From 9e897d44390f3196c6680e1847c3693354de00ed Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Fri, 4 Oct 2024 12:04:54 +0300
Subject: [PATCH] common : sanity check for non-NULL tokens

ggml-ci
---
 common/common.cpp | 30 +++++++++++++++++++++++++++++-
 src/llama-vocab.h | 18 +++++++++---------
 src/llama.cpp     |  2 +-
 3 files changed, 39 insertions(+), 11 deletions(-)

diff --git a/common/common.cpp b/common/common.cpp
index a0611f3d1..29df16c95 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -838,6 +838,31 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
         return iparams;
     }
 
+    if (params.reranking) {
+        bool ok = true;
+
+        if (llama_token_bos(model) == LLAMA_TOKEN_NULL) {
+            LOG_WRN("%s: warning: model does not have a  BOS token, reranking will not work\n", __func__);
+            ok = false;
+        }
+
+        if (llama_token_eos(model) == LLAMA_TOKEN_NULL) {
+            LOG_WRN("%s: warning: model does not have an EOS token, reranking will not work\n", __func__);
+            ok = false;
+        }
+
+        if (llama_token_sep(model) == LLAMA_TOKEN_NULL) {
+            LOG_WRN("%s: warning: model does not have a  SEP token, reranking will not work\n", __func__);
+            ok = false;
+        }
+
+        if (!ok) {
+            llama_free_model(model);
+
+            return iparams;
+        }
+    }
+
     auto cparams = llama_context_params_from_gpt_params(params);
 
     llama_context * lctx = llama_new_context_with_model(model, cparams);
@@ -855,6 +880,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
         if (cvec.n_embd == -1) {
             llama_free(lctx);
             llama_free_model(model);
+
             return iparams;
         }
 
@@ -867,6 +893,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
         if (err) {
             llama_free(lctx);
             llama_free_model(model);
+
             return iparams;
         }
     }
@@ -889,7 +916,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
         llama_lora_adapters_apply(lctx, iparams.lora_adapters);
     }
 
-    if (params.sparams.ignore_eos && llama_token_eos(model) == -1) {
+    if (params.sparams.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) {
         LOG_WRN("%s: warning: model does not have an EOS token, ignoring --ignore-eos\n", __func__);
         params.sparams.ignore_eos = false;
     }
@@ -930,6 +957,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
 
     iparams.model   = model;
     iparams.context = lctx;
+
     return iparams;
 }
 
diff --git a/src/llama-vocab.h b/src/llama-vocab.h
index 069bdc423..28bad9135 100644
--- a/src/llama-vocab.h
+++ b/src/llama-vocab.h
@@ -40,17 +40,17 @@ struct llama_vocab {
     id special_bos_id  = 1;
     id special_eos_id  = 2;
     id special_unk_id  = 0;
-    id special_sep_id  = -1;
-    id special_pad_id  = -1;
-    id special_cls_id  = -1;
-    id special_mask_id = -1;
+    id special_sep_id  = LLAMA_TOKEN_NULL;
+    id special_pad_id  = LLAMA_TOKEN_NULL;
+    id special_cls_id  = LLAMA_TOKEN_NULL;
+    id special_mask_id = LLAMA_TOKEN_NULL;
 
     id linefeed_id       = 13;
-    id special_prefix_id = -1;
-    id special_suffix_id = -1;
-    id special_middle_id = -1;
-    id special_eot_id    = -1; // TODO: move above after "eos_id", and here add "file separator" token
-    id special_eom_id    = -1;
+    id special_prefix_id = LLAMA_TOKEN_NULL;
+    id special_suffix_id = LLAMA_TOKEN_NULL;
+    id special_middle_id = LLAMA_TOKEN_NULL;
+    id special_eot_id    = LLAMA_TOKEN_NULL; // TODO: move above after "eos_id", and here add "file separator" token
+    id special_eom_id    = LLAMA_TOKEN_NULL;
 
     // set of all tokens that cause "end of generation"
     std::set<id> special_eog_ids;
diff --git a/src/llama.cpp b/src/llama.cpp
index 3443b0689..bf6fd9277 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -2412,7 +2412,7 @@ struct llama_hparams {
 
     // needed by encoder-decoder models (e.g. T5, FLAN-T5)
     // ref: https://github.com/ggerganov/llama.cpp/pull/8141
-    llama_token dec_start_token_id = -1;
+    llama_token dec_start_token_id = LLAMA_TOKEN_NULL;
 
     enum llama_pooling_type      pooling_type            = LLAMA_POOLING_TYPE_NONE;
     enum llama_rope_type         rope_type               = LLAMA_ROPE_TYPE_NONE;