mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-11-03 09:22:01 +00:00
01612b7409
* llama : reuse compute graphs ggml-ci * llama-bench : add graph reuse parameter ggml-ci * cont : remove the parameter and the sched resets ggml-ci * graph : rename update() to can_reuse() ggml-ci * params : remove is_same() ggml-ci * graph : set res->params in llm_graph_context constructor ggml-ci * graph : avoid set_max_nodes in llm_graph_result ggml-ci * kv-cache : reuse llama_context's graph result instance ggml-ci * context : reset the previous graph result upon memory updates ggml-ci * batch : llama_ubatch now carries its data instead of pointing to balloc ggml-ci * merge : fix build ggml-ci * graph : fix can_reuse() checks when flash-attention is disabled * graph : move llm_graph_result impl in source file + debug env ggml-ci
400 lines
12 KiB
C++
400 lines
12 KiB
C++
#pragma once
|
|
|
|
#include "llama-batch.h"
|
|
#include "llama-graph.h"
|
|
#include "llama-kv-cells.h"
|
|
#include "llama-memory.h"
|
|
|
|
#include <unordered_map>
|
|
#include <vector>
|
|
|
|
struct llama_cparams;
|
|
struct llama_hparams;
|
|
struct llama_model;
|
|
struct llama_context;
|
|
|
|
//
|
|
// llama_kv_cache_unified
|
|
//
|
|
|
|
class llama_kv_cache_unified : public llama_memory_i {
|
|
public:
|
|
static uint32_t get_padding(const llama_cparams & cparams);
|
|
|
|
// this callback is used to filter out layers that should not be included in the cache
|
|
using layer_filter_cb = std::function<bool(int32_t il)>;
|
|
|
|
struct defrag_info {
|
|
bool empty() const {
|
|
return ids.empty();
|
|
}
|
|
|
|
// contains information about which cell moves where:
|
|
// - cell i moves to ids[i]
|
|
// - if ids[i] == i || ids[i] == ids.size(), then cell i is not moved
|
|
std::vector<uint32_t> ids;
|
|
};
|
|
|
|
struct stream_copy_info {
|
|
bool empty() const {
|
|
assert(ssrc.size() == sdst.size());
|
|
return ssrc.empty();
|
|
}
|
|
|
|
std::vector<uint32_t> ssrc;
|
|
std::vector<uint32_t> sdst;
|
|
};
|
|
|
|
// for each ubatch, create a slot_info that contains information about where the ubatch should be inserted in the
|
|
// KV cells. for example, cell indices for each token, such that: token[i] -> goes to cells[idxs[i]]
|
|
struct slot_info {
|
|
// data for ggml_set_rows
|
|
using idx_vec_t = std::vector<uint32_t>;
|
|
|
|
// number of streams: ns = s1 - s0 + 1
|
|
llama_seq_id s0;
|
|
llama_seq_id s1;
|
|
|
|
std::vector<llama_seq_id> strm; // [ns]
|
|
std::vector<idx_vec_t> idxs; // [ns]
|
|
|
|
uint32_t head() const {
|
|
GGML_ASSERT(idxs.size() == 1);
|
|
GGML_ASSERT(!idxs[0].empty());
|
|
|
|
return idxs[0][0];
|
|
}
|
|
|
|
void resize(size_t n) {
|
|
strm.resize(n);
|
|
idxs.resize(n);
|
|
}
|
|
|
|
size_t size() const {
|
|
GGML_ASSERT(idxs.size() == strm.size());
|
|
GGML_ASSERT(!idxs.empty());
|
|
|
|
return idxs[0].size();
|
|
}
|
|
|
|
size_t n_stream() const {
|
|
return strm.size();
|
|
}
|
|
|
|
bool empty() const {
|
|
return idxs.empty();
|
|
}
|
|
|
|
void clear() {
|
|
idxs.clear();
|
|
}
|
|
};
|
|
|
|
using slot_info_vec_t = std::vector<slot_info>;
|
|
|
|
llama_kv_cache_unified(
|
|
const llama_model & model,
|
|
layer_filter_cb && filter,
|
|
ggml_type type_k,
|
|
ggml_type type_v,
|
|
bool v_trans,
|
|
bool offload,
|
|
bool unified,
|
|
uint32_t kv_size,
|
|
uint32_t n_seq_max,
|
|
uint32_t n_pad,
|
|
uint32_t n_swa,
|
|
llama_swa_type swa_type);
|
|
|
|
~llama_kv_cache_unified() = default;
|
|
|
|
//
|
|
// llama_memory_i
|
|
//
|
|
|
|
llama_memory_context_ptr init_batch(
|
|
llama_batch_allocr & balloc,
|
|
uint32_t n_ubatch,
|
|
bool embd_all) override;
|
|
|
|
llama_memory_context_ptr init_full() override;
|
|
|
|
llama_memory_context_ptr init_update(llama_context * lctx, bool optimize) override;
|
|
|
|
bool get_can_shift() const override;
|
|
|
|
void clear(bool data) override;
|
|
|
|
bool seq_rm (llama_seq_id seq_id, llama_pos p0, llama_pos p1) override;
|
|
void seq_cp (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) override;
|
|
void seq_keep(llama_seq_id seq_id) override;
|
|
void seq_add (llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos shift) override;
|
|
void seq_div (llama_seq_id seq_id, llama_pos p0, llama_pos p1, int d) override;
|
|
|
|
llama_pos seq_pos_min(llama_seq_id seq_id) const override;
|
|
llama_pos seq_pos_max(llama_seq_id seq_id) const override;
|
|
|
|
// state write/load
|
|
|
|
void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1) const override;
|
|
void state_read (llama_io_read_i & io, llama_seq_id seq_id = -1) override;
|
|
|
|
//
|
|
// llama_kv_cache_unified specific API
|
|
//
|
|
|
|
uint32_t get_size() const;
|
|
uint32_t get_n_stream() const;
|
|
|
|
bool get_has_shift() const;
|
|
|
|
//
|
|
// graph_build API
|
|
//
|
|
|
|
uint32_t get_n_kv() const;
|
|
|
|
// TODO: temporary
|
|
bool get_supports_set_rows() const;
|
|
|
|
// get views of the current state of the cache
|
|
ggml_tensor * get_k(ggml_context * ctx, int32_t il, uint32_t n_kv, const slot_info & sinfo) const;
|
|
ggml_tensor * get_v(ggml_context * ctx, int32_t il, uint32_t n_kv, const slot_info & sinfo) const;
|
|
|
|
// store k_cur and v_cur in the cache based on the provided head location
|
|
ggml_tensor * cpy_k(ggml_context * ctx, ggml_tensor * k_cur, ggml_tensor * k_idxs, int32_t il, const slot_info & sinfo) const;
|
|
ggml_tensor * cpy_v(ggml_context * ctx, ggml_tensor * v_cur, ggml_tensor * v_idxs, int32_t il, const slot_info & sinfo) const;
|
|
|
|
//
|
|
// preparation API
|
|
//
|
|
|
|
// find places for the provided ubatches in the cache, returns the slot infos
|
|
// return empty vector on failure
|
|
slot_info_vec_t prepare(const std::vector<llama_ubatch> & ubatches);
|
|
|
|
bool update(llama_context * lctx, bool do_shift, const defrag_info & dinfo, const stream_copy_info & sc_info);
|
|
|
|
// find a slot of kv cells that can hold the ubatch
|
|
// if cont == true, then the slot must be continuous
|
|
// return empty slot_info on failure
|
|
slot_info find_slot(const llama_ubatch & ubatch, bool cont) const;
|
|
|
|
// emplace the ubatch context into slot: [sinfo.idxs[0...ubatch.n_tokens - 1]]
|
|
void apply_ubatch(const slot_info & sinfo, const llama_ubatch & ubatch);
|
|
|
|
//
|
|
// input API
|
|
//
|
|
|
|
ggml_tensor * build_input_k_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const;
|
|
ggml_tensor * build_input_v_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const;
|
|
|
|
void set_input_k_idxs(ggml_tensor * dst, const llama_ubatch * ubatch, const slot_info & sinfo) const;
|
|
void set_input_v_idxs(ggml_tensor * dst, const llama_ubatch * ubatch, const slot_info & sinfo) const;
|
|
|
|
void set_input_k_shift(ggml_tensor * dst) const;
|
|
|
|
void set_input_kq_mask (ggml_tensor * dst, const llama_ubatch * ubatch, bool causal_attn) const;
|
|
void set_input_pos_bucket(ggml_tensor * dst, const llama_ubatch * ubatch) const;
|
|
|
|
private:
|
|
const llama_model & model;
|
|
const llama_hparams & hparams;
|
|
|
|
struct kv_layer {
|
|
// layer index in the model
|
|
// note: can be different from the layer index in the KV cache
|
|
uint32_t il;
|
|
|
|
ggml_tensor * k;
|
|
ggml_tensor * v;
|
|
|
|
std::vector<ggml_tensor *> k_stream;
|
|
std::vector<ggml_tensor *> v_stream;
|
|
};
|
|
|
|
bool v_trans = true; // the value tensor is transposed
|
|
|
|
const uint32_t n_seq_max = 1;
|
|
const uint32_t n_stream = 1;
|
|
|
|
// required padding
|
|
const uint32_t n_pad = 1;
|
|
|
|
// SWA
|
|
const uint32_t n_swa = 0;
|
|
|
|
// env: LLAMA_KV_CACHE_DEBUG
|
|
int debug = 0;
|
|
|
|
// env: LLAMA_SET_ROWS (temporary)
|
|
// ref: https://github.com/ggml-org/llama.cpp/pull/14285
|
|
bool supports_set_rows = false;
|
|
|
|
const llama_swa_type swa_type = LLAMA_SWA_TYPE_NONE;
|
|
|
|
std::vector<ggml_context_ptr> ctxs;
|
|
std::vector<ggml_backend_buffer_ptr> bufs;
|
|
|
|
// the current index from where we start searching for a free slot in the ring buffer of KV cells (see find_slot())
|
|
// note: this is not part of the KV state and it's only used to speed-up the find_slot() method
|
|
std::vector<uint32_t> v_heads;
|
|
|
|
std::vector<llama_kv_cells_unified> v_cells;
|
|
|
|
// maps from a sequence id to a stream id
|
|
std::vector<uint32_t> seq_to_stream;
|
|
|
|
// pending stream copies that will be applied during the next update
|
|
stream_copy_info sc_info;
|
|
|
|
std::vector<kv_layer> layers;
|
|
|
|
// model layer id -> KV cache layer id
|
|
std::unordered_map<int32_t, int32_t> map_layer_ids;
|
|
|
|
// return non-empty vector if cells have been moved
|
|
defrag_info defrag_prepare(int32_t n_max_nodes) const;
|
|
|
|
size_t total_size() const;
|
|
|
|
size_t size_k_bytes() const;
|
|
size_t size_v_bytes() const;
|
|
|
|
bool is_masked_swa(llama_pos p0, llama_pos p1) const;
|
|
|
|
ggml_tensor * build_rope_shift(
|
|
const llama_cparams & cparams,
|
|
ggml_context * ctx,
|
|
ggml_tensor * cur,
|
|
ggml_tensor * shift,
|
|
ggml_tensor * factors,
|
|
float freq_base,
|
|
float freq_scale) const;
|
|
|
|
ggml_cgraph * build_graph_shift(
|
|
llm_graph_result * res,
|
|
llama_context * lctx) const;
|
|
|
|
ggml_cgraph * build_graph_defrag(
|
|
llm_graph_result * res,
|
|
llama_context * lctx,
|
|
const defrag_info & dinfo) const;
|
|
|
|
struct cell_ranges_t {
|
|
uint32_t strm;
|
|
|
|
std::vector<std::pair<uint32_t, uint32_t>> data; // ranges, from inclusive, to exclusive
|
|
};
|
|
|
|
void state_write_meta(llama_io_write_i & io, const cell_ranges_t & cr, llama_seq_id seq_id = -1) const;
|
|
void state_write_data(llama_io_write_i & io, const cell_ranges_t & cr) const;
|
|
|
|
bool state_read_meta(llama_io_read_i & io, uint32_t strm, uint32_t cell_count, llama_seq_id dest_seq_id = -1);
|
|
bool state_read_data(llama_io_read_i & io, uint32_t strm, uint32_t cell_count);
|
|
};
|
|
|
|
class llama_kv_cache_unified_context : public llama_memory_context_i {
|
|
public:
|
|
// some shorthands
|
|
using slot_info_vec_t = llama_kv_cache_unified::slot_info_vec_t;
|
|
using defrag_info = llama_kv_cache_unified::defrag_info;
|
|
using stream_copy_info = llama_kv_cache_unified::stream_copy_info;
|
|
|
|
// used for errors
|
|
llama_kv_cache_unified_context(llama_memory_status status);
|
|
|
|
// used to create a full-cache context
|
|
llama_kv_cache_unified_context(
|
|
llama_kv_cache_unified * kv);
|
|
|
|
// used to create an update context
|
|
llama_kv_cache_unified_context(
|
|
llama_kv_cache_unified * kv,
|
|
llama_context * lctx,
|
|
bool do_shift,
|
|
defrag_info dinfo,
|
|
stream_copy_info sc_info);
|
|
|
|
// used to create a batch procesing context from a batch
|
|
llama_kv_cache_unified_context(
|
|
llama_kv_cache_unified * kv,
|
|
slot_info_vec_t sinfos,
|
|
std::vector<llama_ubatch> ubatches);
|
|
|
|
virtual ~llama_kv_cache_unified_context();
|
|
|
|
//
|
|
// llama_memory_context_i
|
|
//
|
|
|
|
bool next() override;
|
|
bool apply() override;
|
|
|
|
llama_memory_status get_status() const override;
|
|
const llama_ubatch & get_ubatch() const override;
|
|
|
|
//
|
|
// llama_kv_cache_unified_context specific API
|
|
//
|
|
|
|
uint32_t get_n_kv() const;
|
|
|
|
// TODO: temporary
|
|
bool get_supports_set_rows() const;
|
|
|
|
// get views of the current state of the cache
|
|
ggml_tensor * get_k(ggml_context * ctx, int32_t il) const;
|
|
ggml_tensor * get_v(ggml_context * ctx, int32_t il) const;
|
|
|
|
// store k_cur and v_cur in the cache based on the provided head location
|
|
ggml_tensor * cpy_k(ggml_context * ctx, ggml_tensor * k_cur, ggml_tensor * k_idxs, int32_t il) const;
|
|
ggml_tensor * cpy_v(ggml_context * ctx, ggml_tensor * v_cur, ggml_tensor * v_idxs, int32_t il) const;
|
|
|
|
ggml_tensor * build_input_k_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const;
|
|
ggml_tensor * build_input_v_idxs(ggml_context * ctx, const llama_ubatch & ubatch) const;
|
|
|
|
void set_input_k_idxs(ggml_tensor * dst, const llama_ubatch * ubatch) const;
|
|
void set_input_v_idxs(ggml_tensor * dst, const llama_ubatch * ubatch) const;
|
|
|
|
void set_input_k_shift (ggml_tensor * dst) const;
|
|
void set_input_kq_mask (ggml_tensor * dst, const llama_ubatch * ubatch, bool causal_attn) const;
|
|
void set_input_pos_bucket(ggml_tensor * dst, const llama_ubatch * ubatch) const;
|
|
|
|
private:
|
|
llama_memory_status status;
|
|
|
|
llama_kv_cache_unified * kv;
|
|
llama_context * lctx;
|
|
|
|
//
|
|
// update context
|
|
//
|
|
|
|
bool do_shift = false;
|
|
|
|
defrag_info dinfo;
|
|
|
|
stream_copy_info sc_info;
|
|
|
|
//
|
|
// batch processing context
|
|
//
|
|
|
|
// the index of the cur ubatch to process
|
|
size_t i_cur = 0;
|
|
|
|
slot_info_vec_t sinfos;
|
|
|
|
std::vector<llama_ubatch> ubatches;
|
|
|
|
//
|
|
// data needed for building the compute graph for the current ubatch:
|
|
//
|
|
|
|
// a heuristic, to avoid attending the full cache if it is not yet utilized
|
|
// as the cache gets filled, the benefit from this heuristic disappears
|
|
int32_t n_kv;
|
|
};
|