From 9396c7bbaf93a0d0f49c6b92cbb8635e5a2e81be Mon Sep 17 00:00:00 2001
From: toyer <2042519524@qq.com>
Date: Wed, 26 Jun 2024 02:16:12 +0000
Subject: [PATCH] set <|endoftext|> as eos and <|user|> as eot

---
 convert-hf-to-gguf.py |  2 +-
 llama.cpp             | 13 -------------
 2 files changed, 1 insertion(+), 14 deletions(-)

diff --git a/convert-hf-to-gguf.py b/convert-hf-to-gguf.py
index 4f6308d46..363c09720 100755
--- a/convert-hf-to-gguf.py
+++ b/convert-hf-to-gguf.py
@@ -3048,9 +3048,9 @@ class ChatGLMModel(Model):
         special_vocab.chat_template = "ChatGLM4"
         special_vocab.merges = merges
         # only add special tokens when they were not already loaded from config.json
-        # if len(special_vocab.special_token_ids) == 0:
         special_vocab._set_special_token("bos", tokenizer.get_added_vocab()["<|endoftext|>"])
         special_vocab._set_special_token("eos", tokenizer.get_added_vocab()["<|endoftext|>"])
+        special_vocab._set_special_token("eot", tokenizer.get_added_vocab()["<|user|>"])
         # this one is usually not in config.json anyway
         special_vocab._set_special_token("unk", tokenizer.get_added_vocab()["<|endoftext|>"])
         special_vocab.add_to_gguf(self.gguf_writer)
diff --git a/llama.cpp b/llama.cpp
index aea08fa52..7a48c1115 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -18832,19 +18832,6 @@ llama_token_attr llama_token_get_attr(const struct llama_model * model, llama_to
 }
 
 bool llama_token_is_eog(const struct llama_model * model, llama_token token) {
-    auto arch_name = llama_model_arch_name(model->arch);
-    auto vocab_type = model->vocab.type;
-    if (strcmp(arch_name, "chatglm") == 0) {
-        if (LLAMA_VOCAB_TYPE_BPE == vocab_type) { // glm4
-            return token != -1 && (
-                token == llama_token_eos(model) ||
-                token == llama_token_eot(model) ||
-                token == 151329 ||
-                token == 151336 ||
-                token == 151338
-            );
-        }
-    }
     return token != -1 && (
         token == llama_token_eos(model) ||
         token == llama_token_eot(model)