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

llama : remove all_pos_0, all_pos_1, all_seq_id from llama_batch (#9745)

Browse source 
* refactor llama_batch_get_one

* adapt all examples

* fix simple.cpp

* fix llama_bench

* fix

* fix context shifting

* free batch before return

* use common_batch_add, reuse llama_batch in loop

* null terminated seq_id list

* fix save-load-state example

* fix perplexity

* correct token pos in llama_batch_allocr
This commit is contained in:
Xuan Son Nguyen 2024-10-18 23:18:01 +02:00 • committed by GitHub
parent afd9909a64
commit cda0e4b648
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
22 changed files with 205 additions and 118 deletions
Download patch file Download diff file Expand all files Collapse all files

4
common/common.cpp
View file

@ -955,7 +955,7 @@ struct common_init_result common_init_from_params(common_params & params) {
}
if (llama_model_has_encoder(model)) {
llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size(), 0, 0));
llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size()));
llama_token decoder_start_token_id = llama_model_decoder_start_token(model);
if (decoder_start_token_id == -1) {
decoder_start_token_id = bos;
@ -964,7 +964,7 @@ struct common_init_result common_init_from_params(common_params & params) {
tmp.push_back(decoder_start_token_id);
}
if (llama_model_has_decoder(model)) {
llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch), 0, 0));
llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch)));
}
llama_kv_cache_clear(lctx);
llama_synchronize(lctx);