Optimize KV cache and compute buffers, bench results from Pi5.

1 files changed, 4 insertions, 6 deletions

diff --git a/llm.h b/llm.h
index 5bec282..dfe5c39 100644
--- a/llm.h
+++ b/llm.h
@@ -3,11 +3,9 @@
 
 #include "llama.h"
 
-struct llama_model *llm_init(const char *model_path);
-
-void llm_process_request(struct llama_model *model, 
-	const char *user_prompt, FILE *out);
-
-void llm_free(struct llama_model *model);
+struct llm_ctx;
+struct llm_ctx *llm_init(const char *model_path);
+void llm_run(struct llm_ctx *ctx, const char *prompt, FILE *out);
+void llm_free(struct llm_ctx *ctx);
 
 #endif /* LLM_H */