llama : add llama_vocab, functions -> methods, naming (#11110)
* llama : functions -> methods (#11110) * llama : add struct llama_vocab to the API (#11156) ggml-ci * hparams : move vocab params to llama_vocab (#11159) ggml-ci * vocab : more pimpl (#11165) ggml-ci * vocab : minor tokenization optimizations (#11160) ggml-ci Co-authored-by: Diego Devesa <slarengh@gmail.com> * lora : update API names (#11167) ggml-ci * llama : update API names to use correct prefix (#11174) * llama : update API names to use correct prefix ggml-ci * cont ggml-ci * cont ggml-ci * minor [no ci] * vocab : llama_vocab_add_[be]os -> llama_vocab_get_add_[be]os (#11174) ggml-ci * vocab : llama_vocab_n_vocab -> llama_vocab_n_tokens (#11174) ggml-ci --------- Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
Georgi Gerganov
GitHub
parent
c05e8c9934
commit
afa8a9ec9b
68 changed files with 5855 additions and 5400 deletions
|
|
@ -4,79 +4,80 @@
|
|||
#include "llama-arch.h"
|
||||
#include "llama-hparams.h"
|
||||
#include "llama-vocab.h"
|
||||
#include "llama-mmap.h"
|
||||
|
||||
#include "ggml-cpp.h"
|
||||
|
||||
#include <memory>
|
||||
#include <string>
|
||||
#include <unordered_map>
|
||||
#include <vector>
|
||||
|
||||
struct llama_model_loader;
|
||||
|
||||
// available models
|
||||
// TODO: this enum does not follow the enum naming convention
|
||||
enum llm_type {
|
||||
MODEL_UNKNOWN,
|
||||
MODEL_14M,
|
||||
MODEL_17M,
|
||||
MODEL_22M,
|
||||
MODEL_33M,
|
||||
MODEL_60M,
|
||||
MODEL_70M,
|
||||
MODEL_80M,
|
||||
MODEL_109M,
|
||||
MODEL_137M,
|
||||
MODEL_160M,
|
||||
MODEL_220M,
|
||||
MODEL_250M,
|
||||
MODEL_270M,
|
||||
MODEL_335M,
|
||||
MODEL_410M,
|
||||
MODEL_450M,
|
||||
MODEL_770M,
|
||||
MODEL_780M,
|
||||
MODEL_0_5B,
|
||||
MODEL_1B,
|
||||
MODEL_1_3B,
|
||||
MODEL_1_4B,
|
||||
MODEL_1_5B,
|
||||
MODEL_1_6B,
|
||||
MODEL_2B,
|
||||
MODEL_2_8B,
|
||||
MODEL_3B,
|
||||
MODEL_4B,
|
||||
MODEL_6B,
|
||||
MODEL_6_9B,
|
||||
MODEL_7B,
|
||||
MODEL_8B,
|
||||
MODEL_9B,
|
||||
MODEL_11B,
|
||||
MODEL_12B,
|
||||
MODEL_13B,
|
||||
MODEL_14B,
|
||||
MODEL_15B,
|
||||
MODEL_16B,
|
||||
MODEL_20B,
|
||||
MODEL_30B,
|
||||
MODEL_32B,
|
||||
MODEL_34B,
|
||||
MODEL_35B,
|
||||
MODEL_40B,
|
||||
MODEL_65B,
|
||||
MODEL_70B,
|
||||
MODEL_236B,
|
||||
MODEL_314B,
|
||||
MODEL_671B,
|
||||
MODEL_SMALL,
|
||||
MODEL_MEDIUM,
|
||||
MODEL_LARGE,
|
||||
MODEL_XL,
|
||||
MODEL_A1_7B,
|
||||
MODEL_A2_7B,
|
||||
MODEL_8x7B,
|
||||
MODEL_8x22B,
|
||||
MODEL_16x12B,
|
||||
MODEL_16x3_8B,
|
||||
MODEL_10B_128x3_66B,
|
||||
MODEL_57B_A14B,
|
||||
MODEL_27B,
|
||||
LLM_TYPE_UNKNOWN,
|
||||
LLM_TYPE_14M,
|
||||
LLM_TYPE_17M,
|
||||
LLM_TYPE_22M,
|
||||
LLM_TYPE_33M,
|
||||
LLM_TYPE_60M,
|
||||
LLM_TYPE_70M,
|
||||
LLM_TYPE_80M,
|
||||
LLM_TYPE_109M,
|
||||
LLM_TYPE_137M,
|
||||
LLM_TYPE_160M,
|
||||
LLM_TYPE_220M,
|
||||
LLM_TYPE_250M,
|
||||
LLM_TYPE_270M,
|
||||
LLM_TYPE_335M,
|
||||
LLM_TYPE_410M,
|
||||
LLM_TYPE_450M,
|
||||
LLM_TYPE_770M,
|
||||
LLM_TYPE_780M,
|
||||
LLM_TYPE_0_5B,
|
||||
LLM_TYPE_1B,
|
||||
LLM_TYPE_1_3B,
|
||||
LLM_TYPE_1_4B,
|
||||
LLM_TYPE_1_5B,
|
||||
LLM_TYPE_1_6B,
|
||||
LLM_TYPE_2B,
|
||||
LLM_TYPE_2_8B,
|
||||
LLM_TYPE_3B,
|
||||
LLM_TYPE_4B,
|
||||
LLM_TYPE_6B,
|
||||
LLM_TYPE_6_9B,
|
||||
LLM_TYPE_7B,
|
||||
LLM_TYPE_8B,
|
||||
LLM_TYPE_9B,
|
||||
LLM_TYPE_11B,
|
||||
LLM_TYPE_12B,
|
||||
LLM_TYPE_13B,
|
||||
LLM_TYPE_14B,
|
||||
LLM_TYPE_15B,
|
||||
LLM_TYPE_16B,
|
||||
LLM_TYPE_20B,
|
||||
LLM_TYPE_30B,
|
||||
LLM_TYPE_32B,
|
||||
LLM_TYPE_34B,
|
||||
LLM_TYPE_35B,
|
||||
LLM_TYPE_40B,
|
||||
LLM_TYPE_65B,
|
||||
LLM_TYPE_70B,
|
||||
LLM_TYPE_236B,
|
||||
LLM_TYPE_314B,
|
||||
LLM_TYPE_671B,
|
||||
LLM_TYPE_SMALL,
|
||||
LLM_TYPE_MEDIUM,
|
||||
LLM_TYPE_LARGE,
|
||||
LLM_TYPE_XL,
|
||||
LLM_TYPE_A1_7B,
|
||||
LLM_TYPE_A2_7B,
|
||||
LLM_TYPE_8x7B,
|
||||
LLM_TYPE_8x22B,
|
||||
LLM_TYPE_16x12B,
|
||||
LLM_TYPE_16x3_8B,
|
||||
LLM_TYPE_10B_128x3_66B,
|
||||
LLM_TYPE_57B_A14B,
|
||||
LLM_TYPE_27B,
|
||||
};
|
||||
|
||||
struct llama_layer_posnet {
|
||||
|
|
@ -286,11 +287,9 @@ struct llama_layer {
|
|||
};
|
||||
|
||||
struct llama_model {
|
||||
llm_type type = MODEL_UNKNOWN;
|
||||
llm_type type = LLM_TYPE_UNKNOWN;
|
||||
llm_arch arch = LLM_ARCH_UNKNOWN;
|
||||
|
||||
llama_ftype ftype = LLAMA_FTYPE_ALL_F32;
|
||||
|
||||
std::string name = "n/a";
|
||||
|
||||
llama_hparams hparams = {};
|
||||
|
|
@ -319,78 +318,55 @@ struct llama_model {
|
|||
|
||||
std::vector<llama_layer> layers;
|
||||
|
||||
llama_model_params params;
|
||||
|
||||
// gguf metadata
|
||||
std::unordered_map<std::string, std::string> gguf_kv;
|
||||
|
||||
llama_split_mode split_mode;
|
||||
int main_gpu;
|
||||
int n_gpu_layers;
|
||||
|
||||
std::vector<std::string> rpc_servers;
|
||||
|
||||
// list of devices used in this model
|
||||
std::vector<ggml_backend_dev_t> devices;
|
||||
|
||||
|
||||
// lists of buffer types used for each layer
|
||||
using buft_list_t = std::vector<std::pair<ggml_backend_dev_t, ggml_backend_buffer_type_t>>;
|
||||
buft_list_t cpu_buft_list;
|
||||
std::map<ggml_backend_dev_t, buft_list_t> gpu_buft_list;
|
||||
|
||||
struct layer_dev {
|
||||
ggml_backend_dev_t dev;
|
||||
buft_list_t * buft_list;
|
||||
};
|
||||
|
||||
layer_dev dev_input = {};
|
||||
layer_dev dev_output = {};
|
||||
std::vector<layer_dev> dev_layer;
|
||||
|
||||
// contexts where the model tensors metadata is stored
|
||||
std::vector<ggml_context_ptr> ctxs;
|
||||
|
||||
// the model memory buffers for the tensor data
|
||||
std::vector<ggml_backend_buffer_ptr> bufs;
|
||||
|
||||
// model memory mapped files
|
||||
llama_mmaps mappings;
|
||||
|
||||
// objects representing data potentially being locked in memory
|
||||
llama_mlocks mlock_bufs;
|
||||
llama_mlocks mlock_mmaps;
|
||||
|
||||
// for quantize-stats only
|
||||
std::vector<std::pair<std::string, struct ggml_tensor *>> tensors_by_name;
|
||||
|
||||
int64_t t_load_us = 0;
|
||||
int64_t t_start_us = 0;
|
||||
|
||||
// total number of parameters in the model
|
||||
uint64_t n_elements = 0;
|
||||
explicit llama_model(const struct llama_model_params & params);
|
||||
~llama_model();
|
||||
|
||||
// total size of all the tensors in the model in bytes
|
||||
size_t n_bytes = 0;
|
||||
void load_stats (llama_model_loader & ml);
|
||||
void load_arch (llama_model_loader & ml);
|
||||
void load_hparams(llama_model_loader & ml);
|
||||
void load_vocab (llama_model_loader & ml);
|
||||
bool load_tensors(llama_model_loader & ml); // returns false if cancelled by progress_callback
|
||||
|
||||
std::string arch_name() const;
|
||||
std::string type_name() const;
|
||||
|
||||
std::string desc() const;
|
||||
|
||||
size_t size() const;
|
||||
size_t max_nodes() const;
|
||||
size_t n_devices() const;
|
||||
|
||||
// total number of parameters in the model
|
||||
uint64_t n_elements() const;
|
||||
|
||||
void print_info() const;
|
||||
|
||||
ggml_backend_dev_t dev_layer(int il) const;
|
||||
ggml_backend_dev_t dev_output() const;
|
||||
|
||||
ggml_backend_buffer_type_t select_buft(int il) const;
|
||||
|
||||
const struct ggml_tensor * get_tensor(const char * name) const;
|
||||
|
||||
private:
|
||||
struct impl;
|
||||
std::unique_ptr<impl> pimpl;
|
||||
};
|
||||
|
||||
const char * llm_type_name(llm_type type);
|
||||
|
||||
std::string llama_model_arch_name (const llama_model & model);
|
||||
std::string llama_model_type_name (const llama_model & model);
|
||||
std::string llama_model_ftype_name(const llama_model & model);
|
||||
|
||||
// used by llama_adapter_cvec
|
||||
ggml_backend_buffer_type_t llama_model_select_buft(const llama_model & model, int il);
|
||||
|
||||
// used by llama_adapter_lora
|
||||
struct ggml_tensor * llama_model_get_tensor(const struct llama_model & model, const char * name);
|
||||
|
||||
size_t llama_model_max_nodes(const llama_model & model);
|
||||
|
||||
struct llama_model_loader;
|
||||
|
||||
// TODO: become llama_model methods
|
||||
void llm_load_stats (llama_model_loader & ml, llama_model & model);
|
||||
void llm_load_arch (llama_model_loader & ml, llama_model & model);
|
||||
void llm_load_hparams (llama_model_loader & ml, llama_model & model);
|
||||
void llm_load_vocab (llama_model_loader & ml, llama_model & model);
|
||||
void llm_load_print_meta(llama_model_loader & ml, llama_model & model);
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue