matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-12 00:37:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update vendored code to commit 46e3556 (#8308)

Browse Source
This commit is contained in:
Jeffrey Morgan
2025-01-08 11:22:01 -08:00
committed by GitHub
parent 57f038ec7b
commit 1deafd8254
305 changed files with 16048 additions and 12926 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
llama/common.h vendored
View File

@@ -1,5 +1,5 @@
/**
* llama.cpp - commit ba1cb19cdd0d92e012e0f6e009e0620f854b6afd - do not edit this file
* llama.cpp - commit 46e3556e01b824e52395fb050b29804b6cff2a7c - do not edit this file
*
* MIT License
*
@@ -28,7 +28,7 @@
#pragma once
#include "llama.h"
#include "llama-cpp.h"
#include <string>
#include <vector>
@@ -53,10 +53,8 @@
struct common_lora_adapter_info {
std::string path;
float scale;
};
struct common_lora_adapter_container : common_lora_adapter_info {
struct llama_lora_adapter * adapter;
struct llama_lora_adapter * ptr;
};
using llama_tokens = std::vector<llama_token>;
@@ -106,6 +104,7 @@ enum llama_example {
LLAMA_EXAMPLE_LLAVA,
LLAMA_EXAMPLE_LOOKUP,
LLAMA_EXAMPLE_PARALLEL,
LLAMA_EXAMPLE_TTS,
LLAMA_EXAMPLE_COUNT,
};
@@ -121,6 +120,7 @@ enum common_sampler_type {
COMMON_SAMPLER_TYPE_TEMPERATURE = 7,
COMMON_SAMPLER_TYPE_XTC = 8,
COMMON_SAMPLER_TYPE_INFILL = 9,
COMMON_SAMPLER_TYPE_PENALTIES = 10,
};
// dimensionality reduction methods, used by cvector-generator
@@ -156,7 +156,6 @@ struct common_params_sampling {
int32_t mirostat = 0; // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
float mirostat_tau = 5.00f; // target entropy
float mirostat_eta = 0.10f; // learning rate
bool penalize_nl = false; // consider newlines as a repeatable token
bool ignore_eos = false;
bool no_perf = false; // disable performance metrics
bool timing_per_token = false;
@@ -165,6 +164,7 @@ struct common_params_sampling {
std::vector<enum common_sampler_type> samplers = {
COMMON_SAMPLER_TYPE_PENALTIES,
COMMON_SAMPLER_TYPE_DRY,
COMMON_SAMPLER_TYPE_TOP_K,
COMMON_SAMPLER_TYPE_TYPICAL_P,
@@ -184,6 +184,7 @@ struct common_params_sampling {
struct common_params_speculative {
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
int32_t n_ctx = 0; // draft context size
int32_t n_max = 16; // maximum number of tokens to draft during speculative decoding
int32_t n_min = 5; // minimum number of draft tokens to use for speculative decoding
@@ -197,6 +198,14 @@ struct common_params_speculative {
std::string model = ""; // draft model for speculative decoding // NOLINT
};
struct common_params_vocoder {
std::string hf_repo = ""; // HF repo // NOLINT
std::string hf_file = ""; // HF file // NOLINT
std::string model = ""; // model path // NOLINT
std::string model_url = ""; // model url to download // NOLINT
};
struct common_params {
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 4096; // context size
@@ -219,11 +228,13 @@ struct common_params {
float defrag_thold = 0.1f; // KV cache defragmentation threshold
// offload params
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[128] = {0}; // how split tensors should be distributed across GPUs
enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
int32_t n_gpu_layers = -1; // number of layers to store in VRAM (-1 - use default)
int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors
float tensor_split[128] = {0}; // how split tensors should be distributed across GPUs
enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
struct cpu_params cpuparams;
struct cpu_params cpuparams_batch;
@@ -237,8 +248,9 @@ struct common_params {
enum llama_pooling_type pooling_type = LLAMA_POOLING_TYPE_UNSPECIFIED; // pooling type for embeddings
enum llama_attention_type attention_type = LLAMA_ATTENTION_TYPE_UNSPECIFIED; // attention type for embeddings
struct common_params_sampling sampling;
struct common_params_sampling sampling;
struct common_params_speculative speculative;
struct common_params_vocoder vocoder;
std::string model = ""; // model path // NOLINT
std::string model_alias = ""; // model alias // NOLINT
@@ -490,10 +502,12 @@ std::string fs_get_cache_file(const std::string & filename);
// Model utils
//
// note: defines object's lifetime
struct common_init_result {
struct llama_model * model = nullptr;
struct llama_context * context = nullptr;
std::vector<common_lora_adapter_container> lora_adapters;
llama_model_ptr model;
llama_context_ptr context;
std::vector<llama_lora_adapter_ptr> lora;
};
struct common_init_result common_init_from_params(common_params & params);
@@ -515,7 +529,7 @@ struct llama_model * common_load_model_from_hf(
const struct llama_model_params & params);
// clear LoRA adapters from context, then apply new list of adapters
void common_lora_adapters_apply(struct llama_context * ctx, std::vector<common_lora_adapter_container> & lora_adapters);
void common_lora_adapters_apply(struct llama_context * ctx, std::vector<common_lora_adapter_info> & lora);
//
// Batch utils
@@ -583,6 +597,9 @@ struct common_chat_msg {
std::string content;
};
// Get the built-in chat template for the model. Return empty string if not present.
std::string common_get_builtin_chat_template(const struct llama_model * model);
// Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid
bool common_chat_verify_template(const std::string & tmpl);
@@ -619,7 +636,8 @@ void common_kv_cache_dump_view_seqs(const llama_kv_cache_view & view, int row_si
// Embedding utils
//
void common_embd_normalize(const float * inp, float * out, int n, int embd_norm = 2);
// TODO: repace embd_norm with an enum
void common_embd_normalize(const float * inp, float * out, int n, int embd_norm);
float common_embd_similarity_cos(const float * embd1, const float * embd2, int n);
@@ -648,6 +666,10 @@ common_control_vector_data common_control_vector_load(const std::vector<common_c
// Split utils
//
static const char * const LLM_KV_SPLIT_NO = "split.no";
static const char * const LLM_KV_SPLIT_COUNT = "split.count";
static const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
namespace {
const char * const LLM_KV_SPLIT_NO = "split.no";
const char * const LLM_KV_SPLIT_COUNT = "split.count";
const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
}