'优化模型量化损失'

README.md

@@ -16,7 +16,7 @@ tags:
 
 
 ### 【模型更新日期】
-``` 2024-06-05 23:50 ```
+``` 2024-06-18 ```
 
 ### 【模型大小】
 `11.0GB`
@@ -29,6 +29,9 @@ tags:
 ### 【更新日志】
 
 ```
+2004-06-18
+优化模型量化损失
+
 2004-06-05 23:50
 首次commit
 

config.json

@@ -28,7 +28,7 @@
   "hidden_dropout": 0.0,
   "hidden_size": 4096,
   "kv_channels": 128,
-  "layernorm_epsilon": 1e-06,
+  "layernorm_epsilon": 1.5625e-07,
   "model_type": "chatglm",
   "multi_query_attention": true,
   "multi_query_group_num": 2,

generation_config.json

@@ -5,6 +5,5 @@
     151336,
     151338
   ],
-  "pad_token_id": 151329,
   "transformers_version": "4.40.2"
 }

model-00001-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c79e4ae12da0ae7b9b233f2fe8b6b3a46269c58f38c596dd030ab0ed7845505
+oid sha256:1f9f259f678707d482f9e75b80f4a4e964fe4e3757cdcf360d923931e71456f5
 size 4970796416

model-00002-of-00003.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c82ba03bd76f42c804706423fc4be54486cf53b255b86798acce637e318d14b4
+oid sha256:939de54bfc2b87aa79fce72b92d4da8743b09d662b7e92f8466f613fb0aa61eb
 size 4819195016

modeling_chatglm.py

@@ -324,7 +324,7 @@ class SelfAttention(torch.nn.Module):
         )
 
     def forward(
-            self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True
+            self, hidden_states, attention_mask, rotary_pos_emb=None, kv_cache=None, use_cache=True
     ):
         # hidden_states: [b, sq, h]