llama : fix session saving/loading (#3400)

* 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

1 files changed, 3 insertions, 3 deletions

diff --git a/examples/speculative/speculative.cpp b/examples/speculative/speculative.cpp
index c5e5b234..75a2e5e2 100644
--- a/examples/speculative/speculative.cpp
+++ b/examples/speculative/speculative.cpp
@@ -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;