb);
if (gate) {
cur = ggml_mul(ctx0, cur, gate);
if (false && cur->ne[1] == 1) { // we need to add GGML_UNARY_OP_SIGMOID to the ops supported by ggml_fused_mul_unary
cur = ggml_fused_mul_unary(ctx0, cur, gate, GGML_UNARY_OP_SIGMOID);
} else {
gate = ggml_sigmoid(ctx0, gate);
cb(gate, "gate", il_cb);
cur = ggml_mul(ctx0, cur, gate);
}
cb(cur, "qkv_gated", il_cb);
}
@@ -10445,7 +10342,7 @@ ggml_tensor * llm_build_context::build_std_attention(ggml_cgraph * gf, ggml_tens
auto input_normed = cur;
ggml_tensor *Qcur, *Kcur, *Vcur, *gate = nullptr;
if (model.arch == LLM_ARCH_QWEN3NEXT) {
if (model.arch == LLM_ARCH_QWEN3NEXT || model.arch == LLM_ARCH_QWEN35 || model.arch == LLM_ARCH_QWEN35MOE) {
auto [Q, K, V, G] = llm_build_mul_mat_qkv_gated(gf, cur, model.layers[il].wq, model.layers[il].wk, model.layers[il].wv,
model.layers[il].attn_q_norm, model.layers[il].attn_k_norm, il);
Qcur = Q; Kcur = K; Vcur = V; gate = G;
@@ -10506,7 +10403,13 @@ ggml_tensor * llm_build_context::build_std_attention(ggml_cgraph * gf, ggml_tens
if (gate) {
cur = llm_build_kv(ctx0, lctx, kv_self, gf, nullptr, nullptr,
Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, KQ_scale, cb, il, sinks, n_swa);
cur = ggml_mul(ctx0, cur, gate);
if (false && cur->ne[1] == 1) { // we need to add GGML_UNARY_OP_SIGMOID to the ops supported by ggml_fused_mul_unary
cur = ggml_fused_mul_unary(ctx0, cur, gate, GGML_UNARY_OP_SIGMOID);
} else {
gate = ggml_sigmoid(ctx0, gate);
cb(gate, "gate", il);
cur = ggml_mul(ctx0, cur, gate);
}
cb(cur, "qkv_gated", il);
cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
if (model.layers[il].bo) {