Felipe97/llama-cpp-compiled

1#pragma once2 3#include "llama.h"4#include "llama-graph.h"5 6#include <map>7#include <memory>8#include <functional>9 10struct llama_ubatch;11 12class llama_batch_allocr;13 14class llama_io_write_i;15class llama_io_read_i;16 17struct llama_memory_params {18    // kv cache19    ggml_type type_k;20    ggml_type type_v;21 22    // use full-size SWA cache23    bool swa_full;24 25    llama_context_type ctx_type;26 27    llama_memory_t mem_other;28};29 30enum llama_memory_status {31    LLAMA_MEMORY_STATUS_SUCCESS = 0,32    LLAMA_MEMORY_STATUS_NO_UPDATE,33    LLAMA_MEMORY_STATUS_FAILED_PREPARE,34    LLAMA_MEMORY_STATUS_FAILED_COMPUTE,35};36 37// helper function for combining the status of two memory contexts38// useful for implementing hybrid memory types (e.g. iSWA)39llama_memory_status llama_memory_status_combine(llama_memory_status s0, llama_memory_status s1);40 41// helper function for checking if a memory status indicates a failure42bool llama_memory_status_is_fail(llama_memory_status status);43 44// the interface for managing the memory context during batch processing45// this interface is implemented per memory type. see:46//   - llama_kv_cache_context47//   - llama_kv_cache_iswa_context48//   ...49//50// the only method that should mutate the memory and the memory context is llama_memory_i::apply()51struct llama_memory_context_i {52    virtual ~llama_memory_context_i() = default;53 54    // consume the current ubatch from the context and proceed to the next one55    // return false if we are done56    virtual bool next() = 0;57 58    // apply the memory state for the current ubatch to the memory object59    // return false on failure60    virtual bool apply() = 0;61 62    // get the current ubatch63    virtual const llama_ubatch & get_ubatch() const = 0;64 65    // get the status of the memory context - used for error handling and checking if any updates would be applied66    virtual llama_memory_status get_status() const = 0;67};68 69using llama_memory_context_ptr = std::unique_ptr<llama_memory_context_i>;70 71// general concept of LLM memory72// the KV cache is a type of LLM memory, but there can be other types73struct llama_memory_i {74    // this callback is used to filter out layers that should not be included in the cache75    using layer_filter_cb = std::function<bool(int32_t il)>;76 77    // this callback is used to specify which layers should reuse memory from other layers78    // return negative value to indicate that the layer il should not reuse memory79    using layer_reuse_cb = std::function<int32_t(int32_t il)>;80 81    using layer_share_cb = std::function<int32_t(int32_t il)>;82 83    virtual ~llama_memory_i() = default;84 85    // split the input batch into a set of ubatches and verify that they can fit into the cache86    // return a context object containing the ubatches and memory state required to process them87    // check the llama_memory_context_i::get_status() for the result88    virtual llama_memory_context_ptr init_batch(89            llama_batch_allocr & balloc,90            uint32_t n_ubatch,91            bool embd_all) = 0;92 93    // simulate full cache, used for allocating worst-case compute buffers94    virtual llama_memory_context_ptr init_full() = 0;95 96    // prepare for any pending memory updates, such as shifts, copies, etc.97    // status == LLAMA_MEMORY_STATUS_NO_UPDATE if there is nothing to update98    virtual llama_memory_context_ptr init_update(llama_context * lctx, bool optimize) = 0;99 100    // getters101    virtual bool get_can_shift() const = 0;102 103    //104    // ops105    //106 107    // if data == true, the data buffers will also be cleared together with the metadata108    virtual void clear(bool data) = 0;109 110    virtual bool seq_rm  (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1) = 0;111    virtual void seq_cp  (llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) = 0;112    virtual void seq_keep(llama_seq_id seq_id) = 0;113    virtual void seq_add (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1, llama_pos shift) = 0;114    virtual void seq_div (llama_seq_id seq_id,                              llama_pos p0, llama_pos p1, int d) = 0;115 116    virtual llama_pos seq_pos_min(llama_seq_id seq_id) const = 0;117    virtual llama_pos seq_pos_max(llama_seq_id seq_id) const = 0;118 119    virtual std::map<ggml_backend_buffer_type_t, size_t> memory_breakdown() const = 0;120 121    //122    // state write/read123    //124 125    virtual void state_write(llama_io_write_i & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) const = 0;126    virtual void state_read (llama_io_read_i  & io, llama_seq_id seq_id = -1, llama_state_seq_flags flags = 0) = 0;127};128 129using llama_memory_ptr = std::unique_ptr<llama_memory_i>;130