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

llama : fix loading progress bar

Browse source
This commit is contained in:
Georgi Gerganov 2023-08-22 21:53:36 +03:00
parent 2f3c80a845
commit 5c5413dc14
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 23 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

59
llama.cpp
View file

@ -1703,7 +1703,7 @@ static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
if (vocab.linefeed_id != -1) { LLAMA_LOG_INFO( "%s: LF token = %d '%s'\n", __func__, vocab.linefeed_id, vocab.id_to_token[vocab.linefeed_id].text.c_str() ); } if (vocab.linefeed_id != -1) { LLAMA_LOG_INFO( "%s: LF token = %d '%s'\n", __func__, vocab.linefeed_id, vocab.id_to_token[vocab.linefeed_id].text.c_str() ); }
} }
static void llm_load_llama( static void llm_load_tensors(
llm_arch arch, llm_arch arch,
llama_model_loader & ml, llama_model_loader & ml,
llama_model & model, llama_model & model,
@ -2009,24 +2009,10 @@ static bool llama_model_load(
return true; return true;
} }
switch (arch) { llm_load_tensors(
case LLM_ARCH_LLAMA: arch, *ml, model, n_batch, n_gpu_layers,
{ main_gpu, tensor_split, mul_mat_q, low_vram, memory_type,
llm_load_llama( use_mlock, progress_callback, progress_callback_user_data);
arch, *ml, model, n_batch, n_gpu_layers,
main_gpu, tensor_split, mul_mat_q, low_vram, memory_type,
use_mlock, progress_callback, progress_callback_user_data);
} break;
case LLM_ARCH_FALCON:
{
llm_load_llama(
arch, *ml, model, n_batch, n_gpu_layers,
main_gpu, tensor_split, mul_mat_q, low_vram, memory_type,
use_mlock, progress_callback, progress_callback_user_data);
} break;
default:
throw std::runtime_error("unsupported architecture");
};
} catch (const std::exception & err) { } catch (const std::exception & err) {
LLAMA_LOG_ERROR("error loading model: %s\n", err.what()); LLAMA_LOG_ERROR("error loading model: %s\n", err.what());
return false; return false;
@ -4446,6 +4432,22 @@ struct llama_model * llama_load_model_from_file(
ggml_type memory_type = params.f16_kv ? GGML_TYPE_F16 : GGML_TYPE_F32; ggml_type memory_type = params.f16_kv ? GGML_TYPE_F16 : GGML_TYPE_F32;
unsigned cur_percentage = 0;
if (params.progress_callback == NULL) {
params.progress_callback_user_data = &cur_percentage;
params.progress_callback = [](float progress, void * ctx) {
unsigned * cur_percentage_p = (unsigned *) ctx;
unsigned percentage = (unsigned) (100 * progress);
while (percentage > *cur_percentage_p) {
*cur_percentage_p = percentage;
LLAMA_LOG_INFO(".");
if (percentage >= 100) {
LLAMA_LOG_INFO("\n");
}
}
};
}
if (!llama_model_load(path_model, *model, params.n_ctx, params.n_batch, params.n_gpu_layers, if (!llama_model_load(path_model, *model, params.n_ctx, params.n_batch, params.n_gpu_layers,
params.main_gpu, params.tensor_split, params.mul_mat_q, params.rope_freq_base, params.rope_freq_scale, params.main_gpu, params.tensor_split, params.mul_mat_q, params.rope_freq_base, params.rope_freq_scale,
params.low_vram, memory_type, params.use_mmap, params.use_mlock, params.vocab_only, params.low_vram, memory_type, params.use_mmap, params.use_mlock, params.vocab_only,
@ -4476,22 +4478,6 @@ struct llama_context * llama_new_context_with_model(
params.seed = time(NULL); params.seed = time(NULL);
} }
unsigned cur_percentage = 0;
if (params.progress_callback == NULL) {
params.progress_callback_user_data = &cur_percentage;
params.progress_callback = [](float progress, void * ctx) {
unsigned * cur_percentage_p = (unsigned *) ctx;
unsigned percentage = (unsigned) (100 * progress);
while (percentage > *cur_percentage_p) {
*cur_percentage_p = percentage;
LLAMA_LOG_INFO(".");
if (percentage >= 100) {
LLAMA_LOG_INFO("\n");
}
}
};
}
ctx->rng = std::mt19937(params.seed); ctx->rng = std::mt19937(params.seed);
ctx->logits_all = params.logits_all; ctx->logits_all = params.logits_all;
@ -4629,13 +4615,14 @@ struct llama_context * llama_new_context_with_model(
struct llama_context * llama_init_from_file( struct llama_context * llama_init_from_file(
const char * path_model, const char * path_model,
struct llama_context_params params) { struct llama_context_params params) {
struct llama_model * model = llama_load_model_from_file(path_model, params); struct llama_model * model = llama_load_model_from_file(path_model, params);
if (!model) { if (!model) {
return nullptr; return nullptr;
} }
struct llama_context * ctx = llama_new_context_with_model(model, params); struct llama_context * ctx = llama_new_context_with_model(model, params);
ctx->model_owner = true; ctx->model_owner = true;
return ctx; return ctx;
} }