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

style : minor fixes, mostly indentations

Browse source
This commit is contained in:
Georgi Gerganov 2023-07-15 09:57:35 +03:00
parent da730c53bf
commit d0b6c942fc
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
4 changed files with 65 additions and 53 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml-metal.m
View file

@ -885,7 +885,8 @@ void ggml_metal_graph_compute(
const int n_past = ((int32_t *)(src1->data))[0];
float freq_base, freq_scale;
float freq_base;
float freq_scale;
memcpy(&freq_base, (int32_t *) src1->data + 4, sizeof(float));
memcpy(&freq_scale, (int32_t *) src1->data + 5, sizeof(float));

8
ggml.c
View file

@ -12084,11 +12084,13 @@ static void ggml_compute_forward_rope_f32(
return;
}
float freq_base;
float freq_scale;
const int n_past = ((int32_t *) src1->data)[0];
const int n_dims = ((int32_t *) src1->data)[1];
const int mode = ((int32_t *) src1->data)[2];
const int n_ctx = ((int32_t *) src1->data)[3];
float freq_base, freq_scale;
memcpy(&freq_base, (int32_t *) src1->data + 4, sizeof(float));
memcpy(&freq_scale, (int32_t *) src1->data + 5, sizeof(float));
@ -12214,11 +12216,13 @@ static void ggml_compute_forward_rope_f16(
return;
}
float freq_base;
float freq_scale;
const int n_past = ((int32_t *) src1->data)[0];
const int n_dims = ((int32_t *) src1->data)[1];
const int mode = ((int32_t *) src1->data)[2];
const int n_ctx = ((int32_t *) src1->data)[3];
float freq_base, freq_scale;
memcpy(&freq_base, (int32_t *) src1->data + 4, sizeof(float));
memcpy(&freq_scale, (int32_t *) src1->data + 5, sizeof(float));

26
llama.cpp
View file

@ -190,8 +190,10 @@ struct llama_hparams {
uint32_t n_head = 32;
uint32_t n_layer = 32;
uint32_t n_rot = 64;
float rope_freq_base = 10000.0f;
float rope_freq_scale = 1.0f;
enum llama_ftype ftype = LLAMA_FTYPE_MOSTLY_F16;
bool operator!=(const llama_hparams & other) const {
@ -843,12 +845,12 @@ struct llama_context_params llama_context_default_params() {
struct llama_context_params result = {
/*.seed =*/ LLAMA_DEFAULT_SEED,
/*.n_ctx =*/ 512,
/*.rope_freq_base =*/ 10000.0f,
/*.rope_freq_scale =*/ 1.0f,
/*.n_batch =*/ 512,
/*.gpu_layers =*/ 0,
/*.main_gpu =*/ 0,
/*.tensor_split =*/ {0},
/*.rope_freq_base =*/ 10000.0f,
/*.rope_freq_scale =*/ 1.0f,
/*.progress_callback =*/ nullptr,
/*.progress_callback_user_data =*/ nullptr,
/*.low_vram =*/ false,
@ -968,12 +970,12 @@ static void llama_model_load_internal(
llama_model & model,
llama_vocab & vocab,
int n_ctx,
float rope_freq_base,
float rope_freq_scale,
int n_batch,
int n_gpu_layers,
int main_gpu,
const float * tensor_split,
float rope_freq_base,
float rope_freq_scale,
bool low_vram,
ggml_type memory_type,
bool use_mmap,
@ -1008,6 +1010,7 @@ static void llama_model_load_internal(
}
hparams.n_ctx = n_ctx;
hparams.rope_freq_base = rope_freq_base;
hparams.rope_freq_scale = rope_freq_scale;
}
@ -1278,12 +1281,12 @@ static bool llama_model_load(
llama_model & model,
llama_vocab & vocab,
int n_ctx,
float rope_freq_base,
float rope_freq_scale,
int n_batch,
int n_gpu_layers,
int main_gpu,
float * tensor_split,
float rope_freq_base,
float rope_freq_scale,
bool low_vram,
ggml_type memory_type,
bool use_mmap,
@ -1292,7 +1295,7 @@ static bool llama_model_load(
llama_progress_callback progress_callback,
void *progress_callback_user_data) {
try {
llama_model_load_internal(fname, model, vocab, n_ctx, rope_freq_base, rope_freq_scale, n_batch, n_gpu_layers, main_gpu, tensor_split, low_vram, memory_type,
llama_model_load_internal(fname, model, vocab, n_ctx, n_batch, n_gpu_layers, main_gpu, tensor_split, rope_freq_base, rope_freq_scale, low_vram, memory_type,
use_mmap, use_mlock, vocab_only, progress_callback, progress_callback_user_data);
return true;
} catch (const std::exception & err) {
@ -1342,9 +1345,10 @@ static bool llama_eval_internal(
const int n_head = hparams.n_head;
const int n_vocab = hparams.n_vocab;
const int n_rot = hparams.n_embd/hparams.n_head;
const int n_gpu_layers = model.n_gpu_layers;
const float freq_base = hparams.rope_freq_base;
const float freq_scale = hparams.rope_freq_scale;
const int n_gpu_layers = model.n_gpu_layers;
auto & mem_per_token = lctx.mem_per_token;
auto & buf_compute = lctx.buf_compute;
@ -2689,9 +2693,9 @@ struct llama_model * llama_load_model_from_file(
ggml_type memory_type = params.f16_kv ? GGML_TYPE_F16 : GGML_TYPE_F32;
if (!llama_model_load(path_model, *model, model->vocab, params.n_ctx, params.rope_freq_base, params.rope_freq_scale,
params.n_batch, params.n_gpu_layers, params.main_gpu, params.tensor_split, params.low_vram, memory_type,
params.use_mmap, params.use_mlock, params.vocab_only, params.progress_callback,
if (!llama_model_load(path_model, *model, model->vocab, params.n_ctx, params.n_batch, params.n_gpu_layers,
params.main_gpu, params.tensor_split, params.rope_freq_base, params.rope_freq_scale,params.low_vram,
memory_type, params.use_mmap, params.use_mlock, params.vocab_only, params.progress_callback,
params.progress_callback_user_data)) {
delete model;
fprintf(stderr, "%s: failed to load model\n", __func__);

7
llama.h
View file

@ -85,12 +85,15 @@ extern "C" {
struct llama_context_params {
uint32_t seed; // RNG seed, -1 for random
int32_t n_ctx; // text context
float rope_freq_base; // RoPE base frequency
float rope_freq_scale; // RoPE frequency scaling factor
int32_t n_batch; // prompt processing batch size
int32_t n_gpu_layers; // number of layers to store in VRAM
int32_t main_gpu; // the GPU that is used for scratch and small tensors
float tensor_split[LLAMA_MAX_DEVICES]; // how to split layers across multiple GPUs
// ref: https://github.com/ggerganov/llama.cpp/pull/2054
float rope_freq_base; // RoPE base frequency
float rope_freq_scale; // RoPE frequency scaling factor
// called with a progress value between 0 and 1, pass NULL to disable
llama_progress_callback progress_callback;
// context pointer passed to the progress callback