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

Apply suggestions from code review

Browse source 
Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
Michael Klimenko 2024-01-29 08:54:01 +01:00 • committed by GitHub
parent e75d00797e
commit 2dcf37c10b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
13 changed files with 48 additions and 48 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/sampling.cpp
View file

@ -1,7 +1,7 @@
#include "sampling.h"
struct llama_sampling_context * llama_sampling_init(const struct llama_sampling_params & params) {
auto result = new llama_sampling_context();
auto * result = new llama_sampling_context();
result->params = params;
result->grammar = nullptr;
@ -197,8 +197,8 @@ static llama_token llama_sampling_sample_impl(
}
// apply params.logit_bias map
for (const auto & logit_bia : params.logit_bias) {
logits[logit_bia.first] += logit_bia.second;
for (const auto & logit_bias : params.logit_bias) {
logits[logit_bias.first] += logit_bias.second;
}
if (ctx_cfg) {

8
common/train.cpp
View file

@ -18,7 +18,7 @@ struct random_uniform_distribution {
};
struct train_state * init_train_state() {
auto state = new struct train_state;
auto * state = new struct train_state;
state->train_its = 0;
state->train_samples = 0;
state->train_tokens = 0;
@ -46,12 +46,12 @@ void free_train_state(struct train_state * state) {
struct random_normal_distribution * init_random_normal_distribution(
int seed, float mean, float std, float min, float max
) {
auto rnd = new random_normal_distribution{std::mt19937(seed), std::normal_distribution<float>{mean, std}, min, max};
auto * rnd = new random_normal_distribution{std::mt19937(seed), std::normal_distribution<float>{mean, std}, min, max};
return rnd;
}
struct random_uniform_distribution * init_random_uniform_distribution(int seed, float min, float max) {
auto rnd = new random_uniform_distribution{std::mt19937(seed), std::uniform_real_distribution<float>{min, max}};
auto * rnd = new random_uniform_distribution{std::mt19937(seed), std::uniform_real_distribution<float>{min, max}};
return rnd;
}
@ -1379,7 +1379,7 @@ void finish_processing_train_args(struct train_params_common * params) {
}
void train_opt_callback(void * vdata, int accum_step, float * sched, bool * cancel) {
auto data = (struct train_opt_callback_data *) vdata;
auto * data = (struct train_opt_callback_data *) vdata;
struct train_params_common * params = data->params;
struct train_state * train = data->train;
struct ggml_opt_context * opt = train->opt;

10
examples/export-lora/export-lora.cpp
View file

@ -225,7 +225,7 @@ static void free_lora(struct lora_data * lora) {
}
static struct lora_data * load_lora(struct lora_info * info) {
auto result = new struct lora_data;
auto * result = new struct lora_data;
result->info = *info;
result->ctx = NULL;
result->lora_r = 1;
@ -370,8 +370,8 @@ static bool apply_lora(struct ggml_tensor * tensor, struct lora_data * lora, int
static void export_lora(struct export_lora_params * params) {
// load all loras
std::vector<struct lora_data *> loras;
for (auto& i : params->lora) {
auto lora = load_lora(&i);
for (auto & i : params->lora) {
auto * lora = load_lora(&i);
if (lora) {
loras.push_back(lora);
}
@ -431,7 +431,7 @@ static void export_lora(struct export_lora_params * params) {
fin.read_raw(data.data(), data.size());
// apply all loras
for (auto& lora : loras) {
for (auto & lora : loras) {
apply_lora(tensor, lora, params->n_threads);
}
@ -455,7 +455,7 @@ static void export_lora(struct export_lora_params * params) {
gguf_free(gguf_in);
// free loras
for (auto& lora : loras) {
for (auto * lora : loras) {
free_lora(lora);
}
}

16
examples/finetune/finetune.cpp
View file

@ -379,7 +379,7 @@ static void alloc_lora(struct ggml_allocr * alloc, struct my_llama_lora * lora)
ggml_allocr_alloc(alloc, lora->norm_b);
ggml_allocr_alloc(alloc, lora->output_a);
ggml_allocr_alloc(alloc, lora->output_b);
for (auto& layer : lora->layers) {
for (auto & layer : lora->layers) {
ggml_allocr_alloc(alloc, layer.attention_norm_a);
ggml_allocr_alloc(alloc, layer.attention_norm_b);
ggml_allocr_alloc(alloc, layer.wq_a);
@ -405,7 +405,7 @@ static void alloc_lora(struct ggml_allocr * alloc, struct my_llama_lora * lora)
ggml_allocr_alloc(alloc, lora->norm_b->grad);
ggml_allocr_alloc(alloc, lora->output_a->grad);
ggml_allocr_alloc(alloc, lora->output_b->grad);
for (auto& layer : lora->layers) {
for (auto & layer : lora->layers) {
ggml_allocr_alloc(alloc, layer.attention_norm_a->grad);
ggml_allocr_alloc(alloc, layer.attention_norm_b->grad);
ggml_allocr_alloc(alloc, layer.wq_a->grad);
@ -801,7 +801,7 @@ static struct ggml_tensor * llama_build_lora_finetune_graphs(
// allocating checkpoints in one block to reduce memory fragmentation
// note: they will be freed in reverse order
for (auto& checkpoint : checkpoints) {
for (auto * checkpoint : checkpoints) {
if (checkpoint->data == NULL && checkpoint->view_src == NULL) {
ggml_allocr_alloc(alloc, checkpoint);
}
@ -870,7 +870,7 @@ static void load_llama_lora_gguf(struct gguf_context * fctx, struct ggml_context
copy_tensor_by_name(lora->output_a, f_ggml_ctx, ggml_get_name(lora->output_a));
copy_tensor_by_name(lora->output_b, f_ggml_ctx, ggml_get_name(lora->output_b));
for (auto& layer : lora->layers) {
for (auto & layer : lora->layers) {
copy_tensor_by_name(layer.attention_norm_a, f_ggml_ctx, ggml_get_name(layer.attention_norm_a));
copy_tensor_by_name(layer.attention_norm_b, f_ggml_ctx, ggml_get_name(layer.attention_norm_b));
copy_tensor_by_name(layer.wq_a, f_ggml_ctx, ggml_get_name(layer.wq_a));
@ -937,7 +937,7 @@ static void save_llama_lora_gguf(struct gguf_context * fctx, struct my_llama_mod
gguf_add_tensor(fctx, lora->output_a);
gguf_add_tensor(fctx, lora->output_b);
for (auto& layer : lora->layers) {
for (auto & layer : lora->layers) {
gguf_add_tensor(fctx, layer.attention_norm_a);
gguf_add_tensor(fctx, layer.attention_norm_b);
gguf_add_tensor(fctx, layer.wq_a);
@ -1471,7 +1471,7 @@ struct save_train_files_data {
};
static void save_train_files(void * vdata, struct train_state * train) {
auto data = (struct save_train_files_data *) vdata;
auto * data = (struct save_train_files_data *) vdata;
int64_t iter = train->opt->iter;
@ -1494,7 +1494,7 @@ static int64_t get_parameter_count(struct my_llama_lora* lora) {
nx += ggml_nelements(lora->output_a);
nx += ggml_nelements(lora->output_b);
for (auto& layer : lora->layers) {
for (auto & layer : lora->layers) {
nx += ggml_nelements(layer.attention_norm_a);
nx += ggml_nelements(layer.attention_norm_b);
nx += ggml_nelements(layer.wq_a);
@ -1815,7 +1815,7 @@ int main(int argc, char ** argv) {
++token_noccurs[train_token];
}
int n_unique_tokens = 0;
for (unsigned long long token_noccur : token_noccurs) {
for (size_t token_noccur : token_noccurs) {
if (token_noccur == 0) continue;
++n_unique_tokens;
}

2
examples/imatrix/imatrix.cpp
View file

@ -216,7 +216,7 @@ static std::vector<float> softmax(const std::vector<float>& logits) {
sum_exp += exp_logit;
probs[i] = exp_logit;
}
for (float& prob : probs) {
for (float & prob : probs) {
prob /= float(sum_exp);
}
return probs;

6
examples/llava/clip.cpp
View file

@ -1011,21 +1011,21 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
vision_model.mm_2_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "weight"));
vision_model.mm_2_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "bias"));
} catch (std::runtime_error & e) {
static_cast<void>(e);
GGML_UNUSED(e);
}
try {
// Yi-type llava
vision_model.mm_3_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 3, "weight"));
vision_model.mm_3_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 3, "bias"));
} catch (std::runtime_error & e) {
static_cast<void>(e);
GGML_UNUSED(e);
}
try {
// Yi-type llava
vision_model.mm_4_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 4, "weight"));
vision_model.mm_4_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 4, "bias"));
} catch (std::runtime_error & e) {
static_cast<void>(e);
GGML_UNUSED(e);
}
}
else if (new_clip->proj_type == PROJECTOR_TYPE_LDP) {

4
examples/lookahead/lookahead.cpp
View file

@ -268,7 +268,7 @@ int main(int argc, char ** argv) {
// if no active ngrams are left, it means the sampled token does not pass the verification
if (v > 0) {
for (auto& g : ngrams_cur) {
for (auto & g : ngrams_cur) {
if (g.active) {
i_batch = g.i_batch[v];
seq_id_best = g.seq_id;
@ -316,7 +316,7 @@ int main(int argc, char ** argv) {
}
// verify across active n-grams
for (auto& g : ngrams_cur) {
for (auto & g : ngrams_cur) {
if (g.active) {
if (v == N - 1) {
g.active = false;

2
examples/perplexity/perplexity.cpp
View file

@ -94,7 +94,7 @@ static std::vector<float> softmax(const std::vector<float>& logits) {
sum_exp += exp_logit;
probs[i] = exp_logit;
}
for (float& prob : probs) {
for (float & prob : probs) {
prob /= float(sum_exp);
}
return probs;

10
examples/train-text-from-scratch/train-text-from-scratch.cpp
View file

@ -151,7 +151,7 @@ static void alloc_model(struct ggml_allocr * alloc, struct my_llama_model * mode
ggml_allocr_alloc(alloc, model->tok_embeddings);
ggml_allocr_alloc(alloc, model->norm);
ggml_allocr_alloc(alloc, model->output);
for (auto& layer : model->layers) {
for (auto & layer : model->layers) {
ggml_allocr_alloc(alloc, layer.attention_norm);
ggml_allocr_alloc(alloc, layer.wq);
ggml_allocr_alloc(alloc, layer.wk);
@ -165,7 +165,7 @@ static void alloc_model(struct ggml_allocr * alloc, struct my_llama_model * mode
ggml_allocr_alloc(alloc, model->tok_embeddings->grad);
ggml_allocr_alloc(alloc, model->norm->grad);
ggml_allocr_alloc(alloc, model->output->grad);
for (auto& layer : model->layers) {
for (auto & layer : model->layers) {
ggml_allocr_alloc(alloc, layer.attention_norm->grad);
ggml_allocr_alloc(alloc, layer.wq->grad);
ggml_allocr_alloc(alloc, layer.wk->grad);
@ -451,7 +451,7 @@ static struct ggml_tensor * llama_build_train_graphs(
// allocating checkpoints in one block to reduce memory fragmentation
// note: they will be freed in reverse order
for (auto& checkpoint : checkpoints) {
for (auto * checkpoint : checkpoints) {
if (checkpoint->data == NULL && checkpoint->view_src == NULL) {
ggml_allocr_alloc(alloc, checkpoint);
}
@ -923,7 +923,7 @@ struct save_train_files_data {
};
static void save_train_files(void * vdata, struct train_state * train) {
auto data = (struct save_train_files_data *) vdata;
auto * data = (struct save_train_files_data *) vdata;
int64_t iter = train->opt->iter;
if (strlen(data->fn_checkpoint_out) > 0) {
@ -943,7 +943,7 @@ static int64_t get_parameter_count(struct my_llama_model* model) {
nx += ggml_nelements(model->norm);
nx += ggml_nelements(model->output);
for (auto& layer : model->layers) {
for (auto & layer : model->layers) {
nx += ggml_nelements(layer.attention_norm);
nx += ggml_nelements(layer.wq);
nx += ggml_nelements(layer.wk);

16
llama.cpp
View file

@ -3004,7 +3004,7 @@ static void llm_load_hparams(
}
// TODO: This should probably be in llama.h
static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab & vocab, const std::string& raw_text, bool bos, bool special = false);
static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab & vocab, const std::string & raw_text, bool bos, bool special = false);
static llama_token llama_byte_to_token(const llama_vocab & vocab, uint8_t ch);
static void llm_load_vocab(
@ -7374,7 +7374,7 @@ private:
bpe_encoded_words.reserve(text.size());
auto cps = codepoints_from_utf8(text);
for (unsigned int cp : cps)
for (uint32_t cp : cps)
text_utf.emplace_back(codepoint_to_utf8(cp));
for (int i = 0; i < (int)text_utf.size(); i++) {
@ -7633,7 +7633,7 @@ static void tokenizer_st_partition(const llama_vocab & vocab, std::forward_list<
}
}
static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab & vocab, const std::string& raw_text, bool bos, bool special) {
static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab & vocab, const std::string & raw_text, bool bos, bool special) {
std::vector<llama_vocab::id> output;
// OG tokenizer behavior:
@ -8089,7 +8089,7 @@ void llama_grammar_free(struct llama_grammar * grammar) {
}
struct llama_grammar * llama_grammar_copy(const struct llama_grammar * grammar) {
auto result = new llama_grammar{ grammar->rules, grammar->stacks, grammar->partial_utf8 };
auto * result = new llama_grammar{ grammar->rules, grammar->stacks, grammar->partial_utf8 };
// redirect elements in stacks to point to new rules
for (size_t is = 0; is < result->stacks.size(); is++) {
@ -8337,7 +8337,7 @@ void llama_sample_tail_free(struct llama_context * ctx, llama_token_data_array *
}
// Calculate absolute value of second derivatives
for (float& second_derivative : second_derivatives) {
for (float & second_derivative : second_derivatives) {
second_derivative = std::abs(second_derivative);
}
@ -9654,7 +9654,7 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
if (tot_count > 0) {
LLAMA_LOG_INFO(" | hist: ");
for (long long i : hist_cur) {
for (int64_t i : hist_cur) {
LLAMA_LOG_INFO("%5.3f ", i / float(nelements));
}
}
@ -10101,7 +10101,7 @@ struct llama_model * llama_load_model_from_file(
struct llama_model_params params) {
ggml_time_init();
auto model = new llama_model;
auto * model = new llama_model;
unsigned cur_percentage = 0;
if (params.progress_callback == NULL) {
@ -10147,7 +10147,7 @@ struct llama_context * llama_new_context_with_model(
return nullptr;
}
auto ctx = new llama_context(*model);
auto * ctx = new llama_context(*model);
const auto & hparams = model->hparams;
auto & cparams = ctx->cparams;

10
tests/test-backend-ops.cpp
View file

@ -64,7 +64,7 @@ static void init_tensor_uniform(ggml_tensor * tensor, float min = -1.0f, float m
}
}
ggml_quantize_chunk(tensor->type, data.data(), dataq.data(), 0, int(size/tensor->ne[0]),
static_cast<int>(tensor->ne[0]), hist, im);
int(tensor->ne[0]), hist, im);
ggml_backend_tensor_set(tensor, dataq.data(), 0, dataq.size());
} else if (tensor->type == GGML_TYPE_I8 || tensor->type == GGML_TYPE_I16 || tensor->type == GGML_TYPE_I32) {
// This is going to create some weird integers though.
@ -288,9 +288,9 @@ struct test_case {
virtual size_t op_size(ggml_tensor * t) {
size_t size = ggml_nbytes(t);
// add source tensors
for (auto& el : t->src) {
if (el) {
size += ggml_nbytes(el);
for (auto * src : t->src) {
if (src) {
size += ggml_nbytes(src);
}
}
return size;
@ -423,7 +423,7 @@ struct test_case {
};
auto callback = [](int index, ggml_tensor * t1, ggml_tensor * t2, void * user_data) -> bool {
auto ud = (callback_userdata *) user_data;
auto * ud = (callback_userdata *) user_data;
const char * bn1 = ggml_backend_name(ud->backend1);
const char * bn2 = ggml_backend_name(ud->backend2);

4
tests/test-grammar-parser.cpp
View file

@ -29,7 +29,7 @@ term ::= [0-9]+)""";
};
uint32_t index = 0;
for (auto& symbol_id : parsed_grammar.symbol_ids) {
for (auto & symbol_id : parsed_grammar.symbol_ids) {
std::string key = symbol_id.first;
uint32_t value = symbol_id.second;
std::pair<std::string, uint32_t> expected_pair = expected[index];
@ -132,7 +132,7 @@ term ::= [0-9]+)""";
};
index = 0;
for (auto& symbol_id : parsed_grammar.symbol_ids) {
for (auto & symbol_id : parsed_grammar.symbol_ids) {
std::string key = symbol_id.first;
uint32_t value = symbol_id.second;
std::pair<std::string, uint32_t> expected_pair = expected[index];

2
tests/test-llama-grammar.cpp
View file

@ -103,7 +103,7 @@ int main()
parsed_grammar.symbol_ids[pair.first] = pair.second;
}
for (const auto& rule : expected_rules)
for (const auto & rule : expected_rules)
{
parsed_grammar.rules.emplace_back();
for (auto element : rule)