mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-04-22 23:49:23 +00:00
save checkpoint during pp
This commit is contained in:
@@ -104,6 +104,8 @@ struct server_slot {
|
||||
|
||||
void prompt_load(server_prompt_cache& prompt_cache, const server_tokens& tokens);
|
||||
|
||||
size_t checkpoint_pos = 0;
|
||||
|
||||
// sampling
|
||||
llama_token sampled; // in speculative mode, this is the last accepted token
|
||||
llama_tokens drafted;
|
||||
@@ -358,5 +360,7 @@ struct server_context {
|
||||
|
||||
void apply_checkpoint(server_slot & slot);
|
||||
|
||||
void create_checkpoint_at_interval(server_slot & slot, const gpt_params & params_base);
|
||||
|
||||
void release_slot_after_final_response(server_slot & slot);
|
||||
};
|
||||
|
||||
Reference in New Issue
Block a user