@@ -2888,7 +2888,7 @@ struct httpserver {
28882888 if (!support_completion ()) {
28892889 // prompt cache
28902890 cache_prompt = false ;
2891- SRV_INF (" prompt caching %s\n " , cache_prompt ? " unsupported" : " disabled" );
2891+ SRV_INF (" prompt caching %s\n " , params. cache_prompt ? " unsupported" : " disabled" );
28922892
28932893 // context shift
28942894 shift_context = params.llm_params .ctx_shift ;
@@ -2897,8 +2897,7 @@ struct httpserver {
28972897 }
28982898
28992899 // prompt cache
2900- cache_prompt =
2901- llm_model_casual && params.cache_prompt && llm_kv_cache_shift && llama_get_memory (llm_ctx) != nullptr ;
2900+ cache_prompt = params.cache_prompt && llm_kv_cache_shift && llama_get_memory (llm_ctx) != nullptr ;
29022901 if (cache_prompt) {
29032902 cache_prompts.resize (params.llm_params .n_threads_http );
29042903 }
@@ -4049,7 +4048,7 @@ struct httpserver {
40494048 }
40504049
40514050 // prepare cache - clean cache
4052- if (cache_prompt) {
4051+ if (llm_kv_cache_shift && cache_prompt) {
40534052 llama_memory_seq_rm (llama_get_memory (llm_ctx), seq_id, 0 , -1 );
40544053 if (llm_ctx_draft != nullptr ) {
40554054 llama_memory_seq_rm (llama_get_memory (llm_ctx_draft), seq_id, 0 , -1 );
@@ -4721,7 +4720,7 @@ struct httpserver {
47214720 task->embeds [task->embeds .size () - 1 ][0 ] = embed[0 ];
47224721 }
47234722 // clean kv cache
4724- if (cache_prompt ) {
4723+ if (llm_kv_cache_shift ) {
47254724 llama_memory_seq_rm (llama_get_memory (llm_ctx), seq_id, 0 , -1 );
47264725 if (llm_ctx_draft != nullptr ) {
47274726 llama_memory_seq_rm (llama_get_memory (llm_ctx_draft), seq_id, 0 , -1 );
0 commit comments