server : do not create checkpoints right after mtmd chunks (ggml-org#20232)

Author: ggerganov

tools/server/server-context.cpp

@@ -2438,6 +2438,8 @@ struct server_context_impl {
                         slot.n_prompt_tokens_cache = 0;
                     }
 
+                    bool do_checkpoint = params_base.n_ctx_checkpoints > 0;
+
                     // check if we should process the image
                     if (slot.prompt.n_tokens() < slot.task->n_tokens() && input_tokens[slot.prompt.n_tokens()] == LLAMA_TOKEN_NULL) {
                         // process the image
@@ -2457,6 +2459,8 @@ struct server_context_impl {
                             const auto & chunk = input_tokens.find_chunk(slot.prompt.n_tokens());
                             slot.prompt.tokens.push_back(chunk.get()); // copy
                         }
+
+                        do_checkpoint = false; // do not checkpoint right after an image chunk
                     }
 
                     // If using an alora, there may be uncached tokens that come
@@ -2473,8 +2477,6 @@ struct server_context_impl {
                         alora_disabled_id = enabled_loras[0];
                     }
 
-                    bool do_checkpoint = params_base.n_ctx_checkpoints > 0;
-
                     // make checkpoints only for completion tasks
                     do_checkpoint = do_checkpoint && slot.task->type == SERVER_TASK_TYPE_COMPLETION;