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

speculative : refactor and add a simpler example (#10362)

Browse source 
* speculative : refactor and add a simpler example

ggml-ci

* speculative : clean-up and add comments and TODOs [no ci]

* speculative : manage context in common_speculative

ggml-ci

* speculative : simplify

ggml-ci

* speculative : simplify (cont)

ggml-ci

* speculative : add --draft-min CLI arg

* speculative : minor fixup

* make : build fixes

* speculative : do not redraft previous drafts

ggml-ci

* speculative : fix the draft sampling

ggml-ci

* speculative : fix compile warning

* common : refactor args

ggml-ci

* common : change defaults [no ci]

* common : final touches

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-11-25 09:58:41 +02:00 • committed by GitHub
parent cce5a90075
commit d9d54e498d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
28 changed files with 1028 additions and 326 deletions
Download patch file Download diff file Expand all files Collapse all files

3
examples/lookup/lookup-stats.cpp
View file

@ -21,7 +21,7 @@ int main(int argc, char ** argv){
common_init();
const int n_draft = params.n_draft;
const int n_draft = params.speculative.n_max;
// init llama.cpp
llama_backend_init();
@ -40,6 +40,7 @@ int main(int argc, char ** argv){
common_ngram_cache ngram_cache_context;
common_ngram_cache ngram_cache_dynamic;
common_ngram_cache ngram_cache_static;
int64_t t_draft_flat_us = 0;
int64_t t_draft_us = 0;

4
examples/lookup/lookup.cpp
View file

@ -22,7 +22,7 @@ int main(int argc, char ** argv){
common_init();
// max. number of additional tokens to draft if match is found
const int n_draft = params.n_draft;
const int n_draft = params.speculative.n_max;
const bool dump_kv_cache = params.dump_kv_cache;
@ -102,7 +102,7 @@ int main(int argc, char ** argv){
bool has_eos = false;
struct common_sampler * smpl = common_sampler_init(model, params.sparams);
struct common_sampler * smpl = common_sampler_init(model, params.sampling);
std::vector<llama_token> draft;