tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-29 04:35:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : fix session saving/loading (#3400)

Browse Source 
* llama : fix session saving/loading

* llama : temp fix for clearing "future" tokens from the KV cache

* llama : fix handling of "future" tokens when loading sessions

* llama : fix comments for llama_kv_cache API
This commit is contained in:
Georgi Gerganov
2023-10-03 21:04:01 +03:00
committed by GitHub
parent 48be797ffb
commit ac2219fef3
7 changed files with 106 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/speculative/speculative.cpp
View File

@ -172,7 +172,7 @@ int main(int argc, char ** argv) {
LOG("out of drafted tokens\n");
}
llama_kv_cache_seq_rm(ctx_dft, 0, n_past_dft, n_ctx);
llama_kv_cache_seq_rm(ctx_dft, 0, n_past_dft, -1);
llama_decode(ctx_dft, llama_batch_get_one(&id, 1, n_past_dft, 0));
++n_past_dft;
@ -257,7 +257,7 @@ int main(int argc, char ** argv) {
}
// evaluate the drafted token on the draft model
llama_kv_cache_seq_rm(ctx_dft, 0, n_past_cur, n_ctx);
llama_kv_cache_seq_rm(ctx_dft, 0, n_past_cur, -1);
llama_decode(ctx_dft, llama_batch_get_one(&drafted.back(), 1, n_past_cur, 0));
++n_past_cur;
@ -267,7 +267,7 @@ int main(int argc, char ** argv) {
}
// evaluate the target model on the drafted tokens
llama_kv_cache_seq_rm(ctx_tgt, 0, n_past_tgt, n_ctx);
llama_kv_cache_seq_rm(ctx_tgt, 0, n_past_tgt, -1);
llama_decode(ctx_tgt, llama_batch_get_one(drafted.data(), drafted.size(), n_past_tgt, 0));
++n_past_tgt;