Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-12 21:37:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

llama : use _impl suffix instead of _internal (#11060)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-01-06 10:52:01 +02:00 committed by GitHub
parent 46e3556e01
commit 5047dd3546
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 18 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
src/llama-quant.cpp
View File

@ -22,7 +22,7 @@ static void zeros(std::ofstream & file, size_t n) {
} }
} }
struct quantize_state_internal { struct quantize_state_impl {
const llama_model & model; const llama_model & model;
const llama_model_quantize_params * params; const llama_model_quantize_params * params;
@ -43,13 +43,13 @@ struct quantize_state_internal {
// used to figure out if a model shares tok_embd with the output weight // used to figure out if a model shares tok_embd with the output weight
bool has_output = false; bool has_output = false;
quantize_state_internal(const llama_model & model, const llama_model_quantize_params * params) quantize_state_impl(const llama_model & model, const llama_model_quantize_params * params)
: model(model) : model(model)
, params(params) , params(params)
{} {}
}; };
static void llama_tensor_dequantize_internal( static void llama_tensor_dequantize_impl(
struct ggml_tensor * tensor, std::vector<no_init<float>> & output, std::vector<std::thread> & workers, struct ggml_tensor * tensor, std::vector<no_init<float>> & output, std::vector<std::thread> & workers,
const size_t nelements, const int nthread const size_t nelements, const int nthread
) { ) {
@ -121,7 +121,7 @@ static void llama_tensor_dequantize_internal(
workers.clear(); workers.clear();
} }
static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type new_type, const ggml_tensor * tensor, llama_ftype ftype) { static ggml_type llama_tensor_get_type(quantize_state_impl & qs, ggml_type new_type, const ggml_tensor * tensor, llama_ftype ftype) {
const std::string name = ggml_get_name(tensor); const std::string name = ggml_get_name(tensor);
// TODO: avoid hardcoded tensor names - use the TN_* constants // TODO: avoid hardcoded tensor names - use the TN_* constants
@ -410,7 +410,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n
return new_type; return new_type;
} }
static size_t llama_tensor_quantize_internal(enum ggml_type new_type, const float * f32_data, void * new_data, const int64_t chunk_size, int64_t nrows, int64_t n_per_row, const float * imatrix, std::vector<std::thread> & workers, const int nthread) { static size_t llama_tensor_quantize_impl(enum ggml_type new_type, const float * f32_data, void * new_data, const int64_t chunk_size, int64_t nrows, int64_t n_per_row, const float * imatrix, std::vector<std::thread> & workers, const int nthread) {
if (nthread < 2) { if (nthread < 2) {
// single-thread // single-thread
size_t new_size = ggml_quantize_chunk(new_type, f32_data, new_data, 0, nrows, n_per_row, imatrix); size_t new_size = ggml_quantize_chunk(new_type, f32_data, new_data, 0, nrows, n_per_row, imatrix);
@ -464,7 +464,7 @@ static size_t llama_tensor_quantize_internal(enum ggml_type new_type, const floa
return new_size; return new_size;
} }
static void llama_model_quantize_internal(const std::string & fname_inp, const std::string & fname_out, const llama_model_quantize_params * params) { static void llama_model_quantize_impl(const std::string & fname_inp, const std::string & fname_out, const llama_model_quantize_params * params) {
ggml_type default_type; ggml_type default_type;
llama_ftype ftype = params->ftype; llama_ftype ftype = params->ftype;
@ -534,7 +534,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
llm_load_hparams(ml, model); llm_load_hparams(ml, model);
llm_load_stats (ml, model); llm_load_stats (ml, model);
struct quantize_state_internal qs(model, params); struct quantize_state_impl qs(model, params);
if (params->only_copy) { if (params->only_copy) {
ftype = model.ftype; ftype = model.ftype;
@ -837,7 +837,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
} else if (ggml_is_quantized(tensor->type) && !params->allow_requantize) { } else if (ggml_is_quantized(tensor->type) && !params->allow_requantize) {
throw std::runtime_error(format("requantizing from type %s is disabled", ggml_type_name(tensor->type))); throw std::runtime_error(format("requantizing from type %s is disabled", ggml_type_name(tensor->type)));
} else { } else {
llama_tensor_dequantize_internal(tensor, f32_conv_buf, workers, nelements, nthread); llama_tensor_dequantize_impl(tensor, f32_conv_buf, workers, nelements, nthread);
f32_data = (float *) f32_conv_buf.data(); f32_data = (float *) f32_conv_buf.data();
} }
@ -866,7 +866,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
void * new_data_03 = (char *)new_data + ggml_row_size(new_type, n_per_row) * i03 * nrows; void * new_data_03 = (char *)new_data + ggml_row_size(new_type, n_per_row) * i03 * nrows;
const float * imatrix_03 = imatrix ? imatrix + i03 * n_per_row : nullptr; const float * imatrix_03 = imatrix ? imatrix + i03 * n_per_row : nullptr;
new_size += llama_tensor_quantize_internal(new_type, f32_data_03, new_data_03, chunk_size, nrows, n_per_row, imatrix_03, workers, nthread_use); new_size += llama_tensor_quantize_impl(new_type, f32_data_03, new_data_03, chunk_size, nrows, n_per_row, imatrix_03, workers, nthread_use);
} }
LLAMA_LOG_INFO("size = %8.2f MiB -> %8.2f MiB\n", ggml_nbytes(tensor)/1024.0/1024.0, new_size/1024.0/1024.0); LLAMA_LOG_INFO("size = %8.2f MiB -> %8.2f MiB\n", ggml_nbytes(tensor)/1024.0/1024.0, new_size/1024.0/1024.0);
} }
@ -919,7 +919,7 @@ uint32_t llama_model_quantize(
const char * fname_out, const char * fname_out,
const llama_model_quantize_params * params) { const llama_model_quantize_params * params) {
try { try {
llama_model_quantize_internal(fname_inp, fname_out, params); llama_model_quantize_impl(fname_inp, fname_out, params);
} catch (const std::exception & err) { } catch (const std::exception & err) {
LLAMA_LOG_ERROR("%s: failed to quantize: %s\n", __func__, err.what()); LLAMA_LOG_ERROR("%s: failed to quantize: %s\n", __func__, err.what());
return 1; return 1;

16
src/llama.cpp
View File

@ -10717,7 +10717,7 @@ static enum ggml_status llama_graph_compute(
// return positive int on warning // return positive int on warning
// return negative int on error // return negative int on error
// //
static int llama_decode_internal( static int llama_decode_impl(
llama_context & lctx, llama_context & lctx,
llama_batch inp_batch) { llama_batch inp_batch) {
@ -11052,7 +11052,7 @@ static int llama_decode_internal(
// return positive int on warning // return positive int on warning
// return negative int on error // return negative int on error
// //
static int llama_encode_internal( static int llama_encode_impl(
llama_context & lctx, llama_context & lctx,
llama_batch inp_batch) { llama_batch inp_batch) {
@ -11234,7 +11234,7 @@ static int llama_encode_internal(
} }
// find holes from the beginning of the KV cache and fill them by moving data from the end of the cache // find holes from the beginning of the KV cache and fill them by moving data from the end of the cache
static void llama_kv_cache_defrag_internal(struct llama_context & lctx) { static void llama_kv_cache_defrag_impl(struct llama_context & lctx) {
auto & kv_self = lctx.kv_self; auto & kv_self = lctx.kv_self;
const auto & hparams = lctx.model.hparams; const auto & hparams = lctx.model.hparams;
@ -11454,7 +11454,7 @@ static void llama_kv_cache_defrag_internal(struct llama_context & lctx) {
//LLAMA_LOG_INFO("(tmp log) KV defrag time: %.3f ms\n", (t_end - t_start)/1000.0); //LLAMA_LOG_INFO("(tmp log) KV defrag time: %.3f ms\n", (t_end - t_start)/1000.0);
} }
static void llama_kv_cache_update_internal(struct llama_context & lctx) { static void llama_kv_cache_update_impl(struct llama_context & lctx) {
bool need_reserve = false; bool need_reserve = false;
if (lctx.kv_self.has_shift) { if (lctx.kv_self.has_shift) {
@ -11490,7 +11490,7 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
// defragment the KV cache if needed // defragment the KV cache if needed
if (lctx.kv_self.do_defrag) { if (lctx.kv_self.do_defrag) {
llama_kv_cache_defrag_internal(lctx); llama_kv_cache_defrag_impl(lctx);
need_reserve = true; need_reserve = true;
@ -12191,7 +12191,7 @@ void llama_kv_cache_defrag(struct llama_context * ctx) {
} }
void llama_kv_cache_update(struct llama_context * ctx) { void llama_kv_cache_update(struct llama_context * ctx) {
llama_kv_cache_update_internal(*ctx); llama_kv_cache_update_impl(*ctx);
} }
bool llama_kv_cache_can_shift(struct llama_context * ctx) { bool llama_kv_cache_can_shift(struct llama_context * ctx) {
@ -12203,7 +12203,7 @@ bool llama_kv_cache_can_shift(struct llama_context * ctx) {
int32_t llama_encode( int32_t llama_encode(
struct llama_context * ctx, struct llama_context * ctx,
struct llama_batch batch) { struct llama_batch batch) {
const int ret = llama_encode_internal(*ctx, batch); const int ret = llama_encode_impl(*ctx, batch);
if (ret != 0) { if (ret != 0) {
LLAMA_LOG_ERROR("%s: failed to encode, ret = %d\n", __func__, ret); LLAMA_LOG_ERROR("%s: failed to encode, ret = %d\n", __func__, ret);
} }
@ -12214,7 +12214,7 @@ int32_t llama_encode(
int32_t llama_decode( int32_t llama_decode(
struct llama_context * ctx, struct llama_context * ctx,
struct llama_batch batch) { struct llama_batch batch) {
const int ret = llama_decode_internal(*ctx, batch); const int ret = llama_decode_impl(*ctx, batch);
if (ret != 0) { if (ret != 0) {
LLAMA_LOG_ERROR("%s: failed to decode, ret = %d\n", __func__, ret); LLAMA_LOG_ERROR("%s: failed to decode, ret = %d\n", __func__, ret);
} }