Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Optimized DeepSeek V2/V3 implementation (MLA) #11446

Draft
wants to merge 10 commits into
base: master
Choose a base branch
from
+203 −30
Draft

Optimized DeepSeek V2/V3 implementation (MLA) #11446

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
93864cd
llama : experimental DeepSeek2 MLA implementation that caches latent …
sszymczy Jan 22, 2025
f0ce53f
Merge remote-tracking branch 'origin/master' into deepseek2-mla-exp
sszymczy Jan 24, 2025
de538aa
llama : optimize DeepSeek MLA implementation
sszymczy Jan 25, 2025
ce73063
llama : Update tensor names in DeepSeek2 MLA implementation.
sszymczy Jan 26, 2025
202f323
llama : add a second copy of c^KV cache in DeepSeek2 MLA to avoid tra…
sszymczy Jan 26, 2025
93c5937
llama : modified tensor permutations to multiply larger matrices duri…
sszymczy Jan 26, 2025
1eee98f
llama : removed unnecessary code in DeepSeek V2 implementation
sszymczy Jan 27, 2025
8ff0991
convert : make lint happy
sszymczy Jan 28, 2025
8a887de
llama : prompt processing optimizations in DeepSeek V2
sszymczy Jan 28, 2025
7654331
llama : avoid ggml_cont() is possible in DeepSeek V2 implementation
sszymczy Jan 30, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 22 additions & 0 deletions convert_hf_to_gguf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4136,6 +4136,28 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
else:
return []

if name.endswith("kv_b_proj.weight"):
name_kb = name.replace("kv_b_proj", "k_b_proj")
name_vb = name.replace("kv_b_proj", "v_b_proj")

n_head_kv = self.hparams["num_key_value_heads"]
v_head_dim = self.hparams["v_head_dim"]
qk_nope_head_dim = self.hparams["qk_nope_head_dim"]

assert data_torch.shape[0] == n_head_kv * (v_head_dim + qk_nope_head_dim)

kv_b = data_torch.view(n_head_kv, v_head_dim + qk_nope_head_dim, data_torch.shape[-1])
k_b, v_b = torch.split(kv_b, [qk_nope_head_dim, v_head_dim], dim=1)
k_b = k_b.transpose(1, 2)
k_b = k_b.reshape(n_head_kv * data_torch.shape[-1], qk_nope_head_dim)
v_b = v_b.reshape(n_head_kv * v_head_dim, data_torch.shape[-1])

return [
(self.map_tensor_name(name), data_torch),
(self.map_tensor_name(name_kb), k_b),
(self.map_tensor_name(name_vb), v_b)
]

return [(self.map_tensor_name(name), data_torch)]

def prepare_tensors(self):
Expand Down
6 changes: 6 additions & 0 deletions gguf-py/gguf/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -356,6 +356,8 @@ class MODEL_TENSOR(IntEnum):
ATTN_Q_B = auto()
ATTN_KV_A_MQA = auto()
ATTN_KV_B = auto()
ATTN_K_B = auto()
ATTN_V_B = auto()
ATTN_Q_A_NORM = auto()
ATTN_KV_A_NORM = auto()
FFN_SUB_NORM = auto()
Expand Down Expand Up @@ -543,6 +545,8 @@ class MODEL_TENSOR(IntEnum):
MODEL_TENSOR.ATTN_Q_B: "blk.{bid}.attn_q_b",
MODEL_TENSOR.ATTN_KV_A_MQA: "blk.{bid}.attn_kv_a_mqa",
MODEL_TENSOR.ATTN_KV_B: "blk.{bid}.attn_kv_b",
MODEL_TENSOR.ATTN_K_B: "blk.{bid}.attn_k_b",
MODEL_TENSOR.ATTN_V_B: "blk.{bid}.attn_v_b",
MODEL_TENSOR.ATTN_Q_A_NORM: "blk.{bid}.attn_q_a_norm",
MODEL_TENSOR.ATTN_KV_A_NORM: "blk.{bid}.attn_kv_a_norm",
MODEL_TENSOR.ATTN_SUB_NORM: "blk.{bid}.attn_sub_norm",
Expand Down Expand Up @@ -1333,6 +1337,8 @@ class MODEL_TENSOR(IntEnum):
MODEL_TENSOR.ATTN_Q_B,
MODEL_TENSOR.ATTN_KV_A_MQA,
MODEL_TENSOR.ATTN_KV_B,
MODEL_TENSOR.ATTN_K_B,
MODEL_TENSOR.ATTN_V_B,
MODEL_TENSOR.ATTN_Q_A_NORM,
MODEL_TENSOR.ATTN_KV_A_NORM,
MODEL_TENSOR.ATTN_OUT,
Expand Down
8 changes: 8 additions & 0 deletions gguf-py/gguf/tensor_mapping.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -586,6 +586,14 @@ class TensorNameMap:
"model.layers.{bid}.self_attn.kv_b_proj", # deepseek2
),

MODEL_TENSOR.ATTN_K_B: (
"model.layers.{bid}.self_attn.k_b_proj", # deepseek2
),

MODEL_TENSOR.ATTN_V_B: (
"model.layers.{bid}.self_attn.v_b_proj", # deepseek2
),

MODEL_TENSOR.ATTN_Q_A_NORM: (
"model.layers.{bid}.self_attn.q_a_layernorm", # deepseek2
),
Expand Down
6 changes: 6 additions & 0 deletions src/llama-arch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -999,6 +999,8 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
{ LLM_TENSOR_ATTN_Q_B, "blk.%d.attn_q_b" },
{ LLM_TENSOR_ATTN_KV_A_MQA, "blk.%d.attn_kv_a_mqa" },
{ LLM_TENSOR_ATTN_KV_B, "blk.%d.attn_kv_b" },
{ LLM_TENSOR_ATTN_K_B, "blk.%d.attn_k_b" },
{ LLM_TENSOR_ATTN_V_B, "blk.%d.attn_v_b" },
{ LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" },
{ LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" },
{ LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" },
Expand Down Expand Up @@ -1330,6 +1332,8 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
{LLM_TENSOR_ATTN_Q_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_KV_A_MQA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_KV_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_K_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_V_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_DEC_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_DEC_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
Expand All @@ -1347,6 +1351,8 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
{LLM_TENSOR_ATTN_Q_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_KV_A_MQA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_KV_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_K_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_ATTN_V_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_DEC_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_DEC_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
{LLM_TENSOR_DEC_ATTN_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
Expand Down
2 changes: 2 additions & 0 deletions src/llama-arch.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -277,6 +277,8 @@ enum llm_tensor {
LLM_TENSOR_ATTN_Q_B,
LLM_TENSOR_ATTN_KV_A_MQA,
LLM_TENSOR_ATTN_KV_B,
LLM_TENSOR_ATTN_K_B,
LLM_TENSOR_ATTN_V_B,
LLM_TENSOR_ATTN_Q_A_NORM,
LLM_TENSOR_ATTN_KV_A_NORM,
LLM_TENSOR_ATTN_SUB_NORM,
Expand Down
21 changes: 20 additions & 1 deletion src/llama-kv-cache.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,7 @@ bool llama_kv_cache_init(
auto it = ctx_map.find(buft);
if (it == ctx_map.end()) {
struct ggml_init_params params = {
/*.mem_size =*/ size_t(2u*n_layer*ggml_tensor_overhead()),
/*.mem_size =*/ size_t(5u*n_layer*ggml_tensor_overhead()),
/*.mem_buffer =*/ NULL,
/*.no_alloc =*/ true,
};
Expand All @@ -71,6 +71,11 @@ bool llama_kv_cache_init(
cache.k_l.reserve(n_layer);
cache.v_l.reserve(n_layer);

// DeepSeek MLA
cache.kr_l.reserve(n_layer);
cache.kv_l.reserve(n_layer);
cache.kvt_l.reserve(n_layer);

for (int i = 0; i < n_layer; i++) {
const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa(i) + hparams.n_embd_k_s();
const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(i) + hparams.n_embd_v_s();
Expand All @@ -97,6 +102,20 @@ bool llama_kv_cache_init(
ggml_format_name(v, "cache_v_l%d", i);
cache.k_l.push_back(k);
cache.v_l.push_back(v);

// DeepSeek MLA
const uint32_t n_embd_head_qk_rope = hparams.n_rot;
const uint32_t kv_lora_rank = hparams.n_lora_kv;
LLAMA_LOG_DEBUG("%s: layer %d: n_embd_head_qk_rope = %d, kv_lora_rank = %d\n", __func__, i, n_embd_head_qk_rope, kv_lora_rank);
ggml_tensor * kr = ggml_new_tensor_1d(ctx, cache.type_kr, n_embd_head_qk_rope*kv_size);
ggml_tensor * kv = ggml_new_tensor_1d(ctx, cache.type_kv, kv_lora_rank*kv_size);
ggml_tensor * kvt = ggml_new_tensor_1d(ctx, cache.type_kv, kv_lora_rank*kv_size);
ggml_format_name(kr, "cache_kr_l%d", i);
ggml_format_name(kv, "cache_kv_l%d", i);
ggml_format_name(kvt, "cache_kvt_l%d", i);
cache.kr_l.push_back(kr);
cache.kv_l.push_back(kv);
cache.kvt_l.push_back(kvt);
}

// allocate tensors and initialize the buffers to avoid NaNs in the padding
Expand Down
8 changes: 8 additions & 0 deletions src/llama-kv-cache.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,11 +49,19 @@ struct llama_kv_cache {
ggml_type type_k = GGML_TYPE_F16;
ggml_type type_v = GGML_TYPE_F16;

ggml_type type_kr = GGML_TYPE_F16;
ggml_type type_kv = GGML_TYPE_F16;

std::vector<llama_kv_cell> cells;

std::vector<struct ggml_tensor *> k_l; // per layer
std::vector<struct ggml_tensor *> v_l;

// DeepSeek MLA
std::vector<struct ggml_tensor *> kr_l; // per layer
std::vector<struct ggml_tensor *> kv_l;
std::vector<struct ggml_tensor *> kvt_l;

std::vector<ggml_context_ptr> ctxs;
std::vector<ggml_backend_buffer_ptr> bufs;

Expand Down
2 changes: 2 additions & 0 deletions src/llama-model.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2870,6 +2870,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {

layer.wkv_a_mqa = create_tensor(tn(LLM_TENSOR_ATTN_KV_A_MQA, "weight", i), {n_embd, kv_lora_rank + (n_embd_head_qk_rope)}, 0);
layer.wkv_b = create_tensor(tn(LLM_TENSOR_ATTN_KV_B, "weight", i), {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)}, 0);
layer.wk_b = create_tensor(tn(LLM_TENSOR_ATTN_K_B, "weight", i), {n_embd_head_qk_nope, n_head * kv_lora_rank}, 0);
layer.wv_b = create_tensor(tn(LLM_TENSOR_ATTN_V_B, "weight", i), {kv_lora_rank, n_head * n_embd_head_v}, 0);
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_head * ( n_embd_head_v), n_embd}, 0);

layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
Expand Down
2 changes: 2 additions & 0 deletions src/llama-model.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -161,6 +161,8 @@ struct llama_layer {
struct ggml_tensor * wq_b = nullptr;
struct ggml_tensor * wkv_a_mqa = nullptr;
struct ggml_tensor * wkv_b = nullptr;
struct ggml_tensor * wk_b = nullptr;
struct ggml_tensor * wv_b = nullptr;
struct ggml_tensor * wq_cross = nullptr;
struct ggml_tensor * wk_cross = nullptr;
struct ggml_tensor * wv_cross = nullptr;
Expand Down
156 changes: 127 additions & 29 deletions src/llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6403,6 +6403,10 @@ struct llm_build_context {
// KQ_mask (mask for 1 head, it will be broadcasted to all heads)
struct ggml_tensor * KQ_mask = build_inp_KQ_mask();

// whether to use n_tokens as the matrix dimension during multiplication or n_head
// n_tokens is higher during prompt processing, this allows to optimize for this case
bool pp_opt = n_tokens > n_head;

Comment on lines +6406 to +6409
Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not really sure this is the right approach. Haven't followed through the logic yet, but it seems strange to involve so many permutes and conts.

I would first look into improving the FA kernels to support DeepSeek head sizes.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not really sure this is the right approach. Haven't followed through the logic yet, but it seems strange to involve so many permutes and conts.

Hmm? I'm quite sure there's only one ggml_cont() call (excluding the ones for CUDA compatibility that already existed in the previous implementation).

As for the permutes the idea is to multiply by a matrix with a second dimension equal to the number of heads instead of the number of tokens (which is 1) during a single sequence token generation, that increased the performance on a CPU a bit.

So during prompt processing we have 2 permutes and 1 cont. During token generation we have 5 permutes (yeah, that may be a lot) and 0 conts.

Copy link
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the correction - I did imagine the extra conts when I saw the permutes.

for (int il = 0; il < n_layer; ++il) {
struct ggml_tensor * inpSA = inpL;

Expand Down Expand Up @@ -6470,33 +6474,33 @@ struct llm_build_context {
LLM_NORM_RMS, cb, il);
cb(kv_compressed, "kv_compressed", il);

// {kv_lora_rank, n_head * (n_embd_head_qk_nope + n_embd_head_v)} * {kv_lora_rank, n_tokens} -> {n_head * (n_embd_head_qk_nope + n_embd_head_v), n_tokens}
struct ggml_tensor * kv = ggml_mul_mat(ctx0, model.layers[il].wkv_b, kv_compressed);
cb(kv, "kv", il);
struct ggml_tensor * kv_cache_view = ggml_view_1d(ctx0, kv_self.kv_l[il], n_tokens*kv_lora_rank, ggml_row_size(kv_self.kv_l[il]->type, kv_lora_rank)*kv_head);
cb(kv_cache_view, "kv_cache_view", il);

// split into {n_head * n_embd_head_qk_nope, n_tokens}
struct ggml_tensor * k_nope = ggml_view_3d(ctx0, kv, n_embd_head_qk_nope, n_head, n_tokens,
ggml_row_size(kv->type, n_embd_head_qk_nope + hparams.n_embd_head_v),
ggml_row_size(kv->type, n_head * (n_embd_head_qk_nope + hparams.n_embd_head_v)),
0);
cb(k_nope, "k_nope", il);
// note: storing c^KV in the KV cache
ggml_build_forward_expand(gf, ggml_cpy(ctx0, kv_compressed, kv_cache_view));

// and {n_head * n_embd_head_v, n_tokens}
struct ggml_tensor * v_states = ggml_view_3d(ctx0, kv, hparams.n_embd_head_v, n_head, n_tokens,
ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)),
ggml_row_size(kv->type, (n_embd_head_qk_nope + hparams.n_embd_head_v)*n_head),
ggml_row_size(kv->type, (n_embd_head_qk_nope)));
cb(v_states, "v_states", il);
struct ggml_tensor * kv_cache_trans_view = ggml_view_2d(ctx0, kv_self.kvt_l[il], n_tokens, kv_lora_rank, ggml_row_size(kv_self.kv_l[il]->type, kv_self.size), ggml_row_size(kv_self.kv_l[il]->type, kv_head));
cb(kv_cache_trans_view, "kv_cache_trans_view", il);

v_states = ggml_cont(ctx0, v_states);
cb(v_states, "v_states", il);
// note: storing transposed c^KV in the transposed KV cache
ggml_build_forward_expand(gf, ggml_cpy(ctx0, ggml_transpose(ctx0, kv_compressed), kv_cache_trans_view));

v_states = ggml_view_2d(ctx0, v_states, hparams.n_embd_head_v * n_head, n_tokens,
ggml_row_size(kv->type, hparams.n_embd_head_v * n_head),
0);
cb(v_states, "v_states", il);
struct ggml_tensor * kv_cache =
ggml_view_2d(ctx0, kv_self.kv_l[il],
kv_lora_rank, n_kv,
ggml_row_size(kv_self.kv_l[il]->type, kv_lora_rank),
0);
cb(kv_cache, "kv_cache", il);

q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing this
struct ggml_tensor * kv_cache_trans =
ggml_view_2d(ctx0, kv_self.kvt_l[il],
n_kv, kv_lora_rank,
ggml_row_size(kv_self.kv_l[il]->type, kv_self.size),
0);
cb(kv_cache_trans, "kv_cache_trans", il);

q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
q_pe = ggml_rope_ext(
ctx0, q_pe, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
Expand All @@ -6513,15 +6517,91 @@ struct llm_build_context {
);
cb(k_pe, "k_pe", il);

struct ggml_tensor * q_states = ggml_concat(ctx0, q_nope, q_pe, 0);
cb(q_states, "q_states", il);
struct ggml_tensor * kr_cache_view = ggml_view_1d(ctx0, kv_self.kr_l[il], n_tokens*n_embd_head_qk_rope, ggml_row_size(kv_self.kr_l[il]->type, n_embd_head_qk_rope)*kv_head);
cb(kr_cache_view, "kr_cache_view", il);

struct ggml_tensor * k_states = ggml_concat(ctx0, k_nope, ggml_repeat(ctx0, k_pe, q_pe), 0);
cb(k_states, "k_states", il);
// note: storing RoPE-ed version of K^R in the KV cache
ggml_build_forward_expand(gf, ggml_cpy(ctx0, k_pe, kr_cache_view));

cur = llm_build_kv(ctx0, lctx, kv_self, gf,
model.layers[il].wo, NULL,
k_states, v_states, q_states, KQ_mask, n_tokens, kv_head, n_kv, kq_scale, cb, il);
struct ggml_tensor * kr_cache =
ggml_view_2d(ctx0, kv_self.kr_l[il],
n_embd_head_qk_rope, n_kv,
ggml_row_size(kv_self.kr_l[il]->type, n_embd_head_qk_rope),
0);
cb(kr_cache, "kr_cache", il);

struct ggml_tensor * wk_b = ggml_view_3d(ctx0, model.layers[il].wk_b, n_embd_head_qk_nope, kv_lora_rank, n_head, ggml_row_size(model.layers[il].wk_b->type, n_embd_head_qk_nope), ggml_row_size(model.layers[il].wk_b->type, kv_lora_rank * n_embd_head_qk_nope), 0);
cb(wk_b, "wk_b", il);

q_nope = ggml_permute(ctx0, q_nope, 0, 2, 1, 3);
cb(q_nope, "q_nope_perm", il);

struct ggml_tensor * q_nope2 = ggml_mul_mat(ctx0, wk_b, q_nope);
cb(q_nope2, "q_nope2", il);

if (!pp_opt) {
q_nope2 = ggml_permute(ctx0, q_nope2, 0, 2, 1, 3);
cb(q_nope2, "q_nope2_perm", il);
}

struct ggml_tensor * kq_nope = ggml_mul_mat(ctx0, kv_cache, q_nope2);
cb(kq_nope, "kq_nope", il);

if (!pp_opt) {
kq_nope = ggml_permute(ctx0, kq_nope, 0, 2, 1, 3);
cb(kq_nope, "kq_nope_perm", il);
}

if (pp_opt) {
q_pe = ggml_permute(ctx0, q_pe, 0, 2, 1, 3);
cb(q_pe, "q_pe_perm", il);
}

struct ggml_tensor * kq_pe = ggml_mul_mat(ctx0, kr_cache, q_pe);
cb(kq_pe, "kq_pe", il);

if (!pp_opt) {
kq_pe = ggml_permute(ctx0, kq_pe, 0, 2, 1, 3);
cb(kq_pe, "kq_pe_perm", il);
}

struct ggml_tensor * kq = ggml_add(ctx0, kq_nope, kq_pe);
cb(kq, "kq", il);

kq = ggml_soft_max_ext(ctx0, kq, KQ_mask, kq_scale, hparams.f_max_alibi_bias);
cb(kq, "kq_soft_max_ext", il);

if (!pp_opt) {
kq = ggml_permute(ctx0, kq, 0, 2, 1, 3);
cb(kq, "kq_soft_max_ext_perm", il);
}

struct ggml_tensor * kqv_compressed = ggml_mul_mat(ctx0, kv_cache_trans, kq);
cb(kqv_compressed, "kqv_compressed", il);

if (!pp_opt) {
kqv_compressed = ggml_permute(ctx0, kqv_compressed, 0, 2, 3, 1);
cb(kqv_compressed, "kqv_compressed_perm", il);
}

struct ggml_tensor * wv_b = ggml_view_3d(ctx0, model.layers[il].wv_b, kv_lora_rank, n_embd_head_v, n_head, ggml_row_size(model.layers[il].wv_b->type, kv_lora_rank), ggml_row_size(model.layers[il].wv_b->type, kv_lora_rank * n_embd_head_v), 0);
cb(wv_b, "wv_b", il);

struct ggml_tensor * kqv = ggml_mul_mat(ctx0, wv_b, kqv_compressed);
cb(kqv, "kqv", il);

if (pp_opt) {
kqv = ggml_cont(ctx0, ggml_permute(ctx0, kqv, 0, 2, 1, 3));
cb(kqv, "kqv_perm", il);
}

cur = ggml_view_2d(ctx0, kqv, n_embd_head_v*n_head, n_tokens, ggml_row_size(kqv->type, n_embd_head_v*n_head), 0);
cb(cur, "kqv_2d", il);

ggml_build_forward_expand(gf, cur);

cur = llm_build_lora_mm(lctx, ctx0, model.layers[il].wo, cur);
cb(cur, "kqv_out", il);
}

if (il == n_layer - 1) {
Expand Down Expand Up @@ -9708,6 +9788,24 @@ struct llama_context * llama_init_from_model(
ggml_type_name(type_v), (float)memory_size_v / (1024.0f * 1024.0f));
}

{
size_t memory_size_kr = 0;
size_t memory_size_kv = 0;

for (auto & kr : ctx->kv_self.kr_l) {
memory_size_kr += ggml_nbytes(kr);
}

for (auto & kv : ctx->kv_self.kv_l) {
memory_size_kv += ggml_nbytes(kv);
}

LLAMA_LOG_INFO("%s: KV self size = %7.2f MiB, K^R (%s): %7.2f MiB, c^KV (%s): %7.2f MiB\n", __func__,
(float)(memory_size_kr + memory_size_kv) / (1024.0f * 1024.0f),
ggml_type_name(type_k), (float)memory_size_kr / (1024.0f * 1024.0f),
ggml_type_name(type_k), (float)memory_size_kv / (1024.0f * 1024.0f));
}

// graph outputs buffer
{
// resized during inference when a batch uses more outputs
Expand Down
Loading