Added api for getting/setting the kv_cache (#685)

The api provides access methods for retrieving the current memory buffer for the kv_cache and its token number.
It also contains a method for setting the kv_cache from a memory buffer.

This makes it possible to load/save history - maybe support --cache-prompt paramater as well?

Co-authored-by: Pavol Rusnak <pavol@rusnak.io>

1 files changed, 17 insertions, 0 deletions

diff --git a/llama.h b/llama.h
index 258de5a9..04e2bf71 100644
--- a/llama.h
+++ b/llama.h
@@ -83,6 +83,23 @@ extern "C" {
             const char * fname_out,
                    int   itype);
 
+    // Returns the KV cache that will contain the context for the
+    // ongoing prediction with the model.
+    LLAMA_API const uint8_t * llama_get_kv_cache(struct llama_context * ctx);
+
+    // Returns the size of the KV cache
+    LLAMA_API size_t llama_get_kv_cache_size(struct llama_context * ctx);
+
+    // Returns the number of tokens in the KV cache
+    LLAMA_API int llama_get_kv_cache_token_count(struct llama_context * ctx);
+
+    // Sets the KV cache containing the current context for the model
+    LLAMA_API void llama_set_kv_cache(
+            struct llama_context * ctx,
+                   const uint8_t * kv_cache,
+                          size_t   n_size,
+                             int   n_token_count);
+
     // Run the llama inference to obtain the logits and probabilities for the next token.
     // tokens + n_tokens is the provided batch of new tokens to process
     // n_past is the number of tokens to use from previous eval calls