diff --git a/common/speculative.h b/common/speculative.h
index e9595b4bb9..7fb9c46a04 100644
--- a/common/speculative.h
+++ b/common/speculative.h
@@ -99,7 +99,7 @@ struct common_speculative_session {
 
     ~common_speculative_session();
 
-    // dont copy
+    // don't copy
     common_speculative_session(const common_speculative_session &) = delete;
     common_speculative_session & operator=(const common_speculative_session &) = delete;
 
diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp
index bbbd0e2154..db9ac5c9b6 100644
--- a/tools/server/server-context.cpp
+++ b/tools/server/server-context.cpp
@@ -659,7 +659,7 @@ private:
         size_t create_checkpoint() override {
             const auto pos_min = llama_memory_seq_pos_min(llama_get_memory(ctx_impl.ctx), slot.id);
             const auto pos_max = llama_memory_seq_pos_max(llama_get_memory(ctx_impl.ctx), slot.id);
-            const auto n_tokens_cur = batch.n_tokens;
+            const auto n_tokens_cur = 0; // TODO was ctx_impl.batch.n_tokens; The draft model doesn't change the prompt?
             const auto & cur_with_size = ctx_impl.get_checkpoint(slot, n_tokens_cur, pos_min, pos_max);
             auto & cur = cur_with_size.checkpoint;