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

llama : various integer type cast and format string fixes

Browse source 
Some platforms use "%lu" and others "%llu" for uint64_t.
Not sure how to handle that, so casting to size_t when displaying errors.
This commit is contained in:
Francis Couture-Harpin 2024-07-25 22:58:20 -04:00
parent 9e22064a0d
commit cddc899b85
1 changed files with 7 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/llama.cpp
View file

@ -17357,7 +17357,7 @@ struct llama_data_context {
} }
void write_logits(const struct llama_context * ctx) { void write_logits(const struct llama_context * ctx) {
const uint64_t logits_size = std::min(ctx->logits_size, (uint64_t) ctx->n_outputs * ctx->model.hparams.n_vocab); const uint64_t logits_size = std::min((uint64_t) ctx->logits_size, (uint64_t) ctx->n_outputs * ctx->model.hparams.n_vocab);
write(&logits_size, sizeof(logits_size)); write(&logits_size, sizeof(logits_size));
@ -17367,7 +17367,7 @@ struct llama_data_context {
} }
void write_embeddings(const struct llama_context * ctx) { void write_embeddings(const struct llama_context * ctx) {
const uint64_t embeddings_size = std::min(ctx->embd_size, (uint64_t) ctx->n_outputs * ctx->model.hparams.n_embd); const uint64_t embeddings_size = std::min((uint64_t) ctx->embd_size, (uint64_t) ctx->n_outputs * ctx->model.hparams.n_embd);
write(&embeddings_size, sizeof(embeddings_size)); write(&embeddings_size, sizeof(embeddings_size));
@ -17461,7 +17461,7 @@ struct llama_data_context {
write(&v_type_i, sizeof(v_type_i)); write(&v_type_i, sizeof(v_type_i));
// Write element size // Write element size
const uint64_t v_size_el = ggml_type_size(kv_self.v_l[il]->type); const uint32_t v_size_el = ggml_type_size(kv_self.v_l[il]->type);
write(&v_size_el, sizeof(v_size_el)); write(&v_size_el, sizeof(v_size_el));
// Write GQA embedding size // Write GQA embedding size
@ -17710,7 +17710,7 @@ struct llama_data_read_context {
const size_t k_size_row = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa); const size_t k_size_row = ggml_row_size(kv_self.k_l[il]->type, n_embd_k_gqa);
if (k_size_row != k_size_row_ref) { if (k_size_row != k_size_row_ref) {
// llama_kv_cache_seq_rm(kv_self, dest_seq_id, -1, -1); // llama_kv_cache_seq_rm(kv_self, dest_seq_id, -1, -1);
LLAMA_LOG_ERROR("%s: mismatched key row size (%zu != %zu, layer %d)\n", __func__, k_size_row, k_size_row_ref, il); LLAMA_LOG_ERROR("%s: mismatched key row size (%zu != %zu, layer %d)\n", __func__, k_size_row, (size_t) k_size_row_ref, il);
return false; return false;
} }
@ -17739,7 +17739,7 @@ struct llama_data_read_context {
read_to(&v_size_row_ref, sizeof(v_size_row_ref)); read_to(&v_size_row_ref, sizeof(v_size_row_ref));
const size_t v_size_row = ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa); const size_t v_size_row = ggml_row_size(kv_self.v_l[il]->type, n_embd_v_gqa);
if (v_size_row != v_size_row_ref) { if (v_size_row != v_size_row_ref) {
LLAMA_LOG_ERROR("%s: mismatched value row size (%zu != %zu, layer %d)\n", __func__, v_size_row, v_size_row_ref, il); LLAMA_LOG_ERROR("%s: mismatched value row size (%zu != %zu, layer %d)\n", __func__, v_size_row, (size_t) v_size_row_ref, il);
return false; return false;
} }
@ -17763,11 +17763,11 @@ struct llama_data_read_context {
} }
// Read element size of value // Read element size of value
uint64_t v_size_el_ref; uint32_t v_size_el_ref;
read_to(&v_size_el_ref, sizeof(v_size_el_ref)); read_to(&v_size_el_ref, sizeof(v_size_el_ref));
const size_t v_size_el = ggml_type_size(kv_self.v_l[il]->type); const size_t v_size_el = ggml_type_size(kv_self.v_l[il]->type);
if (v_size_el != v_size_el_ref) { if (v_size_el != v_size_el_ref) {
LLAMA_LOG_ERROR("%s: mismatched value element size (%zu != %llu, layer %d)\n", __func__, v_size_el, v_size_el_ref, il); LLAMA_LOG_ERROR("%s: mismatched value element size (%zu != %zu, layer %d)\n", __func__, v_size_el, (size_t) v_size_el_ref, il);
return false; return false;
} }