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

Command line switch to use F16 for memory_k and memory_v (refactor of #154) (#294)

Browse source 
* Use F16 for memory_k and memory_v

* add command line switch to use f16 instead of f32 for memory k+v

---------

Co-authored-by: Ty Everett <ty@tyweb.us>
This commit is contained in:
Erik Scholz 2023-03-19 18:57:00 +01:00 • committed by GitHub
parent 160bfb217d
commit 0b366e7357
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 11 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

13
main.cpp
View file

@ -86,7 +86,7 @@ struct llama_model {
};
// load the model's weights from a file
bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab & vocab, int n_ctx) {
bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab & vocab, int n_ctx, ggml_type memory_type = GGML_TYPE_F32) {
fprintf(stderr, "%s: loading model from '%s' - please wait ...\n", __func__, fname.c_str());
std::vector<char> f_buf(1024*1024);
@ -207,8 +207,8 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab
ctx_size += n_layer*(n_ff*n_embd*ggml_type_sizef(wtype)); // w2
ctx_size += n_layer*(n_ff*n_embd*ggml_type_sizef(wtype)); // w3
ctx_size += n_ctx*n_layer*n_embd*ggml_type_sizef(GGML_TYPE_F32); // memory_k
ctx_size += n_ctx*n_layer*n_embd*ggml_type_sizef(GGML_TYPE_F32); // memory_v
ctx_size += n_ctx*n_layer*n_embd*ggml_type_sizef(memory_type); // memory_k
ctx_size += n_ctx*n_layer*n_embd*ggml_type_sizef(memory_type); // memory_v
ctx_size += (5 + 10*n_layer)*256; // object overhead
@ -293,8 +293,8 @@ bool llama_model_load(const std::string & fname, llama_model & model, gpt_vocab
const int n_mem = n_layer*n_ctx;
const int n_elements = n_embd*n_mem;
model.memory_k = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_elements);
model.memory_v = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, n_elements);
model.memory_k = ggml_new_tensor_1d(ctx, memory_type, n_elements);
model.memory_v = ggml_new_tensor_1d(ctx, memory_type, n_elements);
const size_t memory_size = ggml_nbytes(model.memory_k) + ggml_nbytes(model.memory_v);
@ -814,8 +814,9 @@ int main(int argc, char ** argv) {
// load the model
{
const ggml_type memory_type = params.memory_f16 ? GGML_TYPE_F16 : GGML_TYPE_F32;
const int64_t t_start_us = ggml_time_us();
if (!llama_model_load(params.model, model, vocab, params.n_ctx)) {
if (!llama_model_load(params.model, model, vocab, params.n_ctx, memory_type)) {
fprintf(stderr, "%s: failed to load model from '%s'\n", __func__, params.model.c_str());
return 1;
}