vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : add alternative KV cache merging (EXPERIMENTAL)

Browse source
This commit is contained in:
Georgi Gerganov 2024-02-25 13:57:43 +02:00
parent 0d6f8734b3
commit 9ec749df59
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 59 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

67
llama.cpp
View file

@ -8072,10 +8072,13 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
const auto & hparams = lctx.model.hparams; const auto & hparams = lctx.model.hparams;
const uint32_t n_layer = hparams.n_layer; const uint32_t n_layer = hparams.n_layer;
const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa(); const uint32_t n_embd_k_gqa = hparams.n_embd_k_gqa();
const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa(); const uint32_t n_embd_v_gqa = hparams.n_embd_v_gqa();
const uint32_t kv_size = kv_self.size; const uint32_t n_embd_head_k = hparams.n_embd_head_k; GGML_UNUSED(n_embd_head_k);
const uint32_t n_embd_head_v = hparams.n_embd_head_v; GGML_UNUSED(n_embd_head_v);
const uint32_t n_head_kv = hparams.n_head_kv; GGML_UNUSED(n_head_kv);
const uint32_t kv_size = kv_self.size;
std::vector<uint8_t> buf_q; std::vector<uint8_t> buf_q;
@ -8085,9 +8088,9 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
const int64_t t_start = ggml_time_us(); const int64_t t_start = ggml_time_us();
struct c_pair { uint32_t i0, i1; }; struct c_pair { uint32_t i0, i1; };
struct c_info { bool merged; uint32_t id, cnt;}; struct c_info { bool merged; uint32_t id, cnt, r; };
std::vector<c_info> infos(kv_size, { false, 0, 0 }); std::vector<c_info> infos(kv_size, { false, 0, 0, 0 });
// the destination cell in the new KV cache // the destination cell in the new KV cache
uint32_t id = 0; uint32_t id = 0;
@ -8100,7 +8103,7 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
const auto & cell0 = kv_self.cells[i0]; const auto & cell0 = kv_self.cells[i0];
if (!cell0.is_empty() && !infos[i0].merged) { if (!cell0.is_empty() && !infos[i0].merged) {
infos[i0] = { true, id, 0 }; infos[i0] = { true, id, 0, 0 };
infos[id].cnt = 1; infos[id].cnt = 1;
const llama_pos p0 = cell0.pos; const llama_pos p0 = cell0.pos;
@ -8112,7 +8115,7 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
const llama_pos p1 = cell1.pos; const llama_pos p1 = cell1.pos;
if (std::abs(p0 - p1) <= kv_self.compress_delta) { if (std::abs(p0 - p1) <= kv_self.compress_delta) {
infos[i1] = { true, id, 0 }; infos[i1] = { true, id, 0, 0 };
infos[id].cnt++; infos[id].cnt++;
n_merges++; n_merges++;
} }
@ -8143,6 +8146,10 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
tt_v = ggml_internal_get_type_traits(kv_self.type_v); tt_v = ggml_internal_get_type_traits(kv_self.type_v);
for (uint32_t il = 0; il < n_layer; ++il) { for (uint32_t il = 0; il < n_layer; ++il) {
for (uint32_t i = 0; i < kv_size; ++i) {
infos[i].r = 0;
}
// update keys // update keys
{ {
const int64_t ne = n_embd_k_gqa*kv_size; const int64_t ne = n_embd_k_gqa*kv_size;
@ -8167,6 +8174,7 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
const uint32_t id = infos[i].id; const uint32_t id = infos[i].id;
#if 1
// merge using averaging // merge using averaging
{ {
const float scale = 1.0f/float(infos[id].cnt); const float scale = 1.0f/float(infos[id].cnt);
@ -8178,6 +8186,25 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
buf_dst_f32[od + j] += buf_src_f32[os + j]*scale; buf_dst_f32[od + j] += buf_src_f32[os + j]*scale;
} }
} }
#else
// merge separate heads
{
for (uint32_t h = 0; h < n_head_kv; ++h) {
if ((h + il) % infos[id].cnt != infos[id].r) {
continue;
}
const int64_t os = i*n_embd_k_gqa + h*n_embd_head_k;
const int64_t od = id*n_embd_k_gqa + h*n_embd_head_k;
for (uint32_t j = 0; j < n_embd_head_k; ++j) {
buf_dst_f32[od + j] = buf_src_f32[os + j];
}
}
}
infos[id].r++;
#endif
} }
tt_k.from_float(buf_dst_f32.data(), buf_q.data(), ne); tt_k.from_float(buf_dst_f32.data(), buf_q.data(), ne);
@ -8185,6 +8212,10 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
ggml_backend_tensor_set(kv_self.k_l[il], buf_q.data(), 0, buf_q.size()); ggml_backend_tensor_set(kv_self.k_l[il], buf_q.data(), 0, buf_q.size());
} }
for (uint32_t i = 0; i < kv_size; ++i) {
infos[i].r = 0;
}
// update values (note: they are transposed) // update values (note: they are transposed)
{ {
const int64_t ne = n_embd_v_gqa*kv_size; const int64_t ne = n_embd_v_gqa*kv_size;
@ -8209,6 +8240,7 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
const uint32_t id = infos[i].id; const uint32_t id = infos[i].id;
#if 1
// merge using averaging // merge using averaging
{ {
const float scale = 1.0f/float(infos[id].cnt); const float scale = 1.0f/float(infos[id].cnt);
@ -8221,6 +8253,25 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
buf_dst_f32[od + j*kv_size] += buf_src_f32[os + j*kv_size]*scale; buf_dst_f32[od + j*kv_size] += buf_src_f32[os + j*kv_size]*scale;
} }
} }
#else
// merge separate heads
{
for (uint32_t h = 0; h < n_head_kv; ++h) {
if ((h + il) % infos[id].cnt != infos[id].r) {
continue;
}
const int64_t os = i;
const int64_t od = id;
for (uint32_t j = h*n_embd_head_v; j < (h + 1)*n_embd_head_v; ++j) {
buf_dst_f32[od + j*kv_size] = buf_src_f32[os + j*kv_size];
}
}
}
infos[id].r++;
#endif
} }
tt_v.from_float(buf_dst_f32.data(), buf_q.data(), ne); tt_v.from_float(buf_dst_f32.data(), buf_q.data(), ne);