root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-24 02:14:35 +00:00
Code Issues Actions 11 Packages Projects Releases Wiki Activity

Revert "Fix memory allocation issues and seg faults"

Browse Source 
This reverts commit 4870e455b3.

Will provide the correct fix later
This commit is contained in:
Georgi Gerganov 2023-03-24 06:22:28 +02:00
parent 4870e455b3
commit 3cd8dde0d1
1 changed files with 18 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
llama.cpp
View File

@ -102,9 +102,6 @@ struct llama_context {
// decode output (2-dimensional array: [n_tokens][n_vocab])
std::vector<float> logits;
bool logits_all = false;
// work buffer for transformer evaluation
std::vector<uint8_t> buf_eval;
};
struct llama_context_params llama_context_default_params() {
@ -630,19 +627,27 @@ static bool llama_eval_internal(
const int n_rot = hparams.n_embd/hparams.n_head;
auto & mem_per_token = lctx.mem_per_token;
auto & buf_eval = lctx.buf_eval;
if (mem_per_token*(n_past + N + 16) > buf_eval.size()) {
const size_t buf_size_new = 1.618*buf_eval.size();
// TODO: fix this hardcoded size
static size_t buf_size = 512u*1024*1024;
static void * buf = malloc(buf_size);
//fprintf(stderr, "\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_eval.size(), buf_size_new);
if (mem_per_token > 0 && mem_per_token*N > buf_size) {
const size_t buf_size_new = 1.3*(mem_per_token*N); // add 30% to account for ggml object overhead
//fprintf(stderr, "\n%s: reallocating buffer from %zu to %zu bytes\n", __func__, buf_size, buf_size_new);
buf_eval.resize(buf_size_new);
// reallocate
buf_size = buf_size_new;
buf = realloc(buf, buf_size);
if (buf == nullptr) {
fprintf(stderr, "%s: failed to allocate %zu bytes\n", __func__, buf_size);
return false;
}
}
struct ggml_init_params params = {
/*.mem_size =*/ buf_eval.size(),
/*.mem_buffer =*/ buf_eval.data(),
/*.mem_size =*/ buf_size,
/*.mem_buffer =*/ buf,
};
struct ggml_context * ctx0 = ggml_init(params);
@ -827,11 +832,10 @@ static bool llama_eval_internal(
memcpy(logits_out.data(), (float *) ggml_get_data(inpL) + (n_vocab*(N-1)), sizeof(float)*n_vocab);
}
if (N == 1) {
mem_per_token = ggml_used_mem(ctx0)/(n_past + N);
if (mem_per_token == 0) {
mem_per_token = ggml_used_mem(ctx0)/N;
}
//fprintf(stderr, "\nused_mem = %zu, %zu MB\n", ggml_used_mem(ctx0), ggml_used_mem(ctx0)/1024/1024);
//fprintf(stderr, "used_mem = %zu\n", ggml_used_mem(ctx0));
ggml_free(ctx0);
@ -1412,8 +1416,6 @@ struct llama_context * llama_init_from_file(
return nullptr;
}
ctx->buf_eval.resize(512u*1024u*1024u);
return ctx;
}