Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

lookup: fibonacci hashing, fix crashes (#8548)

Browse Source
This commit is contained in:
Johannes Gäßler 2024-07-17 23:35:44 +02:00 committed by GitHub
parent b3283448ce
commit e02b597be3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
common/ngram-cache.h
View File

@ -37,11 +37,18 @@ struct llama_ngram {
} }
}; };
struct llama_token_hash_function {
size_t operator()(const llama_token token) const {
// see https://probablydance.com/2018/06/16/fibonacci-hashing-the-optimization-that-the-world-forgot-or-a-better-alternative-to-integer-modulo/
return token * 11400714819323198485llu;
}
};
struct llama_ngram_hash_function { struct llama_ngram_hash_function {
size_t operator()(const llama_ngram & ngram) const { size_t operator()(const llama_ngram & ngram) const {
size_t hash = 0; size_t hash = llama_token_hash_function{}(ngram.tokens[0]);
for (int i = 0; i < LLAMA_NGRAM_MAX; ++i) { for (int i = 1; i < LLAMA_NGRAM_MAX; ++i) {
hash ^= std::hash<llama_token>{}(ngram.tokens[i]); hash ^= llama_token_hash_function{}(ngram.tokens[i]);
} }
return hash; return hash;
} }

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

@ -31,7 +31,6 @@ int main(int argc, char ** argv){
// load the model // load the model
std::tie(model, ctx) = llama_init_from_gpt_params(params); std::tie(model, ctx) = llama_init_from_gpt_params(params);
GGML_ASSERT(llama_n_vocab(model) < (1 << 16));
// tokenize the prompt // tokenize the prompt
std::vector<llama_token> inp; std::vector<llama_token> inp;
@ -65,7 +64,7 @@ int main(int argc, char ** argv){
} }
const int n_input = inp.size(); const int n_input = inp.size();
const int n_ctx = params.n_ctx; const int n_ctx = llama_n_ctx(ctx);
int n_drafted = 0; int n_drafted = 0;
int n_accept = 0; int n_accept = 0;

1
examples/lookup/lookup.cpp
View File

@ -39,7 +39,6 @@ int main(int argc, char ** argv){
// load the model // load the model
std::tie(model, ctx) = llama_init_from_gpt_params(params); std::tie(model, ctx) = llama_init_from_gpt_params(params);
GGML_ASSERT(llama_n_vocab(model) < (1 << 16));
// tokenize the prompt // tokenize the prompt
std::vector<llama_token> inp; std::vector<llama_token> inp;