vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : refactor src/llama.cpp (#10902)

Browse source 
* llama : scatter llama.cpp into multiple modules (wip)

* llama : control-vector -> adapter

* llama : arch

* llama : mmap

ggml-ci

* ci : remove BUILD_SHARED_LIBS=OFF

ggml-ci

* llama : arch (cont)

ggml-ci

* llama : chat

ggml-ci

* llama : model

ggml-ci

* llama : hparams

ggml-ci

* llama : adapter

ggml-ci

* examples : fix

ggml-ci

* rebase

ggml-ci

* minor

* llama : kv cache

ggml-ci

* llama : impl

ggml-ci

* llama : batch

ggml-ci

* cont

ggml-ci

* llama : context

ggml-ci

* minor

* llama : context (cont)

ggml-ci

* llama : model loader

ggml-ci

* common : update lora

ggml-ci

* llama : quant

ggml-ci

* llama : quant (cont)

ggml-ci

* minor [no ci]
This commit is contained in:
Georgi Gerganov 2025-01-03 10:18:53 +02:00 • committed by GitHub
parent 2f0ee84b9b
commit f66f582927
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
61 changed files with 12193 additions and 11649 deletions
Download patch file Download diff file Expand all files Collapse all files

4
common/arg.cpp
View file

@ -1512,7 +1512,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
{"--lora"}, "FNAME",
"path to LoRA adapter (can be repeated to use multiple adapters)",
[](common_params & params, const std::string & value) {
params.lora_adapters.push_back({ std::string(value), 1.0 });
params.lora_adapters.push_back({ std::string(value), 1.0, nullptr });
}
// we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
@ -1520,7 +1520,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
{"--lora-scaled"}, "FNAME", "SCALE",
"path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters)",
[](common_params & params, const std::string & fname, const std::string & scale) {
params.lora_adapters.push_back({ fname, std::stof(scale) });
params.lora_adapters.push_back({ fname, std::stof(scale), nullptr });
}
// we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));

25
common/common.cpp
View file

@ -922,20 +922,21 @@ struct common_init_result common_init_from_params(common_params & params) {
// load and optionally apply lora adapters
for (auto & la : params.lora_adapters) {
common_lora_adapter_container loaded_la;
loaded_la.path = la.path;
loaded_la.scale = la.scale;
loaded_la.adapter = llama_lora_adapter_init(model, la.path.c_str());
if (loaded_la.adapter == nullptr) {
llama_lora_adapter_ptr lora;
lora.reset(llama_lora_adapter_init(model, la.path.c_str()));
if (lora == nullptr) {
LOG_ERR("%s: failed to apply lora adapter '%s'\n", __func__, la.path.c_str());
llama_free(lctx);
llama_free_model(model);
return iparams;
}
iparams.lora_adapters.push_back(loaded_la); // copy to list of loaded adapters
la.ptr = lora.get();
iparams.lora.emplace_back(std::move(lora)); // copy to list of loaded adapters
}
if (!params.lora_init_without_apply) {
common_lora_adapters_apply(lctx, iparams.lora_adapters);
common_lora_adapters_apply(lctx, params.lora_adapters);
}
if (params.sampling.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) {
@ -996,17 +997,17 @@ struct common_init_result common_init_from_params(common_params & params) {
llama_perf_context_reset(lctx);
}
iparams.model = model;
iparams.context = lctx;
iparams.model.reset(model);
iparams.context.reset(lctx);
return iparams;
}
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) {
llama_lora_adapter_clear(ctx);
for (auto & la : lora_adapters) {
for (auto & la : lora) {
if (la.scale != 0.0f) {
llama_lora_adapter_set(ctx, la.adapter, la.scale);
llama_lora_adapter_set(ctx, la.ptr, la.scale);
}
}
}

26
common/common.h
View file

@ -2,7 +2,7 @@
#pragma once
#include "llama.h"
#include "llama-cpp.h"
#include <string>
#include <vector>
@ -27,10 +27,8 @@
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>;
@ -478,10 +476,12 @@ 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);
@ -503,7 +503,7 @@ 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
@ -640,6 +640,10 @@ 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";
}