patch: always add token to cache_tokens (#2459)

llm/patches/01-cache.diff

@@ -1,30 +1,21 @@
 diff --git a/examples/server/server.cpp b/examples/server/server.cpp
-index d86d7e04..7d71c766 100644
+index d86d7e04..2694e92e 100644
 --- a/examples/server/server.cpp
 +++ b/examples/server/server.cpp
-@@ -1598,12 +1598,6 @@ struct llama_server_context
-                         LOG_TEE("slot %d : in cache: %i tokens | to process: %i tokens\n", slot.id, slot.n_past, slot.num_prompt_tokens_processed);
-                     }
- 
--                    LOG_TEE("slot %d : kv cache rm - [%d, end)\n", slot.id, (int) system_tokens.size() + slot.n_past);
--
--                    llama_kv_cache_seq_rm(ctx, slot.id, system_tokens.size() + slot.n_past, -1);
--
--                    slot.cache_tokens = prompt_tokens;
--
-                     if (slot.n_past == slot.num_prompt_tokens && slot.n_past > 0)
-                     {
-                         // we have to evaluate at least 1 token to generate logits.
-@@ -1615,6 +1609,12 @@ struct llama_server_context
-                         }
-                     }
- 
-+                    LOG_TEE("slot %d : kv cache rm - [%d, end)\n", slot.id, (int) system_tokens.size() + slot.n_past);
+@@ -901,13 +901,15 @@ struct llama_server_context
+                 slot.sent_count += result.text_to_send.size();
+                 // add the token to slot queue and cache
+             }
+-            slot.add_token_string(result);
 +
-+                    llama_kv_cache_seq_rm(ctx, slot.id, system_tokens.size() + slot.n_past, -1);
-+
-+                    slot.cache_tokens = prompt_tokens;
+             if (slot.params.stream)
+             {
+                 send_partial_response(slot, result);
+             }
+         }
+ 
++        slot.add_token_string(result);
 +
-                     LOG_VERBOSE("prompt ingested", {
-                                                     {"n_past",  slot.n_past},
-                                                     {"cached",  tokens_to_str(ctx, slot.cache_tokens.cbegin(), slot.cache_tokens.cbegin() + slot.n_past)},
+         if (incomplete)
+         {
+             slot.has_next_token = true;