memory : Hybrid recurrent cache (#13979)
* feat: Add llama_model_is_hybrid API call Also, split llama_model_is_recurrent into llm_arch_is_recurrent in llama-arch with llama_model_is_recurrent delegating to llm_arch_is_recurrent. The same split is done for hybird. This is needed because there are places where the llama_model has not yet been initialized but we need to check if the model is recurrent (specifically for the per-layer recurrent check array in hparams). Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Add c++ side constants for attention layer indices hparam Branch: GraniteFour * feat: Add support for distinguishing recurrent vs non-recurrent layers in hparams Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Auto-fill hparams.recurrent_layer_arr based on whether the model is recurrent Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: rename *_is_hybrid -> *_is_hybrid_recurrent The implementation of the hybrid cache intentionally does not specify the types of the child caches, so there was a naming mismatch with these predicate functions that used "hybrid" to imply "hybrid recurrent." Branch: HybridCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Add layer filter to recurrent cache Branch: HybridCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Use per-layer sizing everywhere in kv caches Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: First pass at llama_kv_cache_hybrid_recurrent This follows the pattern in iswa where the two child caches are held explicitly to support the case where a model requires a single attention cache and a single recurrent cache where each layer uses exactly one of the caches. This is a rewrite of the more generic approach in the original hybrid cache PR: https://github.com/ggml-org/llama.cpp/pull/13276 Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Construct hybrid recurrent cache for hybrid recurrent models This includes a refactor of the create_memory logic to avoid needing to use the arch enum explicitly unless a model needs explicit cache instantiation logic beyond the standard logic for recurrent, hybrid, unified, and iswa. Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Fix wrong bool condition for split equal in hybrid cache Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Fix shift logic to defer to unified cache Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Support hybrid recurrent in llama-graph NOTE: I intentionally did not add support for s_mask since it will be going away soon Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Fix logic for initializing inputs and attn layers for hybrid caches Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Update recurrent cache for changes to remove intermediate kv_cache interface Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Fix status for init_update sig for recurrent cache state Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Add missing padding to n_ctx for hybrid cache construction Branch: GraniteFour Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Update clear signature for data argument after rebase Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Remove errant virtual destructor leftover from previous impl attempt Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Use per-layer n_embd_k/v_s calls for mamba (1) layers Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: Remove n_embd_k/v_s from unified cache No longer needed now that unified isn't also supporting recurrent https://github.com/ggml-org/llama.cpp/pull/13979#discussion_r2140761069 Branch: HybridRecurrentCache * refactor: Remove layer index from n_embd_k/v_s Now that it's not used at all in the unified cache, we don't need to use the layer index to zero it out for attention layers. Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: Remove n_embd_k/v_gqa from recurrent cache This is no longer needed now that there are separate implementations https://github.com/ggml-org/llama.cpp/pull/13979#discussion_r2140825128 Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Allow custom layer filters for hybrid recurrent This should help support architectures like Falcon H1 where there is overlap between layers that need attention and recurrent caches. https://github.com/ggml-org/llama.cpp/pull/13979#discussion_r2140748922 Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Remove logits_all after rebase Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Remove llama_model_is_hybrid_Recurrent public API https://github.com/ggml-org/llama.cpp/pull/13979#discussion_r2141728423 Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: Use llama_memory_state_ptr for child states in hybrid memory state Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Overhaul build_recurrent_state / build_inp_s_copy to match attention pattern https://github.com/ggml-org/llama.cpp/pull/13979/files#r2141701738 This is a big overhaul to bring consistency between how inputs and per- layer components are created for attention layers and recurrent layers. The main changes are: - Rename class llm_graph_input_s_copy -> llm_graph_input_rs - Add a corresponding llm_graph_input_rs_hybrid_recurrent - Rename build_inp_s_copy -> build_rs_inp_recurrent - Add a corresponding build_rs_inp_hybrid_recurrent - Rename build_recurrent_state -> build_rs to match build_attn w/ llm_graph_input_rs android-build AUTHORS bamba-9b-2.2T.gguf bamba-9b-2.2T.q4_k_m.gguf broken.log build build-rel build-xcframework.sh build.android build.android.bak ci cmake CMakeLists.txt CMakePresets.json CODEOWNERS common common.o CONTRIBUTING.md convert_hf_to_gguf_update.py convert_hf_to_gguf.py convert_llama_ggml_to_gguf.py convert_lora_to_gguf.py debug.log docs examples flake.lock flake.nix ggml ggml-alloc.o ggml-backend.o ggml-metal.o ggml-model-BF16.gguf ggml-model-Q4_K_M.gguf ggml-quants.o ggml.o gguf-py grammar-parser.o grammars include LICENSE licenses llama.log llama.o llamacpp_trace.log main.log Makefile media models mypy.ini pocs poetry.lock prompts pyproject.toml pyrightconfig.json q4_k_m_boot.log q8_0_boot.log quant.log quant2.log README.md requirements requirements.txt sampling.o scripts SECURITY.md src test-grammar-output.tmp test-json-schema-input.tmp tests tools vendor working.log as the first input - Add a corresponding overload of build_rs w/ llm_graph_input_rs_hybrid_recurrent android-build AUTHORS bamba-9b-2.2T.gguf bamba-9b-2.2T.q4_k_m.gguf broken.log build build-rel build-xcframework.sh build.android build.android.bak ci cmake CMakeLists.txt CMakePresets.json CODEOWNERS common common.o CONTRIBUTING.md convert_hf_to_gguf_update.py convert_hf_to_gguf.py convert_llama_ggml_to_gguf.py convert_lora_to_gguf.py debug.log docs examples flake.lock flake.nix ggml ggml-alloc.o ggml-backend.o ggml-metal.o ggml-model-BF16.gguf ggml-model-Q4_K_M.gguf ggml-quants.o ggml.o gguf-py grammar-parser.o grammars include LICENSE licenses llama.log llama.o llamacpp_trace.log main.log Makefile media models mypy.ini pocs poetry.lock prompts pyproject.toml pyrightconfig.json q4_k_m_boot.log q8_0_boot.log quant.log quant2.log README.md requirements requirements.txt sampling.o scripts SECURITY.md src test-grammar-output.tmp test-json-schema-input.tmp tests tools vendor working.log as the first input - Add a llm_graph_input_attn_kv_hybrid_recurrent analogous to llm_graph_input_attn_kv_unified - Add a build_attn override that takes llm_graph_input_attn_kv_hybrid_recurrent android-build AUTHORS bamba-9b-2.2T.gguf bamba-9b-2.2T.q4_k_m.gguf broken.log build build-rel build-xcframework.sh build.android build.android.bak ci cmake CMakeLists.txt CMakePresets.json CODEOWNERS common common.o CONTRIBUTING.md convert_hf_to_gguf_update.py convert_hf_to_gguf.py convert_llama_ggml_to_gguf.py convert_lora_to_gguf.py debug.log docs examples flake.lock flake.nix ggml ggml-alloc.o ggml-backend.o ggml-metal.o ggml-model-BF16.gguf ggml-model-Q4_K_M.gguf ggml-quants.o ggml.o gguf-py grammar-parser.o grammars include LICENSE licenses llama.log llama.o llamacpp_trace.log main.log Makefile media models mypy.ini pocs poetry.lock prompts pyproject.toml pyrightconfig.json q4_k_m_boot.log q8_0_boot.log quant.log quant2.log README.md requirements requirements.txt sampling.o scripts SECURITY.md src test-grammar-output.tmp test-json-schema-input.tmp tests tools vendor working.log as the first input This makes the two paradigms fully consistent. The main drawback is the code duplication in the build_attn and build_rs implementations where the only difference between implementations is how they cast the memory state. Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Fix resize vs reserve and skip null tensors in size computation https://github.com/ggml-org/llama.cpp/pull/13979/files#r2149469788 Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-Authored-By: @younesbelkada * fix: Fix initialization of child states Since initially writing this PR, the logic in the child state types changed such that using the "init full" signature and keeping the ubatches on the parent struct no longer worked. Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: Use a common build_recurrent_state method that is cache-agnostic This reduces the code duplication between the different build_rs impls and also retains a similar signature to the previous build_recurrent_state method while standardizing on the input-dispatched build_rs implementation. Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * recurrent : rework graph inputs + add TODOs ggml-ci * refactor: Make status and child states const in hybrid and iswa Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: Rename llama_kv_cache_[recurrent|hybrid_recurrent] to remove kv cache This removes the notion of "kv" from the interface names for these memory types. There are still many references to kv in the implementation of the recurrent memory which will need further adjustment. Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor!: Rename all k/v related values for recurrent/hybrid to r/s Anywhere that "kv_<state|cell|size|etc>" is used, I've used the more generic "mem_" prefix. The specifics of "k" (key) translate to "r" (recurrent state) and "v" (value) translate to "s" (state-space embedding states). Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refacor: _recurrent -> _recr for brevity It just _happens_ to have the same number of letters as _attn! Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * style: Fix spacing for ref Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * refactor: recurrent_layer() -> is_recurrent() Branch: HybridRecurrentCache Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * style: Fix spacing for size_s_bytes declaration Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Gabe Goodhart
GitHub
parent
ed3290ab34
commit
edc4a29eff
15 changed files with 1084 additions and 461 deletions
|
|
@ -6,7 +6,8 @@
|
|||
|
||||
#include "llama-kv-cache-unified.h"
|
||||
#include "llama-kv-cache-unified-iswa.h"
|
||||
#include "llama-kv-cache-recurrent.h"
|
||||
#include "llama-memory-hybrid.h"
|
||||
#include "llama-memory-recurrent.h"
|
||||
|
||||
#include <cassert>
|
||||
#include <cmath>
|
||||
|
|
@ -238,18 +239,18 @@ void llm_graph_input_cls::set_input(const llama_ubatch * ubatch) {
|
|||
}
|
||||
}
|
||||
|
||||
void llm_graph_input_s_copy::set_input(const llama_ubatch * ubatch) {
|
||||
void llm_graph_input_rs::set_input(const llama_ubatch * ubatch) {
|
||||
GGML_UNUSED(ubatch);
|
||||
|
||||
const int64_t n_kv = kv_state->get_n_kv();
|
||||
const int64_t n_rs = mem_state->get_n_rs();
|
||||
|
||||
if (s_copy) {
|
||||
GGML_ASSERT(ggml_backend_buffer_is_host(s_copy->buffer));
|
||||
int32_t * data = (int32_t *) s_copy->data;
|
||||
|
||||
// assuming copy destinations ALWAYS happen ONLY on the cells between head and head+n
|
||||
for (uint32_t i = 0; i < n_kv; ++i) {
|
||||
data[i] = kv_state->s_copy(i);
|
||||
for (uint32_t i = 0; i < n_rs; ++i) {
|
||||
data[i] = mem_state->s_copy(i);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -403,6 +404,24 @@ void llm_graph_input_attn_cross::set_input(const llama_ubatch * ubatch) {
|
|||
}
|
||||
}
|
||||
|
||||
void llm_graph_input_mem_hybrid::set_input(const llama_ubatch * ubatch) {
|
||||
if (self_kq_mask) {
|
||||
mem_state->get_state_attn()->set_input_kq_mask(self_kq_mask, ubatch, cparams.causal_attn);
|
||||
}
|
||||
|
||||
const int64_t n_rs = mem_state->get_state_recr()->get_n_rs();
|
||||
|
||||
if (s_copy) {
|
||||
GGML_ASSERT(ggml_backend_buffer_is_host(s_copy->buffer));
|
||||
int32_t * data = (int32_t *) s_copy->data;
|
||||
|
||||
// assuming copy destinations ALWAYS happen ONLY on the cells between head and head+n
|
||||
for (uint32_t i = 0; i < n_rs; ++i) {
|
||||
data[i] = mem_state->get_state_recr()->s_copy(i);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
//
|
||||
// llm_graph_context
|
||||
//
|
||||
|
|
@ -961,23 +980,6 @@ ggml_tensor * llm_graph_context::build_inp_cls() const {
|
|||
return cur;
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_inp_s_copy() const {
|
||||
const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
|
||||
|
||||
auto inp = std::make_unique<llm_graph_input_s_copy>(kv_state);
|
||||
|
||||
const auto n_kv = kv_state->get_n_kv();
|
||||
|
||||
auto & cur = inp->s_copy;
|
||||
|
||||
cur = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_kv);
|
||||
ggml_set_input(cur);
|
||||
|
||||
res->add_input(std::move(inp));
|
||||
|
||||
return cur;
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_inp_cross_embd() const {
|
||||
auto inp = std::make_unique<llm_graph_input_cross_embd>(cross);
|
||||
|
||||
|
|
@ -1047,6 +1049,33 @@ ggml_tensor * llm_graph_context::build_pos_bias(ggml_tensor * pos_bucket, ggml_t
|
|||
return pos_bias;
|
||||
}
|
||||
|
||||
llm_graph_input_mem_hybrid * llm_graph_context::build_inp_mem_hybrid() const {
|
||||
const auto * mem_state = static_cast<const llama_memory_hybrid_state *>(mstate);
|
||||
|
||||
auto inp = std::make_unique<llm_graph_input_mem_hybrid>(hparams, cparams, mem_state);
|
||||
|
||||
{
|
||||
GGML_ASSERT(hparams.swa_type == LLAMA_SWA_TYPE_NONE && "Hybrid recurrent is not supported with SWA attention layers");
|
||||
|
||||
const auto n_kv = inp->mem_state->get_state_attn()->get_n_kv();
|
||||
|
||||
inp->self_kq_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
|
||||
//cb(inp->self_kq_mask, "KQ_mask", -1);
|
||||
ggml_set_input(inp->self_kq_mask);
|
||||
|
||||
inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
|
||||
}
|
||||
|
||||
{
|
||||
const auto n_rs = mem_state->get_state_recr()->get_n_rs();
|
||||
|
||||
inp->s_copy = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_rs);
|
||||
ggml_set_input(inp->s_copy);
|
||||
}
|
||||
|
||||
return (llm_graph_input_mem_hybrid *) res->add_input(std::move(inp));
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_attn_mha(
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * q,
|
||||
|
|
@ -1291,36 +1320,6 @@ ggml_tensor * llm_graph_context::build_attn(
|
|||
return cur;
|
||||
}
|
||||
|
||||
llm_graph_input_attn_kv_unified_iswa * llm_graph_context::build_attn_inp_kv_unified_iswa() const {
|
||||
const auto * kv_state = static_cast<const llama_kv_cache_unified_iswa_state *>(mstate);
|
||||
|
||||
auto inp = std::make_unique<llm_graph_input_attn_kv_unified_iswa>(hparams, cparams, kv_state);
|
||||
|
||||
{
|
||||
const auto n_kv = kv_state->get_base()->get_n_kv();
|
||||
|
||||
inp->self_kq_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
|
||||
//cb(inp->self_kq_mask, "KQ_mask", -1);
|
||||
ggml_set_input(inp->self_kq_mask);
|
||||
|
||||
inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
|
||||
}
|
||||
|
||||
{
|
||||
GGML_ASSERT(hparams.swa_type != LLAMA_SWA_TYPE_NONE && "Use llama_kv_cache_unified for non-SWA");
|
||||
|
||||
const auto n_kv = kv_state->get_swa()->get_n_kv();
|
||||
|
||||
inp->self_kq_mask_swa = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
|
||||
//cb(inp->self_kq_mask_swa, "KQ_mask_swa", -1);
|
||||
ggml_set_input(inp->self_kq_mask_swa);
|
||||
|
||||
inp->self_kq_mask_swa_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask_swa, GGML_TYPE_F16) : inp->self_kq_mask_swa;
|
||||
}
|
||||
|
||||
return (llm_graph_input_attn_kv_unified_iswa *) res->add_input(std::move(inp));
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_attn(
|
||||
llm_graph_input_attn_kv_unified_iswa * inp,
|
||||
ggml_cgraph * gf,
|
||||
|
|
@ -1430,20 +1429,99 @@ ggml_tensor * llm_graph_context::build_attn(
|
|||
return cur;
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_recurrent_state(
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * s,
|
||||
ggml_tensor * state_copy,
|
||||
int32_t state_size,
|
||||
int32_t n_seqs,
|
||||
bool avoid_copies) const {
|
||||
const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
|
||||
ggml_tensor * llm_graph_context::build_attn(
|
||||
llm_graph_input_mem_hybrid * inp,
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * wo,
|
||||
ggml_tensor * wo_b,
|
||||
ggml_tensor * q_cur,
|
||||
ggml_tensor * k_cur,
|
||||
ggml_tensor * v_cur,
|
||||
ggml_tensor * kq_b,
|
||||
ggml_tensor * v_mla,
|
||||
float kq_scale,
|
||||
int il) const {
|
||||
// these nodes are added to the graph together so that they are not reordered
|
||||
// by doing so, the number of splits in the graph is reduced
|
||||
ggml_build_forward_expand(gf, q_cur);
|
||||
ggml_build_forward_expand(gf, k_cur);
|
||||
ggml_build_forward_expand(gf, v_cur);
|
||||
|
||||
const auto n_kv = kv_state->get_n_kv();
|
||||
const auto kv_head = kv_state->get_head();
|
||||
const auto rs_zero = kv_state->get_rs_z();
|
||||
const auto * kv_state = static_cast<const llama_memory_hybrid_state *>(mstate)->get_state_attn();
|
||||
|
||||
ggml_tensor * states = ggml_reshape_2d(ctx0, s, state_size, kv_state->get_size());
|
||||
// store to KV cache
|
||||
{
|
||||
ggml_build_forward_expand(gf, kv_state->cpy_k(ctx0, k_cur, il));
|
||||
ggml_build_forward_expand(gf, kv_state->cpy_v(ctx0, v_cur, il));
|
||||
}
|
||||
|
||||
const auto & kq_mask = inp->get_kq_mask();
|
||||
|
||||
ggml_tensor * q = q_cur;
|
||||
ggml_tensor * k = kv_state->get_k(ctx0, il);
|
||||
ggml_tensor * v = kv_state->get_v(ctx0, il);
|
||||
|
||||
ggml_tensor * cur = build_attn_mha(gf, q, k, v, kq_b, kq_mask, v_mla, kq_scale);
|
||||
cb(cur, "kqv_out", il);
|
||||
|
||||
if (wo) {
|
||||
cur = build_lora_mm(wo, cur);
|
||||
if (arch == LLM_ARCH_GLM4) {
|
||||
// GLM4 seems to have numerical issues with half-precision accumulators
|
||||
ggml_mul_mat_set_prec(cur, GGML_PREC_F32);
|
||||
}
|
||||
}
|
||||
|
||||
if (wo_b) {
|
||||
cur = ggml_add(ctx0, cur, wo_b);
|
||||
}
|
||||
|
||||
return cur;
|
||||
}
|
||||
|
||||
llm_graph_input_attn_kv_unified_iswa * llm_graph_context::build_attn_inp_kv_unified_iswa() const {
|
||||
const auto * kv_state = static_cast<const llama_kv_cache_unified_iswa_state *>(mstate);
|
||||
|
||||
auto inp = std::make_unique<llm_graph_input_attn_kv_unified_iswa>(hparams, cparams, kv_state);
|
||||
|
||||
{
|
||||
const auto n_kv = kv_state->get_base()->get_n_kv();
|
||||
|
||||
inp->self_kq_mask = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
|
||||
//cb(inp->self_kq_mask, "KQ_mask", -1);
|
||||
ggml_set_input(inp->self_kq_mask);
|
||||
|
||||
inp->self_kq_mask_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask, GGML_TYPE_F16) : inp->self_kq_mask;
|
||||
}
|
||||
|
||||
{
|
||||
GGML_ASSERT(hparams.swa_type != LLAMA_SWA_TYPE_NONE && "Use llama_kv_cache_unified for non-SWA");
|
||||
|
||||
const auto n_kv = kv_state->get_swa()->get_n_kv();
|
||||
|
||||
inp->self_kq_mask_swa = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, n_kv, GGML_PAD(n_tokens, GGML_KQ_MASK_PAD));
|
||||
//cb(inp->self_kq_mask_swa, "KQ_mask_swa", -1);
|
||||
ggml_set_input(inp->self_kq_mask_swa);
|
||||
|
||||
inp->self_kq_mask_swa_cnv = cparams.flash_attn ? ggml_cast(ctx0, inp->self_kq_mask_swa, GGML_TYPE_F16) : inp->self_kq_mask_swa;
|
||||
}
|
||||
|
||||
return (llm_graph_input_attn_kv_unified_iswa *) res->add_input(std::move(inp));
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_rs(
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * s,
|
||||
ggml_tensor * state_copy,
|
||||
int32_t state_size,
|
||||
int32_t n_seqs,
|
||||
uint32_t n_kv,
|
||||
uint32_t kv_head,
|
||||
uint32_t kv_size,
|
||||
int32_t rs_zero,
|
||||
bool avoid_copies) const {
|
||||
|
||||
ggml_tensor * states = ggml_reshape_2d(ctx0, s, state_size, kv_size);
|
||||
|
||||
// Clear a single state which will then be copied to the other cleared states.
|
||||
// Note that this is a no-op when the view is zero-sized.
|
||||
|
|
@ -1474,22 +1552,59 @@ ggml_tensor * llm_graph_context::build_recurrent_state(
|
|||
return output_states;
|
||||
}
|
||||
|
||||
llm_graph_input_rs * llm_graph_context::build_rs_inp() const {
|
||||
const auto * kv_state = static_cast<const llama_memory_recurrent_state *>(mstate);
|
||||
|
||||
auto inp = std::make_unique<llm_graph_input_rs>(kv_state);
|
||||
|
||||
const auto n_rs = kv_state->get_n_rs();
|
||||
|
||||
inp->s_copy = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_rs);
|
||||
ggml_set_input(inp->s_copy);
|
||||
|
||||
return (llm_graph_input_rs *) res->add_input(std::move(inp));
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_rs(
|
||||
llm_graph_input_rs * inp,
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * s,
|
||||
int32_t state_size,
|
||||
int32_t n_seqs,
|
||||
bool avoid_copies) const {
|
||||
const auto * kv_state = static_cast<const llama_memory_recurrent_state *>(mstate);
|
||||
|
||||
return build_rs(gf, s, inp->s_copy, state_size, n_seqs, kv_state->get_n_rs(), kv_state->get_head(), kv_state->get_size(), kv_state->get_rs_z(), avoid_copies);
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_rs(
|
||||
llm_graph_input_mem_hybrid * inp,
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * s,
|
||||
int32_t state_size,
|
||||
int32_t n_seqs,
|
||||
bool avoid_copies) const {
|
||||
const auto * kv_state = static_cast<const llama_memory_hybrid_state *>(mstate)->get_state_recr();
|
||||
|
||||
return build_rs(gf, s, inp->s_copy, state_size, n_seqs, kv_state->get_n_rs(), kv_state->get_head(), kv_state->get_size(), kv_state->get_rs_z(), avoid_copies);
|
||||
}
|
||||
|
||||
ggml_tensor * llm_graph_context::build_rwkv_token_shift_load(
|
||||
ggml_cgraph * gf,
|
||||
ggml_tensor * state_copy,
|
||||
const llama_ubatch & ubatch,
|
||||
llm_graph_input_rs * inp,
|
||||
ggml_cgraph * gf,
|
||||
const llama_ubatch & ubatch,
|
||||
int il) const {
|
||||
const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
|
||||
const auto * kv_state = static_cast<const llama_memory_recurrent_state *>(mstate);
|
||||
|
||||
const auto token_shift_count = hparams.token_shift_count;
|
||||
|
||||
const int64_t n_seqs = ubatch.n_seqs;
|
||||
|
||||
ggml_tensor * token_shift_all = kv_state->get_k_l(il);
|
||||
ggml_tensor * token_shift_all = kv_state->get_r_l(il);
|
||||
|
||||
ggml_tensor * token_shift = build_recurrent_state(
|
||||
gf, token_shift_all, state_copy,
|
||||
hparams.n_embd_k_s(), n_seqs);
|
||||
ggml_tensor * token_shift = build_rs(
|
||||
inp, gf, token_shift_all,
|
||||
hparams.n_embd_r(), n_seqs);
|
||||
|
||||
token_shift = ggml_reshape_3d(ctx0, token_shift, hparams.n_embd, token_shift_count, n_seqs);
|
||||
|
||||
|
|
@ -1500,7 +1615,7 @@ ggml_tensor * llm_graph_context::build_rwkv_token_shift_store(
|
|||
ggml_tensor * token_shift,
|
||||
const llama_ubatch & ubatch,
|
||||
int il) const {
|
||||
const auto * kv_state = static_cast<const llama_kv_cache_recurrent_state *>(mstate);
|
||||
const auto * kv_state = static_cast<const llama_memory_recurrent_state *>(mstate);
|
||||
|
||||
const auto token_shift_count = hparams.token_shift_count;
|
||||
const auto n_embd = hparams.n_embd;
|
||||
|
|
@ -1512,7 +1627,7 @@ ggml_tensor * llm_graph_context::build_rwkv_token_shift_store(
|
|||
return ggml_cpy(
|
||||
ctx0,
|
||||
ggml_view_1d(ctx0, token_shift, n_embd * n_seqs * token_shift_count, 0),
|
||||
ggml_view_1d(ctx0, kv_state->get_k_l(il), hparams.n_embd_k_s()*n_seqs, hparams.n_embd_k_s()*kv_head*ggml_element_size(kv_state->get_k_l(il)))
|
||||
ggml_view_1d(ctx0, kv_state->get_r_l(il), hparams.n_embd_r()*n_seqs, hparams.n_embd_r()*kv_head*ggml_element_size(kv_state->get_r_l(il)))
|
||||
);
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue