Sengxian commited on
Commit
d11c6aa
0 Parent(s):

Add chatglm-6b

Browse files
.gitattributes ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tflite filter=lfs diff=lfs merge=lfs -text
29
+ *.tgz filter=lfs diff=lfs merge=lfs -text
30
+ *.wasm filter=lfs diff=lfs merge=lfs -text
31
+ *.xz filter=lfs diff=lfs merge=lfs -text
32
+ *.zip filter=lfs diff=lfs merge=lfs -text
33
+ *.zst filter=lfs diff=lfs merge=lfs -text
34
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
LICENSE ADDED
@@ -0,0 +1,201 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Apache License
2
+ Version 2.0, January 2004
3
+ http://www.apache.org/licenses/
4
+
5
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6
+
7
+ 1. Definitions.
8
+
9
+ "License" shall mean the terms and conditions for use, reproduction,
10
+ and distribution as defined by Sections 1 through 9 of this document.
11
+
12
+ "Licensor" shall mean the copyright owner or entity authorized by
13
+ the copyright owner that is granting the License.
14
+
15
+ "Legal Entity" shall mean the union of the acting entity and all
16
+ other entities that control, are controlled by, or are under common
17
+ control with that entity. For the purposes of this definition,
18
+ "control" means (i) the power, direct or indirect, to cause the
19
+ direction or management of such entity, whether by contract or
20
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
21
+ outstanding shares, or (iii) beneficial ownership of such entity.
22
+
23
+ "You" (or "Your") shall mean an individual or Legal Entity
24
+ exercising permissions granted by this License.
25
+
26
+ "Source" form shall mean the preferred form for making modifications,
27
+ including but not limited to software source code, documentation
28
+ source, and configuration files.
29
+
30
+ "Object" form shall mean any form resulting from mechanical
31
+ transformation or translation of a Source form, including but
32
+ not limited to compiled object code, generated documentation,
33
+ and conversions to other media types.
34
+
35
+ "Work" shall mean the work of authorship, whether in Source or
36
+ Object form, made available under the License, as indicated by a
37
+ copyright notice that is included in or attached to the work
38
+ (an example is provided in the Appendix below).
39
+
40
+ "Derivative Works" shall mean any work, whether in Source or Object
41
+ form, that is based on (or derived from) the Work and for which the
42
+ editorial revisions, annotations, elaborations, or other modifications
43
+ represent, as a whole, an original work of authorship. For the purposes
44
+ of this License, Derivative Works shall not include works that remain
45
+ separable from, or merely link (or bind by name) to the interfaces of,
46
+ the Work and Derivative Works thereof.
47
+
48
+ "Contribution" shall mean any work of authorship, including
49
+ the original version of the Work and any modifications or additions
50
+ to that Work or Derivative Works thereof, that is intentionally
51
+ submitted to Licensor for inclusion in the Work by the copyright owner
52
+ or by an individual or Legal Entity authorized to submit on behalf of
53
+ the copyright owner. For the purposes of this definition, "submitted"
54
+ means any form of electronic, verbal, or written communication sent
55
+ to the Licensor or its representatives, including but not limited to
56
+ communication on electronic mailing lists, source code control systems,
57
+ and issue tracking systems that are managed by, or on behalf of, the
58
+ Licensor for the purpose of discussing and improving the Work, but
59
+ excluding communication that is conspicuously marked or otherwise
60
+ designated in writing by the copyright owner as "Not a Contribution."
61
+
62
+ "Contributor" shall mean Licensor and any individual or Legal Entity
63
+ on behalf of whom a Contribution has been received by Licensor and
64
+ subsequently incorporated within the Work.
65
+
66
+ 2. Grant of Copyright License. Subject to the terms and conditions of
67
+ this License, each Contributor hereby grants to You a perpetual,
68
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69
+ copyright license to reproduce, prepare Derivative Works of,
70
+ publicly display, publicly perform, sublicense, and distribute the
71
+ Work and such Derivative Works in Source or Object form.
72
+
73
+ 3. Grant of Patent License. Subject to the terms and conditions of
74
+ this License, each Contributor hereby grants to You a perpetual,
75
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76
+ (except as stated in this section) patent license to make, have made,
77
+ use, offer to sell, sell, import, and otherwise transfer the Work,
78
+ where such license applies only to those patent claims licensable
79
+ by such Contributor that are necessarily infringed by their
80
+ Contribution(s) alone or by combination of their Contribution(s)
81
+ with the Work to which such Contribution(s) was submitted. If You
82
+ institute patent litigation against any entity (including a
83
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
84
+ or a Contribution incorporated within the Work constitutes direct
85
+ or contributory patent infringement, then any patent licenses
86
+ granted to You under this License for that Work shall terminate
87
+ as of the date such litigation is filed.
88
+
89
+ 4. Redistribution. You may reproduce and distribute copies of the
90
+ Work or Derivative Works thereof in any medium, with or without
91
+ modifications, and in Source or Object form, provided that You
92
+ meet the following conditions:
93
+
94
+ (a) You must give any other recipients of the Work or
95
+ Derivative Works a copy of this License; and
96
+
97
+ (b) You must cause any modified files to carry prominent notices
98
+ stating that You changed the files; and
99
+
100
+ (c) You must retain, in the Source form of any Derivative Works
101
+ that You distribute, all copyright, patent, trademark, and
102
+ attribution notices from the Source form of the Work,
103
+ excluding those notices that do not pertain to any part of
104
+ the Derivative Works; and
105
+
106
+ (d) If the Work includes a "NOTICE" text file as part of its
107
+ distribution, then any Derivative Works that You distribute must
108
+ include a readable copy of the attribution notices contained
109
+ within such NOTICE file, excluding those notices that do not
110
+ pertain to any part of the Derivative Works, in at least one
111
+ of the following places: within a NOTICE text file distributed
112
+ as part of the Derivative Works; within the Source form or
113
+ documentation, if provided along with the Derivative Works; or,
114
+ within a display generated by the Derivative Works, if and
115
+ wherever such third-party notices normally appear. The contents
116
+ of the NOTICE file are for informational purposes only and
117
+ do not modify the License. You may add Your own attribution
118
+ notices within Derivative Works that You distribute, alongside
119
+ or as an addendum to the NOTICE text from the Work, provided
120
+ that such additional attribution notices cannot be construed
121
+ as modifying the License.
122
+
123
+ You may add Your own copyright statement to Your modifications and
124
+ may provide additional or different license terms and conditions
125
+ for use, reproduction, or distribution of Your modifications, or
126
+ for any such Derivative Works as a whole, provided Your use,
127
+ reproduction, and distribution of the Work otherwise complies with
128
+ the conditions stated in this License.
129
+
130
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
131
+ any Contribution intentionally submitted for inclusion in the Work
132
+ by You to the Licensor shall be under the terms and conditions of
133
+ this License, without any additional terms or conditions.
134
+ Notwithstanding the above, nothing herein shall supersede or modify
135
+ the terms of any separate license agreement you may have executed
136
+ with Licensor regarding such Contributions.
137
+
138
+ 6. Trademarks. This License does not grant permission to use the trade
139
+ names, trademarks, service marks, or product names of the Licensor,
140
+ except as required for reasonable and customary use in describing the
141
+ origin of the Work and reproducing the content of the NOTICE file.
142
+
143
+ 7. Disclaimer of Warranty. Unless required by applicable law or
144
+ agreed to in writing, Licensor provides the Work (and each
145
+ Contributor provides its Contributions) on an "AS IS" BASIS,
146
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147
+ implied, including, without limitation, any warranties or conditions
148
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149
+ PARTICULAR PURPOSE. You are solely responsible for determining the
150
+ appropriateness of using or redistributing the Work and assume any
151
+ risks associated with Your exercise of permissions under this License.
152
+
153
+ 8. Limitation of Liability. In no event and under no legal theory,
154
+ whether in tort (including negligence), contract, or otherwise,
155
+ unless required by applicable law (such as deliberate and grossly
156
+ negligent acts) or agreed to in writing, shall any Contributor be
157
+ liable to You for damages, including any direct, indirect, special,
158
+ incidental, or consequential damages of any character arising as a
159
+ result of this License or out of the use or inability to use the
160
+ Work (including but not limited to damages for loss of goodwill,
161
+ work stoppage, computer failure or malfunction, or any and all
162
+ other commercial damages or losses), even if such Contributor
163
+ has been advised of the possibility of such damages.
164
+
165
+ 9. Accepting Warranty or Additional Liability. While redistributing
166
+ the Work or Derivative Works thereof, You may choose to offer,
167
+ and charge a fee for, acceptance of support, warranty, indemnity,
168
+ or other liability obligations and/or rights consistent with this
169
+ License. However, in accepting such obligations, You may act only
170
+ on Your own behalf and on Your sole responsibility, not on behalf
171
+ of any other Contributor, and only if You agree to indemnify,
172
+ defend, and hold each Contributor harmless for any liability
173
+ incurred by, or claims asserted against, such Contributor by reason
174
+ of your accepting any such warranty or additional liability.
175
+
176
+ END OF TERMS AND CONDITIONS
177
+
178
+ APPENDIX: How to apply the Apache License to your work.
179
+
180
+ To apply the Apache License to your work, attach the following
181
+ boilerplate notice, with the fields enclosed by brackets "[]"
182
+ replaced with your own identifying information. (Don't include
183
+ the brackets!) The text should be enclosed in the appropriate
184
+ comment syntax for the file format. We also recommend that a
185
+ file or class name and description of purpose be included on the
186
+ same "printed page" as the copyright notice for easier
187
+ identification within third-party archives.
188
+
189
+ Copyright Zhengxiao Du
190
+
191
+ Licensed under the Apache License, Version 2.0 (the "License");
192
+ you may not use this file except in compliance with the License.
193
+ You may obtain a copy of the License at
194
+
195
+ http://www.apache.org/licenses/LICENSE-2.0
196
+
197
+ Unless required by applicable law or agreed to in writing, software
198
+ distributed under the License is distributed on an "AS IS" BASIS,
199
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200
+ See the License for the specific language governing permissions and
201
+ limitations under the License.
MODEL_LICENSE ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ The GLM-130B License
2
+
3
+ 1. Definitions
4
+
5
+ “Licensor” means the GLM-130B Model Team that distributes its Software.
6
+
7
+ “Software” means the GLM-130B model parameters made available under this license.
8
+
9
+ 2. License Grant
10
+
11
+ Subject to the terms and conditions of this License, the Licensor hereby grants to you a non-exclusive, worldwide, non-transferable, non-sublicensable, revocable, royalty-free copyright license to use the Software solely for your non-commercial research purposes.
12
+
13
+ The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
14
+
15
+ 3. Restriction
16
+
17
+ You will not use, copy, modify, merge, publish, distribute, reproduce, or create derivative works of the Software, in whole or in part, for any commercial, military, or illegal purposes.
18
+
19
+ You will not use the Software for any act that may undermine China's national security and national unity, harm the public interest of society, or infringe upon the rights and interests of human beings.
20
+
21
+ 4. Disclaimer
22
+
23
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
24
+
25
+ 5. Limitation of Liability
26
+
27
+ EXCEPT TO THE EXTENT PROHIBITED BY APPLICABLE LAW, IN NO EVENT AND UNDER NO LEGAL THEORY, WHETHER BASED IN TORT, NEGLIGENCE, CONTRACT, LIABILITY, OR OTHERWISE WILL ANY LICENSOR BE LIABLE TO YOU FOR ANY DIRECT, INDIRECT, SPECIAL, INCIDENTAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES, OR ANY OTHER COMMERCIAL LOSSES, EVEN IF THE LICENSOR HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
28
+
29
+ 6. Dispute Resolution
30
+
31
+ This license shall be governed and construed in accordance with the laws of People’s Republic of China. Any dispute arising from or in connection with this License shall be submitted to Haidian District People's Court in Beijing.
32
+
33
+ Note that the license is subject to update to a more comprehensive version. For any questions related to the license and copyright, please contact us at glm-130b@googlegroups.com.
README.md ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - zh
4
+ - en
5
+ tags:
6
+ - glm
7
+ - chatglm
8
+ - thudm
9
+ ---
10
+ # ChatGLM-6B
11
+ ## 介绍
12
+ ChatGLM-6B 是一个开源的、支持中英双语问答和对话的预训练语言模型,基于 [GLM](https://github.com/THUDM/GLM) 架构,具有 62 亿参数。ChatGLM-6B 使用了和 ChatGLM(内测中,地址 [https://chatglm.cn](https://chatglm.cn))相同的技术面向中文问答和对话进行优化。
13
+
14
+ ## 使用方式
15
+ 使用前请先安装`transformers>=4.23.1`和`icetk`。
16
+
17
+ ```shell
18
+ pip install "transformers>=4.23.1,icetk"
19
+ ```
20
+
21
+ ### 代码调用
22
+
23
+ 可以通过如下代码调用 ChatGLM-6B 模型来生成对话。
24
+
25
+ ```python
26
+ from transformers import AutoTokenizer, AutoModel
27
+
28
+ tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True)
29
+ model = AutoModel.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True).half().cuda()
30
+ model = model.eval()
31
+
32
+ history = []
33
+ query = "你好"
34
+ response, history = model.chat(tokenizer, query, history=history)
35
+ print(response)
36
+
37
+ query = "晚上睡不着应该怎么办"
38
+ response, history = model.chat(tokenizer, query, history=history)
39
+ print(history)
40
+ ```
41
+
42
+ 关于更多的使用说明,以及如何运行命令行和Web版本的demo,请参考我们的[Github repo](https://github.com/THUDM/ChatGLM-6B)。
43
+
44
+ ## INT8 量化
45
+ 默认情况下,模型以 FP16 精度加载,运行上述代码需要大概 13GB 显存。如果你的 GPU 显存有限,可以尝试使用 `transformers` 提供的 8bit 量化功能,即将代码中的
46
+
47
+ ```python
48
+ model = AutoModel.from_pretrained("THUDM/chatglm-6b", trust_remote_code=True).half().cuda()
49
+ ```
50
+
51
+ 替换为
52
+
53
+ ```python
54
+ model = AutoModel.from_pretrained("THUDM/chatglm-6b", device_map="auto", load_in_8bit=True, trust_remote_code=True)
55
+ ```
56
+
57
+ 使用 8-bit 量化之后大约需要 9.5GB 的 GPU 显存。
58
+
59
+ ## 引用
60
+
61
+ 如果你觉得我们的工作有帮助的话,请考虑引用下列论文
62
+
63
+ ```
64
+ @inproceedings{
65
+ zeng2023glm-130b,
66
+ title={{GLM}-130B: An Open Bilingual Pre-trained Model},
67
+ author={Aohan Zeng and Xiao Liu and Zhengxiao Du and Zihan Wang and Hanyu Lai and Ming Ding and Zhuoyi Yang and Yifan Xu and Wendi Zheng and Xiao Xia and Weng Lam Tam and Zixuan Ma and Yufei Xue and Jidong Zhai and Wenguang Chen and Zhiyuan Liu and Peng Zhang and Yuxiao Dong and Jie Tang},
68
+ booktitle={The Eleventh International Conference on Learning Representations (ICLR)},
69
+ year={2023},
70
+ url={https://openreview.net/forum?id=-Aw0rrrPUF}
71
+ }
72
+ ```
73
+ ```
74
+ @inproceedings{du2022glm,
75
+ title={GLM: General Language Model Pretraining with Autoregressive Blank Infilling},
76
+ author={Du, Zhengxiao and Qian, Yujie and Liu, Xiao and Ding, Ming and Qiu, Jiezhong and Yang, Zhilin and Tang, Jie},
77
+ booktitle={Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
78
+ pages={320--335},
79
+ year={2022}
80
+ }
81
+ ```
config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "THUDM/chatglm-6b",
3
+ "architectures": [
4
+ "ChatGLMModel"
5
+ ],
6
+ "auto_map": {
7
+ "AutoConfig": "configuration_chatglm.ChatGLMConfig",
8
+ "AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration",
9
+ "AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration"
10
+ },
11
+ "bos_token_id": 150004,
12
+ "eos_token_id": 150005,
13
+ "hidden_size": 4096,
14
+ "inner_hidden_size": 16384,
15
+ "layernorm_epsilon": 1e-05,
16
+ "max_sequence_length": 2048,
17
+ "model_type": "chatglm",
18
+ "num_attention_heads": 32,
19
+ "num_layers": 28,
20
+ "position_encoding_2d": true,
21
+ "torch_dtype": "float16",
22
+ "transformers_version": "4.23.1",
23
+ "use_cache": true,
24
+ "vocab_size": 150528
25
+ }
configuration_chatglm.py ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """ ChatGLM model configuration """
2
+
3
+ from transformers.configuration_utils import PretrainedConfig
4
+ from transformers.utils import logging
5
+
6
+ logger = logging.get_logger(__name__)
7
+
8
+
9
+ class ChatGLMConfig(PretrainedConfig):
10
+ r"""
11
+ This is the configuration class to store the configuration of a [`~ChatGLMModel`].
12
+ It is used to instantiate an ChatGLM model according to the specified arguments, defining the model
13
+ architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of
14
+ the ChatGLM-6B [THUDM/ChatGLM-6B](https://huggingface.co/THUDM/chatglm-6b) architecture.
15
+
16
+ Configuration objects inherit from [`PretrainedConfig`] and can be used
17
+ to control the model outputs. Read the documentation from [`PretrainedConfig`]
18
+ for more information.
19
+
20
+
21
+ Args:
22
+ vocab_size (`int`, *optional*, defaults to 150528):
23
+ Vocabulary size of the ChatGLM-6B model. Defines the number of different tokens that can be represented by the
24
+ `inputs_ids` passed when calling [`~ChatGLMModel`] or
25
+ [`~TFChatGLMModel`].
26
+ hidden_size (`int`, *optional*, defaults to 4096):
27
+ Dimension of the encoder layers and the pooler layer.
28
+ num_hidden_layers (`int`, *optional*, defaults to 28):
29
+ Number of hidden layers in the Transformer encoder.
30
+ num_attention_heads (`int`, *optional*, defaults to 32):
31
+ Number of attention heads for each attention layer in the Transformer encoder.
32
+ inner_hidden_size (`int`, *optional*, defaults to 16384):
33
+ Dimension of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
34
+ max_sequence_length (`int`, *optional*, defaults to 512):
35
+ The maximum sequence length that this model might ever be used with.
36
+ Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
37
+ layernorm_epsilon (`float`, *optional*, defaults to 1e-5):
38
+ The epsilon used by the layer normalization layers.
39
+ use_cache (`bool`, *optional*, defaults to `True`):
40
+ Whether the model should return the last key/values attentions (not used by all models).
41
+ Example:
42
+
43
+ ```python
44
+ >>> from configuration_chatglm import ChatGLMConfig
45
+ >>> from modeling_chatglm import ChatGLMModel
46
+
47
+ >>> # Initializing a ChatGLM-6B THUDM/ChatGLM-6B style configuration
48
+ >>> configuration = ChatGLMConfig()
49
+
50
+ >>> # Initializing a model from the THUDM/ChatGLM-6B style configuration
51
+ >>> model = ChatGLMModel(configuration)
52
+
53
+ >>> # Accessing the model configuration
54
+ >>> configuration = model.config
55
+ ```
56
+ """
57
+ model_type = "chatglm"
58
+
59
+ def __init__(
60
+ self,
61
+ vocab_size=150528,
62
+ hidden_size=4096,
63
+ num_layers=28,
64
+ num_attention_heads=32,
65
+ layernorm_epsilon=1e-5,
66
+ use_cache=False,
67
+ bos_token_id=150004,
68
+ eos_token_id=150005,
69
+ pad_token_id=0,
70
+ max_sequence_length=2048,
71
+ inner_hidden_size=16384,
72
+ position_encoding_2d=True,
73
+ **kwargs
74
+ ):
75
+ self.num_layers = num_layers
76
+ self.vocab_size = vocab_size
77
+ self.hidden_size = hidden_size
78
+ self.num_attention_heads = num_attention_heads
79
+ self.max_sequence_length = max_sequence_length
80
+ self.layernorm_epsilon = layernorm_epsilon
81
+ self.inner_hidden_size = inner_hidden_size
82
+ self.use_cache = use_cache
83
+ self.bos_token_id = bos_token_id
84
+ self.eos_token_id = eos_token_id
85
+ self.pad_token_id = pad_token_id
86
+ self.position_encoding_2d = position_encoding_2d
87
+ super().__init__(
88
+ pad_token_id=pad_token_id,
89
+ bos_token_id=bos_token_id,
90
+ eos_token_id=eos_token_id,
91
+ **kwargs
92
+ )
ice_text.model ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:99871e0c85db81ad7af1028854fd091cd5778c8414ae9d94bbbc10d02c831c21
3
+ size 2699926
modeling_chatglm.py ADDED
@@ -0,0 +1,1152 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """ PyTorch ChatGLM model. """
2
+
3
+ import math
4
+ import copy
5
+ import os
6
+
7
+ import torch
8
+ import torch.utils.checkpoint
9
+ import torch.nn.functional as F
10
+ from torch import nn
11
+ from torch.nn import CrossEntropyLoss, LayerNorm
12
+ from torch.nn.utils import skip_init
13
+ from typing import Optional, Tuple, Union, List
14
+
15
+ from transformers.utils import (
16
+ add_code_sample_docstrings,
17
+ add_start_docstrings,
18
+ add_start_docstrings_to_model_forward,
19
+ )
20
+ from transformers.modeling_outputs import (
21
+ BaseModelOutputWithPast,
22
+ CausalLMOutputWithPast,
23
+ BaseModelOutputWithPastAndCrossAttentions,
24
+ )
25
+ from transformers.modeling_utils import PreTrainedModel
26
+
27
+ from transformers.utils import logging
28
+ from .configuration_chatglm import ChatGLMConfig
29
+
30
+ # flags required to enable jit fusion kernels
31
+ torch._C._jit_set_profiling_mode(False)
32
+ torch._C._jit_set_profiling_executor(False)
33
+ torch._C._jit_override_can_fuse_on_cpu(True)
34
+ torch._C._jit_override_can_fuse_on_gpu(True)
35
+
36
+ logger = logging.get_logger(__name__)
37
+
38
+ _CHECKPOINT_FOR_DOC = "THUDM/ChatGLM-6B"
39
+ _CONFIG_FOR_DOC = "ChatGLM6BConfig"
40
+
41
+ CHATGLM_6B_PRETRAINED_MODEL_ARCHIVE_LIST = [
42
+ "THUDM/chatglm-6b",
43
+ # See all ChatGLM-6B models at https://huggingface.co/models?filter=chatglm
44
+ ]
45
+
46
+
47
+ def load_tf_weights_in_chatglm_6b(model, config, tf_checkpoint_path):
48
+ """Load tf checkpoints in a pytorch model."""
49
+ try:
50
+ import re
51
+
52
+ import numpy as np
53
+ import tensorflow as tf
54
+ except ImportError:
55
+ logger.error(
56
+ "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
57
+ "https://www.tensorflow.org/install/ for installation instructions."
58
+ )
59
+ raise
60
+ tf_path = os.path.abspath(tf_checkpoint_path)
61
+ logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
62
+ # Load weights from TF model
63
+ init_vars = tf.train.list_variables(tf_path)
64
+ names = []
65
+ arrays = []
66
+ for name, shape in init_vars:
67
+ logger.info(f"Loading TF weight {name} with shape {shape}")
68
+ array = tf.train.load_variable(tf_path, name)
69
+ names.append(name)
70
+ arrays.append(array)
71
+
72
+ for name, array in zip(names, arrays):
73
+ name = name.split("/")
74
+ # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
75
+ # which are not required for using pretrained model
76
+ if any(
77
+ n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
78
+ for n in name
79
+ ):
80
+ logger.info(f"Skipping {'/'.join(name)}")
81
+ continue
82
+ pointer = model
83
+ for m_name in name:
84
+ if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
85
+ scope_names = re.split(r"_(\d+)", m_name)
86
+ else:
87
+ scope_names = [m_name]
88
+ if scope_names[0] == "kernel" or scope_names[0] == "gamma":
89
+ pointer = getattr(pointer, "weight")
90
+ elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
91
+ pointer = getattr(pointer, "bias")
92
+ elif scope_names[0] == "output_weights":
93
+ pointer = getattr(pointer, "weight")
94
+ elif scope_names[0] == "squad":
95
+ pointer = getattr(pointer, "classifier")
96
+ else:
97
+ try:
98
+ pointer = getattr(pointer, scope_names[0])
99
+ except AttributeError:
100
+ logger.info(f"Skipping {'/'.join(name)}")
101
+ continue
102
+ if len(scope_names) >= 2:
103
+ num = int(scope_names[1])
104
+ pointer = pointer[num]
105
+ if m_name[-11:] == "_embeddings":
106
+ pointer = getattr(pointer, "weight")
107
+ elif m_name == "kernel":
108
+ array = np.transpose(array)
109
+ try:
110
+ assert (
111
+ pointer.shape == array.shape
112
+ ), f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
113
+ except AssertionError as e:
114
+ e.args += (pointer.shape, array.shape)
115
+ raise
116
+ logger.info(f"Initialize PyTorch weight {name}")
117
+ pointer.data = torch.from_numpy(array)
118
+ return model
119
+
120
+
121
+ @torch.jit.script
122
+ def gelu_impl(x):
123
+ """OpenAI's gelu implementation."""
124
+ return 0.5 * x * (1.0 + torch.tanh(0.7978845608028654 * x *
125
+ (1.0 + 0.044715 * x * x)))
126
+
127
+
128
+ def gelu(x):
129
+ return gelu_impl(x)
130
+
131
+
132
+ class RotaryEmbedding(torch.nn.Module):
133
+ def __init__(self, dim, base=10000, precision=torch.half, learnable=False):
134
+ super().__init__()
135
+ inv_freq = 1. / (base ** (torch.arange(0, dim, 2).float() / dim))
136
+ inv_freq = inv_freq.half()
137
+ self.learnable = learnable
138
+ if learnable:
139
+ self.inv_freq = torch.nn.Parameter(inv_freq)
140
+ self.max_seq_len_cached = None
141
+ else:
142
+ self.register_buffer('inv_freq', inv_freq)
143
+ self.max_seq_len_cached = None
144
+ self.cos_cached = None
145
+ self.sin_cached = None
146
+ self.precision = precision
147
+
148
+ def _load_from_state_dict(self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys,
149
+ error_msgs):
150
+ pass
151
+
152
+ def forward(self, x, seq_dim=1, seq_len=None):
153
+ if seq_len is None:
154
+ seq_len = x.shape[seq_dim]
155
+ if self.max_seq_len_cached is None or (seq_len > self.max_seq_len_cached):
156
+ self.max_seq_len_cached = None if self.learnable else seq_len
157
+ t = torch.arange(seq_len, device=x.device, dtype=self.inv_freq.dtype)
158
+ freqs = torch.einsum('i,j->ij', t, self.inv_freq)
159
+ # Different from paper, but it uses a different permutation in order to obtain the same calculation
160
+ emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
161
+ if self.precision == torch.bfloat16:
162
+ emb = emb.float()
163
+
164
+ # [sx, 1 (b * np), hn]
165
+ cos_cached = emb.cos()[:, None, :]
166
+ sin_cached = emb.sin()[:, None, :]
167
+ if self.precision == torch.bfloat16:
168
+ cos_cached = cos_cached.bfloat16()
169
+ sin_cached = sin_cached.bfloat16()
170
+ if self.learnable:
171
+ return cos_cached, sin_cached
172
+ self.cos_cached, self.sin_cached = cos_cached, sin_cached
173
+ return self.cos_cached[:seq_len, ...], self.sin_cached[:seq_len, ...]
174
+
175
+
176
+ def rotate_half(x):
177
+ x1, x2 = x[..., :x.shape[-1] // 2], x[..., x.shape[-1] // 2:]
178
+ return torch.cat((-x2, x1), dim=x1.ndim - 1) # dim=-1 triggers a bug in earlier torch versions
179
+
180
+
181
+ @torch.jit.script
182
+ def apply_rotary_pos_emb_index(q, k, cos, sin, position_id):
183
+ # position_id: [sq, b], q, k: [sq, b, np, hn], cos: [sq, 1, hn] -> [sq, b, 1, hn]
184
+ cos, sin = F.embedding(position_id, cos.squeeze(1)).unsqueeze(2), \
185
+ F.embedding(position_id, sin.squeeze(1)).unsqueeze(2)
186
+ q, k = (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
187
+ return q, k
188
+
189
+
190
+ def attention_fn(
191
+ self,
192
+ query_layer,
193
+ key_layer,
194
+ value_layer,
195
+ attention_mask,
196
+ hidden_size_per_partition,
197
+ layer_id,
198
+ layer_past=None,
199
+ scaling_attention_score=True,
200
+ use_cache=False,
201
+ ):
202
+ if layer_past is not None:
203
+ past_key, past_value = layer_past
204
+ key_layer = torch.cat((past_key, key_layer), dim=0)
205
+ value_layer = torch.cat((past_value, value_layer), dim=0)
206
+
207
+ # seqlen, batch, num_attention_heads, hidden_size_per_attention_head
208
+ seq_len, b, nh, hidden_size = key_layer.shape
209
+
210
+ if use_cache:
211
+ present = (key_layer, value_layer)
212
+ else:
213
+ present = None
214
+
215
+ query_key_layer_scaling_coeff = float(layer_id + 1)
216
+ if scaling_attention_score:
217
+ query_layer = query_layer / (math.sqrt(hidden_size) * query_key_layer_scaling_coeff)
218
+
219
+ # ===================================
220
+ # Raw attention scores. [b, np, s, s]
221
+ # ===================================
222
+
223
+ # [b, np, sq, sk]
224
+ output_size = (query_layer.size(1), query_layer.size(2), query_layer.size(0), key_layer.size(0))
225
+
226
+ # [sq, b, np, hn] -> [sq, b * np, hn]
227
+ query_layer = query_layer.view(output_size[2], output_size[0] * output_size[1], -1)
228
+ # [sk, b, np, hn] -> [sk, b * np, hn]
229
+ key_layer = key_layer.view(output_size[3], output_size[0] * output_size[1], -1)
230
+
231
+ matmul_result = torch.empty(
232
+ output_size[0] * output_size[1],
233
+ output_size[2],
234
+ output_size[3],
235
+ dtype=query_layer.dtype,
236
+ device=query_layer.device,
237
+ )
238
+
239
+ matmul_result = torch.baddbmm(
240
+ matmul_result,
241
+ query_layer.transpose(0, 1), # [b * np, sq, hn]
242
+ key_layer.transpose(0, 1).transpose(1, 2), # [b * np, hn, sk]
243
+ beta=0.0,
244
+ alpha=1.0,
245
+ )
246
+
247
+ # change view to [b, np, sq, sk]
248
+ attention_scores = matmul_result.view(*output_size)
249
+
250
+ if self.scale_mask_softmax:
251
+ self.scale_mask_softmax.scale = query_key_layer_scaling_coeff
252
+ attention_probs = self.scale_mask_softmax(attention_scores, attention_mask.contiguous())
253
+ else:
254
+ if not (attention_mask == 0).all():
255
+ # if auto-regressive, skip
256
+ attention_scores.masked_fill_(attention_mask, -10000.0)
257
+
258
+ attention_scores = attention_scores.float()
259
+ attention_scores = attention_scores * query_key_layer_scaling_coeff
260
+
261
+ attention_probs = F.softmax(attention_scores, dim=-1)
262
+
263
+ attention_probs = attention_probs.half()
264
+
265
+ # =========================
266
+ # Context layer. [sq, b, hp]
267
+ # =========================
268
+
269
+ # value_layer -> context layer.
270
+ # [sk, b, np, hn] --> [b, np, sq, hn]
271
+
272
+ # context layer shape: [b, np, sq, hn]
273
+ output_size = (value_layer.size(1), value_layer.size(2), query_layer.size(0), value_layer.size(3))
274
+
275
+ # change view [sk, b * np, hn]
276
+ value_layer = value_layer.view(value_layer.size(0), output_size[0] * output_size[1], -1)
277
+
278
+ # change view [b * np, sq, sk]
279
+ attention_probs = attention_probs.view(output_size[0] * output_size[1], output_size[2], -1)
280
+
281
+ # matmul: [b * np, sq, hn]
282
+ context_layer = torch.bmm(attention_probs, value_layer.transpose(0, 1))
283
+
284
+ # change view [b, np, sq, hn]
285
+ context_layer = context_layer.view(*output_size)
286
+
287
+ # [b, np, sq, hn] --> [sq, b, np, hn]
288
+ context_layer = context_layer.permute(2, 0, 1, 3).contiguous()
289
+
290
+ # [sq, b, np, hn] --> [sq, b, hp]
291
+ new_context_layer_shape = context_layer.size()[:-2] + (hidden_size_per_partition,)
292
+ context_layer = context_layer.view(*new_context_layer_shape)
293
+
294
+ outputs = (context_layer, present, attention_probs)
295
+
296
+ return outputs
297
+
298
+
299
+ class SelfAttention(torch.nn.Module):
300
+ def __init__(self, hidden_size, num_attention_heads,
301
+ layer_id, hidden_size_per_attention_head=None, bias=True,
302
+ params_dtype=torch.float, position_encoding_2d=True):
303
+ super(SelfAttention, self).__init__()
304
+
305
+ self.layer_id = layer_id
306
+ self.hidden_size = hidden_size
307
+ self.hidden_size_per_partition = hidden_size
308
+ self.num_attention_heads = num_attention_heads
309
+ self.num_attention_heads_per_partition = num_attention_heads
310
+ self.position_encoding_2d = position_encoding_2d
311
+ self.rotary_emb = RotaryEmbedding(
312
+ self.hidden_size // (self.num_attention_heads * 2)
313
+ if position_encoding_2d
314
+ else self.hidden_size // self.num_attention_heads,
315
+ base=10000,
316
+ precision=torch.half,
317
+ learnable=False,
318
+ )
319
+
320
+ self.scale_mask_softmax = None
321
+
322
+ if hidden_size_per_attention_head is None:
323
+ self.hidden_size_per_attention_head = hidden_size // num_attention_heads
324
+ else:
325
+ self.hidden_size_per_attention_head = hidden_size_per_attention_head
326
+
327
+ self.inner_hidden_size = num_attention_heads * self.hidden_size_per_attention_head
328
+
329
+ # Strided linear layer.
330
+ self.query_key_value = skip_init(
331
+ torch.nn.Linear,
332
+ hidden_size,
333
+ 3 * self.inner_hidden_size,
334
+ bias=bias,
335
+ dtype=params_dtype,
336
+ )
337
+
338
+ self.dense = skip_init(
339
+ torch.nn.Linear,
340
+ self.inner_hidden_size,
341
+ hidden_size,
342
+ bias=bias,
343
+ dtype=params_dtype,
344
+ )
345
+
346
+ @staticmethod
347
+ def attention_mask_func(attention_scores, attention_mask):
348
+ attention_scores.masked_fill_(attention_mask, -10000.0)
349
+ return attention_scores
350
+
351
+ def split_tensor_along_last_dim(self, tensor, num_partitions,
352
+ contiguous_split_chunks=False):
353
+ """Split a tensor along its last dimension.
354
+ Arguments:
355
+ tensor: input tensor.
356
+ num_partitions: number of partitions to split the tensor
357
+ contiguous_split_chunks: If True, make each chunk contiguous
358
+ in memory.
359
+ """
360
+ # Get the size and dimension.
361
+ last_dim = tensor.dim() - 1
362
+ last_dim_size = tensor.size()[last_dim] // num_partitions
363
+ # Split.
364
+ tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
365
+ # Note: torch.split does not create contiguous tensors by default.
366
+ if contiguous_split_chunks:
367
+ return tuple(chunk.contiguous() for chunk in tensor_list)
368
+
369
+ return tensor_list
370
+
371
+ def forward(
372
+ self,
373
+ hidden_states: torch.Tensor,
374
+ position_ids,
375
+ attention_mask: torch.Tensor,
376
+ layer_id,
377
+ layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
378
+ use_cache: bool = False,
379
+ output_attentions: bool = False,
380
+ ):
381
+ """
382
+ hidden_states: [seq_len, batch, hidden_size]
383
+ attention_mask: [(1, 1), seq_len, seq_len]
384
+ """
385
+
386
+ # [seq_len, batch, 3 * hidden_size]
387
+ mixed_raw_layer = self.query_key_value(hidden_states)
388
+
389
+ # [seq_len, batch, 3 * hidden_size] --> [seq_len, batch, num_attention_heads, 3 * hidden_size_per_attention_head]
390
+ new_tensor_shape = mixed_raw_layer.size()[:-1] + (
391
+ self.num_attention_heads_per_partition,
392
+ 3 * self.hidden_size_per_attention_head,
393
+ )
394
+ mixed_raw_layer = mixed_raw_layer.view(*new_tensor_shape)
395
+
396
+ # [seq_len, batch, num_attention_heads, hidden_size_per_attention_head]
397
+ (query_layer, key_layer, value_layer) = self.split_tensor_along_last_dim(mixed_raw_layer, 3)
398
+
399
+ if self.position_encoding_2d:
400
+ q1, q2 = query_layer.chunk(2, dim=(query_layer.ndim - 1))
401
+ k1, k2 = key_layer.chunk(2, dim=(key_layer.ndim - 1))
402
+ cos, sin = self.rotary_emb(q1, seq_len=position_ids.max() + 1)
403
+ position_ids, block_position_ids = position_ids[:, 0, :].transpose(0, 1).contiguous(), \
404
+ position_ids[:, 1, :].transpose(0, 1).contiguous()
405
+ q1, k1 = apply_rotary_pos_emb_index(q1, k1, cos, sin, position_ids)
406
+ q2, k2 = apply_rotary_pos_emb_index(q2, k2, cos, sin, block_position_ids)
407
+ query_layer = torch.concat([q1, q2], dim=(q1.ndim - 1))
408
+ key_layer = torch.concat([k1, k2], dim=(k1.ndim - 1))
409
+ else:
410
+ position_ids = position_ids.transpose(0, 1)
411
+ cos, sin = self.rotary_emb(value_layer, seq_len=position_ids.max() + 1)
412
+ # [seq_len, batch, num_attention_heads, hidden_size_per_attention_head]
413
+ query_layer, key_layer = apply_rotary_pos_emb_index(query_layer, key_layer, cos, sin, position_ids)
414
+
415
+ # [seq_len, batch, hidden_size]
416
+ context_layer, present, attention_probs = attention_fn(
417
+ self=self,
418
+ query_layer=query_layer,
419
+ key_layer=key_layer,
420
+ value_layer=value_layer,
421
+ attention_mask=attention_mask,
422
+ hidden_size_per_partition=self.hidden_size_per_partition,
423
+ layer_id=layer_id,
424
+ layer_past=layer_past,
425
+ use_cache=use_cache
426
+ )
427
+
428
+ output = self.dense(context_layer)
429
+
430
+ outputs = (output, present)
431
+
432
+ if output_attentions:
433
+ outputs += (attention_probs,)
434
+
435
+ return outputs # output, present, attention_probs
436
+
437
+
438
+ class GEGLU(torch.nn.Module):
439
+ def __init__(self):
440
+ super().__init__()
441
+ self.activation_fn = F.gelu
442
+
443
+ def forward(self, x):
444
+ # dim=-1 breaks in jit for pt<1.10
445
+ x1, x2 = x.chunk(2, dim=(x.ndim - 1))
446
+ return x1 * self.activation_fn(x2)
447
+
448
+
449
+ class GLU(torch.nn.Module):
450
+ def __init__(self, hidden_size, inner_hidden_size=None,
451
+ layer_id=None, bias=True, activation_func=gelu, params_dtype=torch.float):
452
+ super(GLU, self).__init__()
453
+ self.layer_id = layer_id
454
+ self.activation_func = activation_func
455
+
456
+ # Project to 4h.
457
+ self.hidden_size = hidden_size
458
+ if inner_hidden_size is None:
459
+ inner_hidden_size = 4 * hidden_size
460
+ self.inner_hidden_size = inner_hidden_size
461
+ self.dense_h_to_4h = skip_init(
462
+ torch.nn.Linear,
463
+ self.hidden_size,
464
+ self.inner_hidden_size,
465
+ bias=bias,
466
+ dtype=params_dtype,
467
+ )
468
+ # Project back to h.
469
+ self.dense_4h_to_h = skip_init(
470
+ torch.nn.Linear,
471
+ self.inner_hidden_size,
472
+ self.hidden_size,
473
+ bias=bias,
474
+ dtype=params_dtype,
475
+ )
476
+
477
+ def forward(self, hidden_states):
478
+ """
479
+ hidden_states: [seq_len, batch, hidden_size]
480
+ """
481
+
482
+ # [seq_len, batch, inner_hidden_size]
483
+ intermediate_parallel = self.dense_h_to_4h(hidden_states)
484
+
485
+ intermediate_parallel = self.activation_func(intermediate_parallel)
486
+
487
+ output = self.dense_4h_to_h(intermediate_parallel)
488
+
489
+ return output
490
+
491
+
492
+ class GLMBlock(torch.nn.Module):
493
+ def __init__(
494
+ self,
495
+ hidden_size,
496
+ num_attention_heads,
497
+ layernorm_epsilon,
498
+ layer_id,
499
+ inner_hidden_size=None,
500
+ hidden_size_per_attention_head=None,
501
+ layernorm=LayerNorm,
502
+ use_bias=True,
503
+ params_dtype=torch.float,
504
+ num_layers=28,
505
+ position_encoding_2d=True
506
+ ):
507
+ super(GLMBlock, self).__init__()
508
+ # Set output layer initialization if not provided.
509
+
510
+ self.layer_id = layer_id
511
+
512
+ # Layernorm on the input data.
513
+ self.input_layernorm = layernorm(hidden_size, eps=layernorm_epsilon)
514
+
515
+ self.position_encoding_2d = position_encoding_2d
516
+
517
+ # Self attention.
518
+ self.attention = SelfAttention(
519
+ hidden_size,
520
+ num_attention_heads,
521
+ layer_id,
522
+ hidden_size_per_attention_head=hidden_size_per_attention_head,
523
+ bias=use_bias,
524
+ params_dtype=params_dtype,
525
+ position_encoding_2d=self.position_encoding_2d
526
+ )
527
+
528
+ # Layernorm on the input data.
529
+ self.post_attention_layernorm = layernorm(hidden_size, eps=layernorm_epsilon)
530
+
531
+ self.num_layers = num_layers
532
+
533
+ # GLU
534
+ self.mlp = GLU(
535
+ hidden_size,
536
+ inner_hidden_size=inner_hidden_size,
537
+ bias=use_bias,
538
+ layer_id=layer_id,
539
+ params_dtype=params_dtype,
540
+ )
541
+
542
+ def forward(
543
+ self,
544
+ hidden_states: torch.Tensor,
545
+ position_ids,
546
+ attention_mask: torch.Tensor,
547
+ layer_id,
548
+ layer_past: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
549
+ use_cache: bool = False,
550
+ output_attentions: bool = False,
551
+ ):
552
+ """
553
+ hidden_states: [seq_len, batch, hidden_size]
554
+ attention_mask: [(1, 1), seq_len, seq_len]
555
+ """
556
+
557
+ # Layer norm at the begining of the transformer layer.
558
+ # [seq_len, batch, hidden_size]
559
+ attention_input = self.input_layernorm(hidden_states)
560
+
561
+ # Self attention.
562
+ attention_outputs = self.attention(
563
+ attention_input,
564
+ position_ids,
565
+ attention_mask=attention_mask,
566
+ layer_id=layer_id,
567
+ layer_past=layer_past,
568
+ use_cache=use_cache,
569
+ output_attentions=output_attentions
570
+ )
571
+
572
+ attention_output = attention_outputs[0]
573
+
574
+ outputs = attention_outputs[1:]
575
+
576
+ # Residual connection.
577
+ alpha = (2 * self.num_layers) ** 0.5
578
+ hidden_states = attention_input * alpha + attention_output
579
+
580
+ mlp_input = self.post_attention_layernorm(hidden_states)
581
+
582
+ # MLP.
583
+ mlp_output = self.mlp(mlp_input)
584
+
585
+ # Second residual connection.
586
+ output = mlp_input * alpha + mlp_output
587
+
588
+ if use_cache:
589
+ outputs = (output,) + outputs
590
+ else:
591
+ outputs = (output,) + outputs[1:]
592
+
593
+ return outputs # hidden_states, present, attentions
594
+
595
+
596
+ class ChatGLMPreTrainedModel(PreTrainedModel):
597
+ """
598
+ An abstract class to handle weights initialization and
599
+ a simple interface for downloading and loading pretrained models.
600
+ """
601
+
602
+ is_parallelizable = True
603
+ supports_gradient_checkpointing = False
604
+ config_class = ChatGLMConfig
605
+ base_model_prefix = "transformer"
606
+ _no_split_modules = ["GLM6BBlock"]
607
+
608
+ def __init__(self, *inputs, **kwargs):
609
+ super().__init__(*inputs, **kwargs)
610
+
611
+ def _init_weights(self, module: nn.Module):
612
+ """Initialize the weights."""
613
+ return
614
+
615
+
616
+ CHATGLM_6B_START_DOCSTRING = r"""
617
+ This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class.
618
+ Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general
619
+ usage and behavior.
620
+
621
+ Parameters:
622
+ config ([`~ChatGLM6BConfig`]): Model configuration class with all the parameters of the model.
623
+ Initializing with a config file does not load the weights associated with the model, only the configuration.
624
+ Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
625
+ """
626
+
627
+ CHATGLM_6B_INPUTS_DOCSTRING = r"""
628
+ Args:
629
+ input_ids (`torch.LongTensor` of shape `({0})`):
630
+ Indices of input sequence tokens in the vocabulary.
631
+
632
+ Indices can be obtained using [`ChatGLM6BTokenizer`].
633
+ See [`PreTrainedTokenizer.encode`] and
634
+ [`PreTrainedTokenizer.__call__`] for details.
635
+
636
+ [What are input IDs?](../glossary#input-ids)
637
+ attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
638
+ Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
639
+
640
+ - 1 for tokens that are **not masked**,
641
+ - 0 for tokens that are **masked**.
642
+
643
+ [What are attention masks?](../glossary#attention-mask)
644
+ token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
645
+ Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`:
646
+
647
+ - 0 corresponds to a *sentence A* token,
648
+ - 1 corresponds to a *sentence B* token.
649
+
650
+ [What are token type IDs?](../glossary#token-type-ids)
651
+ position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
652
+ Indices of positions of each input sequence tokens in the position embeddings.
653
+ Selected in the range `[0, config.max_position_embeddings - 1]`.
654
+
655
+ [What are position IDs?](../glossary#position-ids)
656
+ head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
657
+ Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
658
+
659
+ - 1 indicates the head is **not masked**,
660
+ - 0 indicates the head is **masked**.
661
+
662
+ inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*):
663
+ Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
664
+ This is useful if you want more control over how to convert *input_ids* indices into associated vectors
665
+ than the model's internal embedding lookup matrix.
666
+ output_attentions (`bool`, *optional*):
667
+ Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
668
+ tensors for more detail.
669
+ output_hidden_states (`bool`, *optional*):
670
+ Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
671
+ more detail.
672
+ return_dict (`bool`, *optional*):
673
+ Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
674
+ """
675
+
676
+
677
+ @add_start_docstrings(
678
+ "The bare ChatGLM-6B Model transformer outputting raw hidden-states without any specific head on top.",
679
+ CHATGLM_6B_START_DOCSTRING,
680
+ )
681
+ class ChatGLMModel(ChatGLMPreTrainedModel):
682
+ """
683
+
684
+ The model can behave as an encoder (with only self-attention) as well
685
+ as a decoder, in which case a layer of cross-attention is added between
686
+ the self-attention layers, following the architecture described in [Attention is
687
+ all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani,
688
+ Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin.
689
+
690
+ To behave as an decoder the model needs to be initialized with the
691
+ `is_decoder` argument of the configuration set to `True`.
692
+ To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder`
693
+ argument and `add_cross_attention` set to `True`; an
694
+ `encoder_hidden_states` is then expected as an input to the forward pass.
695
+ """
696
+
697
+ def __init__(self, config: ChatGLMConfig):
698
+ super().__init__(config)
699
+
700
+ # recording parameters
701
+ self.max_sequence_length = config.max_sequence_length
702
+ self.hidden_size = config.hidden_size
703
+ self.params_dtype = torch.half
704
+ self.num_attention_heads = config.num_attention_heads
705
+ self.vocab_size = config.vocab_size
706
+ self.num_layers = config.num_layers
707
+ self.layernorm_epsilon = config.layernorm_epsilon
708
+ self.inner_hidden_size = config.inner_hidden_size
709
+ self.hidden_size_per_attention_head = self.hidden_size // self.num_attention_heads
710
+ self.position_encoding_2d = config.position_encoding_2d
711
+
712
+ self.word_embeddings = skip_init(
713
+ torch.nn.Embedding,
714
+ num_embeddings=self.vocab_size, embedding_dim=self.hidden_size,
715
+ dtype=self.params_dtype
716
+ )
717
+
718
+ def get_layer(layer_id):
719
+ return GLMBlock(
720
+ self.hidden_size,
721
+ self.num_attention_heads,
722
+ self.layernorm_epsilon,
723
+ layer_id,
724
+ inner_hidden_size=self.inner_hidden_size,
725
+ hidden_size_per_attention_head=self.hidden_size_per_attention_head,
726
+ layernorm=LayerNorm,
727
+ use_bias=True,
728
+ params_dtype=self.params_dtype,
729
+ position_encoding_2d=self.position_encoding_2d,
730
+ )
731
+
732
+ self.layers = torch.nn.ModuleList(
733
+ [get_layer(layer_id) for layer_id in range(self.num_layers)]
734
+ )
735
+
736
+ # Final layer norm before output.
737
+ self.final_layernorm = LayerNorm(self.hidden_size, eps=self.layernorm_epsilon)
738
+
739
+ def get_input_embeddings(self):
740
+ return self.word_embeddings
741
+
742
+ def set_input_embeddings(self, new_embeddings: torch.Tensor):
743
+ self.word_embeddings = new_embeddings
744
+
745
+ @staticmethod
746
+ def get_masks(seq, device):
747
+ context_length = seq.index(150004) + 1
748
+
749
+ attention_mask = torch.ones((1, len(seq), len(seq)), device=device)
750
+ attention_mask.tril_()
751
+ attention_mask[..., :context_length - 1] = 1
752
+ attention_mask.unsqueeze_(1)
753
+ attention_mask = (attention_mask < 0.5).bool()
754
+
755
+ return attention_mask
756
+
757
+ def get_position_ids(self, seq, mask_position, device, gmask=False):
758
+ context_length = seq.index(150004) + 1
759
+ if self.position_encoding_2d:
760
+ seq_length = seq.index(150004)
761
+ position_ids = torch.arange(context_length, dtype=torch.long, device=device)
762
+ if not gmask:
763
+ position_ids[seq_length:] = mask_position
764
+ block_position_ids = torch.cat((
765
+ torch.zeros(seq_length, dtype=torch.long, device=device),
766
+ torch.arange(context_length - seq_length, dtype=torch.long, device=device) + 1
767
+ ))
768
+ position_ids = torch.stack((position_ids, block_position_ids), dim=0)
769
+ else:
770
+ position_ids = torch.arange(context_length, dtype=torch.long, device=device)
771
+ if not gmask:
772
+ position_ids[context_length - 1:] = mask_position
773
+
774
+ position_ids = position_ids.unsqueeze(0)
775
+
776
+ return position_ids
777
+
778
+ @add_start_docstrings_to_model_forward(CHATGLM_6B_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
779
+ @add_code_sample_docstrings(
780
+ checkpoint=_CHECKPOINT_FOR_DOC,
781
+ output_type=BaseModelOutputWithPastAndCrossAttentions,
782
+ config_class=_CONFIG_FOR_DOC,
783
+ )
784
+ def forward(
785
+ self,
786
+ input_ids: Optional[torch.LongTensor] = None,
787
+ position_ids: Optional[torch.LongTensor] = None,
788
+ attention_mask: Optional[torch.Tensor] = None,
789
+ past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
790
+ inputs_embeds: Optional[torch.LongTensor] = None,
791
+ use_cache: Optional[bool] = None,
792
+ output_attentions: Optional[bool] = None,
793
+ output_hidden_states: Optional[bool] = None,
794
+ return_dict: Optional[bool] = None,
795
+ ) -> Union[Tuple[torch.Tensor, ...], BaseModelOutputWithPast]:
796
+
797
+ output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
798
+ output_hidden_states = (
799
+ output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
800
+ )
801
+ use_cache = use_cache if use_cache is not None else self.config.use_cache
802
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
803
+
804
+ if input_ids is not None and inputs_embeds is not None:
805
+ raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
806
+ elif input_ids is not None:
807
+ batch_size, seq_length = input_ids.shape[:2]
808
+ elif inputs_embeds is not None:
809
+ batch_size, seq_length, _ = inputs_embeds.shape[:2]
810
+ else:
811
+ raise ValueError("You have to specify either input_ids or inputs_embeds")
812
+
813
+ if past_key_values is None:
814
+ past_key_values = tuple([None] * len(self.layers))
815
+
816
+ MASK, gMASK = 150000, 150001
817
+ mask_token = MASK if MASK in input_ids else gMASK
818
+ use_gmask = False if MASK in input_ids else gMASK
819
+ seq = input_ids[0].tolist()
820
+
821
+ mask_position = seq.index(mask_token)
822
+
823
+ if attention_mask is None:
824
+ attention_mask = self.get_masks(
825
+ seq=seq,
826
+ device=input_ids.device
827
+ )
828
+
829
+ if position_ids is None:
830
+ position_ids = self.get_position_ids(
831
+ seq=seq,
832
+ mask_position=mask_position,
833
+ device=input_ids.device,
834
+ gmask=use_gmask
835
+ )
836
+
837
+ if inputs_embeds is None:
838
+ inputs_embeds = self.word_embeddings(input_ids)
839
+
840
+ # [seq_len, batch, hidden_size]
841
+ hidden_states = inputs_embeds.transpose(0, 1)
842
+
843
+ presents = () if use_cache else None
844
+ all_self_attentions = () if output_attentions else None
845
+ all_hidden_states = () if output_hidden_states else None
846
+
847
+ seq_length_with_past = seq_length
848
+ past_key_values_length = 0
849
+ if past_key_values[0] is not None:
850
+ past_key_values_length = past_key_values[0][0].shape[0]
851
+ seq_length_with_past = seq_length_with_past + past_key_values_length
852
+ if attention_mask is None:
853
+ attention_mask = torch.zeros(1, 1, device=input_ids.device).bool()
854
+
855
+ else:
856
+ attention_mask = attention_mask.to(input_ids.device)
857
+
858
+ for i, layer in enumerate(self.layers):
859
+
860
+ if output_hidden_states:
861
+ all_hidden_states = all_hidden_states + (hidden_states,)
862
+
863
+ layer_ret = layer(
864
+ hidden_states,
865
+ position_ids=position_ids,
866
+ attention_mask=attention_mask,
867
+ layer_id=torch.tensor(i),
868
+ layer_past=past_key_values[i],
869
+ use_cache=use_cache,
870
+ output_attentions=output_attentions
871
+ )
872
+
873
+ hidden_states = layer_ret[0]
874
+
875
+ if use_cache:
876
+ presents = presents + (layer_ret[1],)
877
+
878
+ if output_attentions:
879
+ all_self_attentions = all_self_attentions + (layer_ret[2 if use_cache else 1],)
880
+
881
+ # Final layer norm.
882
+ hidden_states = self.final_layernorm(hidden_states)
883
+
884
+ if output_hidden_states:
885
+ all_hidden_states = all_hidden_states + (hidden_states,)
886
+
887
+ if not return_dict:
888
+ return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
889
+
890
+ return BaseModelOutputWithPast(
891
+ last_hidden_state=hidden_states,
892
+ past_key_values=presents,
893
+ hidden_states=all_hidden_states,
894
+ attentions=all_self_attentions,
895
+ )
896
+
897
+
898
+ class ChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):
899
+ def __init__(self, config):
900
+ super().__init__(config)
901
+
902
+ # self.hidden_size = config.hidden_size
903
+ # self.params_dtype = torch.half
904
+ # self.vocab_size = config.vocab_size
905
+ self.max_sequence_length = config.max_sequence_length
906
+
907
+ self.position_encoding_2d = config.position_encoding_2d
908
+
909
+ self.transformer = ChatGLMModel(config)
910
+
911
+ self.lm_head = skip_init(
912
+ nn.Linear,
913
+ config.hidden_size,
914
+ config.vocab_size,
915
+ bias=False,
916
+ dtype=torch.half
917
+ )
918
+
919
+ def get_output_embeddings(self):
920
+ return self.lm_head
921
+
922
+ def set_output_embeddings(self, new_embeddings):
923
+ self.lm_head = new_embeddings
924
+
925
+ def get_masks_and_position_ids(self, seq, mask_position, context_length, device, gmask=False):
926
+ attention_mask = torch.ones((1, context_length, context_length), device=device)
927
+ attention_mask.tril_()
928
+ attention_mask[..., :context_length - 1] = 1
929
+ attention_mask.unsqueeze_(1)
930
+ attention_mask = (attention_mask < 0.5).bool()
931
+
932
+ if self.position_encoding_2d:
933
+ seq_length = seq.index(150004)
934
+ position_ids = torch.arange(context_length, dtype=torch.long, device=device)
935
+ if not gmask:
936
+ position_ids[seq_length:] = mask_position
937
+ block_position_ids = torch.cat((
938
+ torch.zeros(seq_length, dtype=torch.long, device=device),
939
+ torch.arange(context_length - seq_length, dtype=torch.long, device=device) + 1
940
+ ))
941
+ position_ids = torch.stack((position_ids, block_position_ids), dim=0)
942
+ else:
943
+ position_ids = torch.arange(context_length, dtype=torch.long, device=device)
944
+ if not gmask:
945
+ position_ids[context_length - 1:] = mask_position
946
+
947
+ position_ids = position_ids.unsqueeze(0)
948
+
949
+ return attention_mask, position_ids
950
+
951
+ def prepare_inputs_for_generation(
952
+ self,
953
+ input_ids: torch.LongTensor,
954
+ past: Optional[torch.Tensor] = None,
955
+ attention_mask: Optional[torch.Tensor] = None,
956
+ **kwargs
957
+ ) -> dict:
958
+
959
+ MASK, gMASK = 150000, 150001
960
+ mask_token = MASK if MASK in input_ids else gMASK
961
+ use_gmask = False if MASK in input_ids else gMASK
962
+ seq = input_ids[0].tolist()
963
+ mask_position = seq.index(mask_token)
964
+
965
+ if mask_token not in seq:
966
+ raise ValueError("You have to add either [MASK] or [gMASK] in your input")
967
+
968
+ # only last token for input_ids if past is not None
969
+ if past:
970
+ context_length = seq.index(150004)
971
+ last_token = input_ids[:, -1].unsqueeze(-1)
972
+ if self.position_encoding_2d:
973
+ position_ids = torch.tensor([[[mask_position], [len(seq) - context_length]]], dtype=torch.long,
974
+ device=input_ids.device)
975
+ else:
976
+ position_ids = torch.tensor([[mask_position]], dtype=torch.long, device=input_ids.device)
977
+
978
+ return {
979
+ "input_ids": last_token,
980
+ "past_key_values": past,
981
+ "position_ids": position_ids,
982
+ }
983
+ else:
984
+ attention_mask, position_ids = self.get_masks_and_position_ids(
985
+ seq=seq,
986
+ mask_position=mask_position,
987
+ context_length=len(seq),
988
+ device=input_ids.device,
989
+ gmask=use_gmask
990
+ )
991
+
992
+ return {
993
+ "input_ids": input_ids,
994
+ "past_key_values": past,
995
+ "position_ids": position_ids,
996
+ "attention_mask": attention_mask
997
+ }
998
+
999
+ def forward(
1000
+ self,
1001
+ input_ids: Optional[torch.Tensor] = None,
1002
+ position_ids: Optional[torch.Tensor] = None,
1003
+ attention_mask: Optional[torch.Tensor] = None,
1004
+ past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
1005
+ inputs_embeds: Optional[torch.Tensor] = None,
1006
+ labels: Optional[torch.Tensor] = None,
1007
+ use_cache: Optional[bool] = None,
1008
+ output_attentions: Optional[bool] = None,
1009
+ output_hidden_states: Optional[bool] = None,
1010
+ return_dict: Optional[bool] = None,
1011
+ ):
1012
+ use_cache = use_cache if use_cache is not None else self.config.use_cache
1013
+ return_dict = return_dict if return_dict is not None else self.config.use_return_dict
1014
+
1015
+ transformer_outputs = self.transformer(
1016
+ input_ids=input_ids,
1017
+ position_ids=position_ids,
1018
+ attention_mask=attention_mask,
1019
+ past_key_values=past_key_values,
1020
+ inputs_embeds=inputs_embeds,
1021
+ use_cache=use_cache,
1022
+ output_attentions=output_attentions,
1023
+ output_hidden_states=output_hidden_states,
1024
+ return_dict=return_dict,
1025
+ )
1026
+
1027
+ hidden_states = transformer_outputs[0]
1028
+
1029
+ lm_logits = self.lm_head(hidden_states).permute(1, 0, 2).contiguous()
1030
+
1031
+ loss = None
1032
+ if labels is not None:
1033
+ lm_logits = lm_logits.to(torch.float32)
1034
+
1035
+ # Shift so that tokens < n predict n
1036
+ shift_logits = lm_logits[..., :-1, :].contiguous()
1037
+ shift_labels = labels[..., 1:].contiguous()
1038
+ # Flatten the tokens
1039
+ loss_fct = CrossEntropyLoss()
1040
+ loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
1041
+
1042
+ lm_logits = lm_logits.to(hidden_states.dtype)
1043
+ loss = loss.to(hidden_states.dtype)
1044
+
1045
+ if not return_dict:
1046
+ output = (lm_logits,) + transformer_outputs[1:]
1047
+ return ((loss,) + output) if loss is not None else output
1048
+
1049
+ return CausalLMOutputWithPast(
1050
+ loss=loss,
1051
+ logits=lm_logits,
1052
+ past_key_values=transformer_outputs.past_key_values,
1053
+ hidden_states=transformer_outputs.hidden_states,
1054
+ attentions=transformer_outputs.attentions,
1055
+ )
1056
+
1057
+ @staticmethod
1058
+ def _reorder_cache(
1059
+ past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
1060
+ ) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
1061
+ """
1062
+ This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
1063
+ [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
1064
+ beam_idx at every generation step.
1065
+
1066
+ Output shares the same memory storage as `past`.
1067
+ """
1068
+ return tuple(
1069
+ (
1070
+ layer_past[0].index_select(1, beam_idx.to(layer_past[0].device)),
1071
+ layer_past[1].index_select(1, beam_idx.to(layer_past[1].device)),
1072
+ )
1073
+ for layer_past in past
1074
+ )
1075
+
1076
+ @torch.no_grad()
1077
+ def chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = [], max_length: int = 2048, num_beams=1,
1078
+ do_sample=True, top_p=0.7, temperature=0.95, **kwargs):
1079
+ gen_kwargs = {"max_length": max_length, "num_beams": num_beams, "do_sample": do_sample, "top_p": top_p,
1080
+ "temperature": temperature, **kwargs}
1081
+ if not history:
1082
+ prompt = query
1083
+ else:
1084
+ prompt = ""
1085
+ for i, (old_query, response) in enumerate(history):
1086
+ prompt += "[Round {}]\n问:{}\n答:{}\n".format(i, old_query, response)
1087
+ prompt += "[Round {}]\n问:{}\n答:".format(len(history), query)
1088
+ input_ids = tokenizer([prompt], return_tensors="pt", padding=True)
1089
+ input_ids = input_ids.to(self.device)
1090
+ outputs = self.generate(**input_ids, **gen_kwargs)
1091
+ outputs = outputs.tolist()[0][len(input_ids["input_ids"][0]) - 2:]
1092
+ response = tokenizer.decode(outputs)
1093
+ response = response.strip()
1094
+ response = response.replace("[[训练时间]]", "2023年")
1095
+ history.append((query, response))
1096
+ return response, history
1097
+
1098
+ @torch.no_grad()
1099
+ def generate(
1100
+ self,
1101
+ **kwargs,
1102
+ ):
1103
+ MASK, gMASK = 150000, 150001
1104
+ bos, eos = 150004, 150005
1105
+
1106
+ if "eos_token_id" not in kwargs:
1107
+ kwargs["eos_token_id"] = eos
1108
+
1109
+ stop = False
1110
+
1111
+ return_seqs = []
1112
+
1113
+ while True:
1114
+ output_ids = super().generate(**kwargs)
1115
+
1116
+ return_seqs = []
1117
+ max_length = 0
1118
+
1119
+ for i in range(output_ids.shape[0]):
1120
+ output_seq = output_ids[i].tolist()
1121
+ mask_token = MASK if MASK in output_seq else gMASK
1122
+ mask_position = output_seq.index(mask_token)
1123
+ bos_position = output_seq.index(bos)
1124
+ if eos in output_seq:
1125
+ eos_position = output_seq.index(eos)
1126
+ else:
1127
+ eos_position = len(output_seq)
1128
+
1129
+ return_seq = output_seq[:mask_position] + output_seq[bos_position + 1:eos_position] + output_seq[
1130
+ mask_position + 1:bos_position]
1131
+ max_length = max(max_length, len(return_seq))
1132
+ return_seqs.append(return_seq)
1133
+
1134
+ for i in range(output_ids.shape[0]):
1135
+ return_seqs[i] = [0] * (max_length - len(return_seqs[i])) + return_seqs[i] # padding
1136
+ if mask_token not in return_seqs[i]:
1137
+ stop = True
1138
+
1139
+ if stop:
1140
+ break
1141
+
1142
+ for return_seq in return_seqs:
1143
+ return_seq += [bos]
1144
+
1145
+ kwargs['input_ids'] = torch.tensor(return_seqs, dtype=torch.long, device=kwargs['input_ids'].device)
1146
+
1147
+ return torch.tensor(return_seqs, dtype=torch.long, device=kwargs['input_ids'].device)
1148
+
1149
+ def quantize(self, bits: int):
1150
+ from .quantization import quantize
1151
+ self.transformer = quantize(self.transformer, bits)
1152
+ return self
pytorch_model-00001-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00001-of-00008.bin
pytorch_model-00002-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00002-of-00008.bin
pytorch_model-00003-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00003-of-00008.bin
pytorch_model-00004-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00004-of-00008.bin
pytorch_model-00005-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00005-of-00008.bin
pytorch_model-00006-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00006-of-00008.bin
pytorch_model-00007-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00007-of-00008.bin
pytorch_model-00008-of-00008.bin ADDED
@@ -0,0 +1 @@
 
1
+ /mnt/vepfs/zxdu/checkpoints/qa-glm-6b-sft-v0.8-v2-original-lr/pytorch_model-00008-of-00008.bin
pytorch_model.bin.index.json ADDED
@@ -0,0 +1,375 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 13744473856
4
+ },
5
+ "weight_map": {
6
+ "lm_head.weight": "pytorch_model-00008-of-00008.bin",
7
+ "transformer.final_layernorm.bias": "pytorch_model-00007-of-00008.bin",
8
+ "transformer.final_layernorm.weight": "pytorch_model-00007-of-00008.bin",
9
+ "transformer.layers.0.attention.dense.bias": "pytorch_model-00001-of-00008.bin",
10
+ "transformer.layers.0.attention.dense.weight": "pytorch_model-00001-of-00008.bin",
11
+ "transformer.layers.0.attention.query_key_value.bias": "pytorch_model-00001-of-00008.bin",
12
+ "transformer.layers.0.attention.query_key_value.weight": "pytorch_model-00001-of-00008.bin",
13
+ "transformer.layers.0.attention.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
14
+ "transformer.layers.0.input_layernorm.bias": "pytorch_model-00001-of-00008.bin",
15
+ "transformer.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
16
+ "transformer.layers.0.mlp.dense_4h_to_h.bias": "pytorch_model-00001-of-00008.bin",
17
+ "transformer.layers.0.mlp.dense_4h_to_h.weight": "pytorch_model-00001-of-00008.bin",
18
+ "transformer.layers.0.mlp.dense_h_to_4h.bias": "pytorch_model-00001-of-00008.bin",
19
+ "transformer.layers.0.mlp.dense_h_to_4h.weight": "pytorch_model-00001-of-00008.bin",
20
+ "transformer.layers.0.post_attention_layernorm.bias": "pytorch_model-00001-of-00008.bin",
21
+ "transformer.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
22
+ "transformer.layers.1.attention.dense.bias": "pytorch_model-00001-of-00008.bin",
23
+ "transformer.layers.1.attention.dense.weight": "pytorch_model-00001-of-00008.bin",
24
+ "transformer.layers.1.attention.query_key_value.bias": "pytorch_model-00001-of-00008.bin",
25
+ "transformer.layers.1.attention.query_key_value.weight": "pytorch_model-00001-of-00008.bin",
26
+ "transformer.layers.1.attention.rotary_emb.inv_freq": "pytorch_model-00001-of-00008.bin",
27
+ "transformer.layers.1.input_layernorm.bias": "pytorch_model-00001-of-00008.bin",
28
+ "transformer.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00008.bin",
29
+ "transformer.layers.1.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
30
+ "transformer.layers.1.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
31
+ "transformer.layers.1.mlp.dense_h_to_4h.bias": "pytorch_model-00001-of-00008.bin",
32
+ "transformer.layers.1.mlp.dense_h_to_4h.weight": "pytorch_model-00001-of-00008.bin",
33
+ "transformer.layers.1.post_attention_layernorm.bias": "pytorch_model-00001-of-00008.bin",
34
+ "transformer.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00008.bin",
35
+ "transformer.layers.10.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
36
+ "transformer.layers.10.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
37
+ "transformer.layers.10.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
38
+ "transformer.layers.10.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
39
+ "transformer.layers.10.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
40
+ "transformer.layers.10.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
41
+ "transformer.layers.10.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
42
+ "transformer.layers.10.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
43
+ "transformer.layers.10.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
44
+ "transformer.layers.10.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
45
+ "transformer.layers.10.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
46
+ "transformer.layers.10.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
47
+ "transformer.layers.10.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
48
+ "transformer.layers.11.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
49
+ "transformer.layers.11.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
50
+ "transformer.layers.11.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
51
+ "transformer.layers.11.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
52
+ "transformer.layers.11.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
53
+ "transformer.layers.11.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
54
+ "transformer.layers.11.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
55
+ "transformer.layers.11.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
56
+ "transformer.layers.11.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
57
+ "transformer.layers.11.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
58
+ "transformer.layers.11.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
59
+ "transformer.layers.11.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
60
+ "transformer.layers.11.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
61
+ "transformer.layers.12.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
62
+ "transformer.layers.12.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
63
+ "transformer.layers.12.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
64
+ "transformer.layers.12.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
65
+ "transformer.layers.12.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
66
+ "transformer.layers.12.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
67
+ "transformer.layers.12.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
68
+ "transformer.layers.12.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
69
+ "transformer.layers.12.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
70
+ "transformer.layers.12.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
71
+ "transformer.layers.12.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
72
+ "transformer.layers.12.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
73
+ "transformer.layers.12.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
74
+ "transformer.layers.13.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
75
+ "transformer.layers.13.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
76
+ "transformer.layers.13.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
77
+ "transformer.layers.13.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
78
+ "transformer.layers.13.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
79
+ "transformer.layers.13.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
80
+ "transformer.layers.13.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
81
+ "transformer.layers.13.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
82
+ "transformer.layers.13.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
83
+ "transformer.layers.13.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
84
+ "transformer.layers.13.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
85
+ "transformer.layers.13.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
86
+ "transformer.layers.13.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
87
+ "transformer.layers.14.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
88
+ "transformer.layers.14.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
89
+ "transformer.layers.14.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
90
+ "transformer.layers.14.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
91
+ "transformer.layers.14.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
92
+ "transformer.layers.14.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
93
+ "transformer.layers.14.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
94
+ "transformer.layers.14.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
95
+ "transformer.layers.14.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
96
+ "transformer.layers.14.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
97
+ "transformer.layers.14.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
98
+ "transformer.layers.14.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
99
+ "transformer.layers.14.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
100
+ "transformer.layers.15.attention.dense.bias": "pytorch_model-00004-of-00008.bin",
101
+ "transformer.layers.15.attention.dense.weight": "pytorch_model-00004-of-00008.bin",
102
+ "transformer.layers.15.attention.query_key_value.bias": "pytorch_model-00004-of-00008.bin",
103
+ "transformer.layers.15.attention.query_key_value.weight": "pytorch_model-00004-of-00008.bin",
104
+ "transformer.layers.15.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
105
+ "transformer.layers.15.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
106
+ "transformer.layers.15.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
107
+ "transformer.layers.15.mlp.dense_4h_to_h.bias": "pytorch_model-00004-of-00008.bin",
108
+ "transformer.layers.15.mlp.dense_4h_to_h.weight": "pytorch_model-00004-of-00008.bin",
109
+ "transformer.layers.15.mlp.dense_h_to_4h.bias": "pytorch_model-00004-of-00008.bin",
110
+ "transformer.layers.15.mlp.dense_h_to_4h.weight": "pytorch_model-00004-of-00008.bin",
111
+ "transformer.layers.15.post_attention_layernorm.bias": "pytorch_model-00004-of-00008.bin",
112
+ "transformer.layers.15.post_attention_layernorm.weight": "pytorch_model-00004-of-00008.bin",
113
+ "transformer.layers.16.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
114
+ "transformer.layers.16.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
115
+ "transformer.layers.16.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
116
+ "transformer.layers.16.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
117
+ "transformer.layers.16.attention.rotary_emb.inv_freq": "pytorch_model-00004-of-00008.bin",
118
+ "transformer.layers.16.input_layernorm.bias": "pytorch_model-00004-of-00008.bin",
119
+ "transformer.layers.16.input_layernorm.weight": "pytorch_model-00004-of-00008.bin",
120
+ "transformer.layers.16.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
121
+ "transformer.layers.16.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
122
+ "transformer.layers.16.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
123
+ "transformer.layers.16.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
124
+ "transformer.layers.16.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
125
+ "transformer.layers.16.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
126
+ "transformer.layers.17.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
127
+ "transformer.layers.17.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
128
+ "transformer.layers.17.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
129
+ "transformer.layers.17.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
130
+ "transformer.layers.17.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
131
+ "transformer.layers.17.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
132
+ "transformer.layers.17.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
133
+ "transformer.layers.17.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
134
+ "transformer.layers.17.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
135
+ "transformer.layers.17.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
136
+ "transformer.layers.17.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
137
+ "transformer.layers.17.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
138
+ "transformer.layers.17.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
139
+ "transformer.layers.18.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
140
+ "transformer.layers.18.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
141
+ "transformer.layers.18.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
142
+ "transformer.layers.18.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
143
+ "transformer.layers.18.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
144
+ "transformer.layers.18.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
145
+ "transformer.layers.18.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
146
+ "transformer.layers.18.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
147
+ "transformer.layers.18.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
148
+ "transformer.layers.18.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
149
+ "transformer.layers.18.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
150
+ "transformer.layers.18.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
151
+ "transformer.layers.18.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
152
+ "transformer.layers.19.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
153
+ "transformer.layers.19.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
154
+ "transformer.layers.19.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
155
+ "transformer.layers.19.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
156
+ "transformer.layers.19.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
157
+ "transformer.layers.19.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
158
+ "transformer.layers.19.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
159
+ "transformer.layers.19.mlp.dense_4h_to_h.bias": "pytorch_model-00005-of-00008.bin",
160
+ "transformer.layers.19.mlp.dense_4h_to_h.weight": "pytorch_model-00005-of-00008.bin",
161
+ "transformer.layers.19.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
162
+ "transformer.layers.19.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
163
+ "transformer.layers.19.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
164
+ "transformer.layers.19.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
165
+ "transformer.layers.2.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
166
+ "transformer.layers.2.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
167
+ "transformer.layers.2.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
168
+ "transformer.layers.2.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
169
+ "transformer.layers.2.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
170
+ "transformer.layers.2.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
171
+ "transformer.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
172
+ "transformer.layers.2.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
173
+ "transformer.layers.2.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
174
+ "transformer.layers.2.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
175
+ "transformer.layers.2.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
176
+ "transformer.layers.2.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
177
+ "transformer.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
178
+ "transformer.layers.20.attention.dense.bias": "pytorch_model-00005-of-00008.bin",
179
+ "transformer.layers.20.attention.dense.weight": "pytorch_model-00005-of-00008.bin",
180
+ "transformer.layers.20.attention.query_key_value.bias": "pytorch_model-00005-of-00008.bin",
181
+ "transformer.layers.20.attention.query_key_value.weight": "pytorch_model-00005-of-00008.bin",
182
+ "transformer.layers.20.attention.rotary_emb.inv_freq": "pytorch_model-00005-of-00008.bin",
183
+ "transformer.layers.20.input_layernorm.bias": "pytorch_model-00005-of-00008.bin",
184
+ "transformer.layers.20.input_layernorm.weight": "pytorch_model-00005-of-00008.bin",
185
+ "transformer.layers.20.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
186
+ "transformer.layers.20.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
187
+ "transformer.layers.20.mlp.dense_h_to_4h.bias": "pytorch_model-00005-of-00008.bin",
188
+ "transformer.layers.20.mlp.dense_h_to_4h.weight": "pytorch_model-00005-of-00008.bin",
189
+ "transformer.layers.20.post_attention_layernorm.bias": "pytorch_model-00005-of-00008.bin",
190
+ "transformer.layers.20.post_attention_layernorm.weight": "pytorch_model-00005-of-00008.bin",
191
+ "transformer.layers.21.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
192
+ "transformer.layers.21.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
193
+ "transformer.layers.21.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
194
+ "transformer.layers.21.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
195
+ "transformer.layers.21.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
196
+ "transformer.layers.21.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
197
+ "transformer.layers.21.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
198
+ "transformer.layers.21.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
199
+ "transformer.layers.21.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
200
+ "transformer.layers.21.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
201
+ "transformer.layers.21.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
202
+ "transformer.layers.21.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
203
+ "transformer.layers.21.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
204
+ "transformer.layers.22.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
205
+ "transformer.layers.22.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
206
+ "transformer.layers.22.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
207
+ "transformer.layers.22.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
208
+ "transformer.layers.22.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
209
+ "transformer.layers.22.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
210
+ "transformer.layers.22.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
211
+ "transformer.layers.22.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
212
+ "transformer.layers.22.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
213
+ "transformer.layers.22.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
214
+ "transformer.layers.22.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
215
+ "transformer.layers.22.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
216
+ "transformer.layers.22.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
217
+ "transformer.layers.23.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
218
+ "transformer.layers.23.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
219
+ "transformer.layers.23.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
220
+ "transformer.layers.23.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
221
+ "transformer.layers.23.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
222
+ "transformer.layers.23.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
223
+ "transformer.layers.23.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
224
+ "transformer.layers.23.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
225
+ "transformer.layers.23.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
226
+ "transformer.layers.23.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
227
+ "transformer.layers.23.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
228
+ "transformer.layers.23.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
229
+ "transformer.layers.23.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
230
+ "transformer.layers.24.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
231
+ "transformer.layers.24.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
232
+ "transformer.layers.24.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
233
+ "transformer.layers.24.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
234
+ "transformer.layers.24.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
235
+ "transformer.layers.24.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
236
+ "transformer.layers.24.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
237
+ "transformer.layers.24.mlp.dense_4h_to_h.bias": "pytorch_model-00006-of-00008.bin",
238
+ "transformer.layers.24.mlp.dense_4h_to_h.weight": "pytorch_model-00006-of-00008.bin",
239
+ "transformer.layers.24.mlp.dense_h_to_4h.bias": "pytorch_model-00006-of-00008.bin",
240
+ "transformer.layers.24.mlp.dense_h_to_4h.weight": "pytorch_model-00006-of-00008.bin",
241
+ "transformer.layers.24.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
242
+ "transformer.layers.24.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
243
+ "transformer.layers.25.attention.dense.bias": "pytorch_model-00006-of-00008.bin",
244
+ "transformer.layers.25.attention.dense.weight": "pytorch_model-00006-of-00008.bin",
245
+ "transformer.layers.25.attention.query_key_value.bias": "pytorch_model-00006-of-00008.bin",
246
+ "transformer.layers.25.attention.query_key_value.weight": "pytorch_model-00006-of-00008.bin",
247
+ "transformer.layers.25.attention.rotary_emb.inv_freq": "pytorch_model-00006-of-00008.bin",
248
+ "transformer.layers.25.input_layernorm.bias": "pytorch_model-00006-of-00008.bin",
249
+ "transformer.layers.25.input_layernorm.weight": "pytorch_model-00006-of-00008.bin",
250
+ "transformer.layers.25.mlp.dense_4h_to_h.bias": "pytorch_model-00007-of-00008.bin",
251
+ "transformer.layers.25.mlp.dense_4h_to_h.weight": "pytorch_model-00007-of-00008.bin",
252
+ "transformer.layers.25.mlp.dense_h_to_4h.bias": "pytorch_model-00007-of-00008.bin",
253
+ "transformer.layers.25.mlp.dense_h_to_4h.weight": "pytorch_model-00007-of-00008.bin",
254
+ "transformer.layers.25.post_attention_layernorm.bias": "pytorch_model-00006-of-00008.bin",
255
+ "transformer.layers.25.post_attention_layernorm.weight": "pytorch_model-00006-of-00008.bin",
256
+ "transformer.layers.26.attention.dense.bias": "pytorch_model-00007-of-00008.bin",
257
+ "transformer.layers.26.attention.dense.weight": "pytorch_model-00007-of-00008.bin",
258
+ "transformer.layers.26.attention.query_key_value.bias": "pytorch_model-00007-of-00008.bin",
259
+ "transformer.layers.26.attention.query_key_value.weight": "pytorch_model-00007-of-00008.bin",
260
+ "transformer.layers.26.attention.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
261
+ "transformer.layers.26.input_layernorm.bias": "pytorch_model-00007-of-00008.bin",
262
+ "transformer.layers.26.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
263
+ "transformer.layers.26.mlp.dense_4h_to_h.bias": "pytorch_model-00007-of-00008.bin",
264
+ "transformer.layers.26.mlp.dense_4h_to_h.weight": "pytorch_model-00007-of-00008.bin",
265
+ "transformer.layers.26.mlp.dense_h_to_4h.bias": "pytorch_model-00007-of-00008.bin",
266
+ "transformer.layers.26.mlp.dense_h_to_4h.weight": "pytorch_model-00007-of-00008.bin",
267
+ "transformer.layers.26.post_attention_layernorm.bias": "pytorch_model-00007-of-00008.bin",
268
+ "transformer.layers.26.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
269
+ "transformer.layers.27.attention.dense.bias": "pytorch_model-00007-of-00008.bin",
270
+ "transformer.layers.27.attention.dense.weight": "pytorch_model-00007-of-00008.bin",
271
+ "transformer.layers.27.attention.query_key_value.bias": "pytorch_model-00007-of-00008.bin",
272
+ "transformer.layers.27.attention.query_key_value.weight": "pytorch_model-00007-of-00008.bin",
273
+ "transformer.layers.27.attention.rotary_emb.inv_freq": "pytorch_model-00007-of-00008.bin",
274
+ "transformer.layers.27.input_layernorm.bias": "pytorch_model-00007-of-00008.bin",
275
+ "transformer.layers.27.input_layernorm.weight": "pytorch_model-00007-of-00008.bin",
276
+ "transformer.layers.27.mlp.dense_4h_to_h.bias": "pytorch_model-00007-of-00008.bin",
277
+ "transformer.layers.27.mlp.dense_4h_to_h.weight": "pytorch_model-00007-of-00008.bin",
278
+ "transformer.layers.27.mlp.dense_h_to_4h.bias": "pytorch_model-00007-of-00008.bin",
279
+ "transformer.layers.27.mlp.dense_h_to_4h.weight": "pytorch_model-00007-of-00008.bin",
280
+ "transformer.layers.27.post_attention_layernorm.bias": "pytorch_model-00007-of-00008.bin",
281
+ "transformer.layers.27.post_attention_layernorm.weight": "pytorch_model-00007-of-00008.bin",
282
+ "transformer.layers.3.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
283
+ "transformer.layers.3.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
284
+ "transformer.layers.3.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
285
+ "transformer.layers.3.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
286
+ "transformer.layers.3.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
287
+ "transformer.layers.3.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
288
+ "transformer.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
289
+ "transformer.layers.3.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
290
+ "transformer.layers.3.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
291
+ "transformer.layers.3.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
292
+ "transformer.layers.3.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
293
+ "transformer.layers.3.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
294
+ "transformer.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
295
+ "transformer.layers.4.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
296
+ "transformer.layers.4.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
297
+ "transformer.layers.4.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
298
+ "transformer.layers.4.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
299
+ "transformer.layers.4.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
300
+ "transformer.layers.4.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
301
+ "transformer.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
302
+ "transformer.layers.4.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
303
+ "transformer.layers.4.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
304
+ "transformer.layers.4.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
305
+ "transformer.layers.4.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
306
+ "transformer.layers.4.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
307
+ "transformer.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
308
+ "transformer.layers.5.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
309
+ "transformer.layers.5.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
310
+ "transformer.layers.5.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
311
+ "transformer.layers.5.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
312
+ "transformer.layers.5.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
313
+ "transformer.layers.5.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
314
+ "transformer.layers.5.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
315
+ "transformer.layers.5.mlp.dense_4h_to_h.bias": "pytorch_model-00002-of-00008.bin",
316
+ "transformer.layers.5.mlp.dense_4h_to_h.weight": "pytorch_model-00002-of-00008.bin",
317
+ "transformer.layers.5.mlp.dense_h_to_4h.bias": "pytorch_model-00002-of-00008.bin",
318
+ "transformer.layers.5.mlp.dense_h_to_4h.weight": "pytorch_model-00002-of-00008.bin",
319
+ "transformer.layers.5.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
320
+ "transformer.layers.5.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
321
+ "transformer.layers.6.attention.dense.bias": "pytorch_model-00002-of-00008.bin",
322
+ "transformer.layers.6.attention.dense.weight": "pytorch_model-00002-of-00008.bin",
323
+ "transformer.layers.6.attention.query_key_value.bias": "pytorch_model-00002-of-00008.bin",
324
+ "transformer.layers.6.attention.query_key_value.weight": "pytorch_model-00002-of-00008.bin",
325
+ "transformer.layers.6.attention.rotary_emb.inv_freq": "pytorch_model-00002-of-00008.bin",
326
+ "transformer.layers.6.input_layernorm.bias": "pytorch_model-00002-of-00008.bin",
327
+ "transformer.layers.6.input_layernorm.weight": "pytorch_model-00002-of-00008.bin",
328
+ "transformer.layers.6.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
329
+ "transformer.layers.6.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
330
+ "transformer.layers.6.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
331
+ "transformer.layers.6.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
332
+ "transformer.layers.6.post_attention_layernorm.bias": "pytorch_model-00002-of-00008.bin",
333
+ "transformer.layers.6.post_attention_layernorm.weight": "pytorch_model-00002-of-00008.bin",
334
+ "transformer.layers.7.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
335
+ "transformer.layers.7.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
336
+ "transformer.layers.7.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
337
+ "transformer.layers.7.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
338
+ "transformer.layers.7.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
339
+ "transformer.layers.7.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
340
+ "transformer.layers.7.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
341
+ "transformer.layers.7.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
342
+ "transformer.layers.7.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
343
+ "transformer.layers.7.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
344
+ "transformer.layers.7.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
345
+ "transformer.layers.7.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
346
+ "transformer.layers.7.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
347
+ "transformer.layers.8.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
348
+ "transformer.layers.8.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
349
+ "transformer.layers.8.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
350
+ "transformer.layers.8.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
351
+ "transformer.layers.8.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
352
+ "transformer.layers.8.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
353
+ "transformer.layers.8.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
354
+ "transformer.layers.8.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
355
+ "transformer.layers.8.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
356
+ "transformer.layers.8.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
357
+ "transformer.layers.8.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
358
+ "transformer.layers.8.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
359
+ "transformer.layers.8.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
360
+ "transformer.layers.9.attention.dense.bias": "pytorch_model-00003-of-00008.bin",
361
+ "transformer.layers.9.attention.dense.weight": "pytorch_model-00003-of-00008.bin",
362
+ "transformer.layers.9.attention.query_key_value.bias": "pytorch_model-00003-of-00008.bin",
363
+ "transformer.layers.9.attention.query_key_value.weight": "pytorch_model-00003-of-00008.bin",
364
+ "transformer.layers.9.attention.rotary_emb.inv_freq": "pytorch_model-00003-of-00008.bin",
365
+ "transformer.layers.9.input_layernorm.bias": "pytorch_model-00003-of-00008.bin",
366
+ "transformer.layers.9.input_layernorm.weight": "pytorch_model-00003-of-00008.bin",
367
+ "transformer.layers.9.mlp.dense_4h_to_h.bias": "pytorch_model-00003-of-00008.bin",
368
+ "transformer.layers.9.mlp.dense_4h_to_h.weight": "pytorch_model-00003-of-00008.bin",
369
+ "transformer.layers.9.mlp.dense_h_to_4h.bias": "pytorch_model-00003-of-00008.bin",
370
+ "transformer.layers.9.mlp.dense_h_to_4h.weight": "pytorch_model-00003-of-00008.bin",
371
+ "transformer.layers.9.post_attention_layernorm.bias": "pytorch_model-00003-of-00008.bin",
372
+ "transformer.layers.9.post_attention_layernorm.weight": "pytorch_model-00003-of-00008.bin",
373
+ "transformer.word_embeddings.weight": "pytorch_model-00001-of-00008.bin"
374
+ }
375
+ }
quantization.py ADDED
@@ -0,0 +1,187 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from torch.nn import Linear
2
+ from torch.nn.parameter import Parameter
3
+
4
+ import bz2
5
+ import torch
6
+ import base64
7
+ import ctypes
8
+
9
+ from typing import List
10
+ from cpm_kernels.kernels.base import LazyKernelCModule, KernelFunction, round_up
11
+
12
+
13
+ class W8A16Linear(torch.autograd.Function):
14
+ @staticmethod
15
+ def forward(ctx, inp: torch.Tensor, quant_w: torch.Tensor, scale_w: torch.Tensor, weight_bit_width):
16
+ ctx.inp_shape = inp.size()
17
+ ctx.weight_shape = quant_w.size()
18
+ ctx.weight_bit_width = weight_bit_width
19
+ out_features = quant_w.size(0)
20
+ inp = inp.contiguous().view(-1, inp.size(-1))
21
+ weight = extract_weight_to_half(quant_w, scale_w, weight_bit_width)
22
+ output = inp.mm(weight.t())
23
+ ctx.save_for_backward(inp, quant_w, scale_w)
24
+ return output.view(*(ctx.inp_shape[:-1] + (out_features,)))
25
+
26
+ @staticmethod
27
+ def backward(ctx, grad_output: torch.Tensor):
28
+ inp, quant_w, scale_w = ctx.saved_tensors
29
+ weight = extract_weight_to_half(quant_w, scale_w, ctx.weight_bit_width)
30
+ grad_output = grad_output.contiguous().view(-1, weight.size(0))
31
+ grad_input = grad_output.mm(weight)
32
+ grad_weight = grad_output.t().mm(inp)
33
+ return grad_input.view(ctx.inp_shape), grad_weight.view(ctx.weight_shape), None
34
+
35
+
36
+ class Kernel:
37
+ def __init__(self, code: bytes, function_names: List[str]):
38
+ self.code = code
39
+ self._function_names = function_names
40
+ self._cmodule = LazyKernelCModule(self.code)
41
+
42
+ for name in self._function_names:
43
+ setattr(self, name, KernelFunction(self._cmodule, name))
44
+
45
+
46
+ quantization_code = "$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"
47
+
48
+ kernels = Kernel(
49
+ bz2.decompress(base64.b64decode(quantization_code)),
50
+ [
51
+ "int4WeightCompression",
52
+ "int4WeightExtractionFloat",
53
+ "int4WeightExtractionHalf",
54
+ "int8WeightExtractionFloat",
55
+ "int8WeightExtractionHalf",
56
+ ],
57
+ )
58
+
59
+
60
+ def compress_int4_weight(weight: torch.Tensor): # (n, m)
61
+ with torch.cuda.device(weight.device):
62
+ n, m = weight.size(0), weight.size(1)
63
+ assert m % 2 == 0
64
+ m = m // 2
65
+ out = torch.empty(n, m, dtype=torch.int8, device="cuda")
66
+ stream = torch.cuda.current_stream()
67
+
68
+ gridDim = (n, 1, 1)
69
+ blockDim = (min(round_up(m, 32), 1024), 1, 1)
70
+
71
+ kernels.int4WeightCompression(
72
+ gridDim,
73
+ blockDim,
74
+ 0,
75
+ stream,
76
+ [ctypes.c_void_p(weight.data_ptr()), ctypes.c_void_p(out.data_ptr()), ctypes.c_int32(n), ctypes.c_int32(m)],
77
+ )
78
+ return out
79
+
80
+
81
+ def extract_weight_to_half(weight: torch.Tensor, scale_list: torch.Tensor, source_bit_width: int):
82
+ if source_bit_width == 8:
83
+ func = kernels.int8WeightExtractionHalf
84
+ elif source_bit_width == 4:
85
+ func = kernels.int4WeightExtractionHalf
86
+ else:
87
+ assert False, "Unsupported bit-width"
88
+
89
+ with torch.cuda.device(weight.device):
90
+ n, m = weight.size(0), weight.size(1)
91
+ out = torch.empty(n, m * (8 // source_bit_width), dtype=torch.half, device="cuda")
92
+ stream = torch.cuda.current_stream()
93
+
94
+ gridDim = (n, 1, 1)
95
+ blockDim = (min(round_up(m, 32), 1024), 1, 1)
96
+
97
+ func(
98
+ gridDim,
99
+ blockDim,
100
+ 0,
101
+ stream,
102
+ [
103
+ ctypes.c_void_p(weight.data_ptr()),
104
+ ctypes.c_void_p(scale_list.data_ptr()),
105
+ ctypes.c_void_p(out.data_ptr()),
106
+ ctypes.c_int32(n),
107
+ ctypes.c_int32(m),
108
+ ],
109
+ )
110
+ return out
111
+
112
+
113
+ class QuantizedLinear(Linear):
114
+ def __init__(self, weight_bit_width: int, weight_tensor=None, bias_tensor=None, *args, **kwargs):
115
+ super(QuantizedLinear, self).__init__(*args, **kwargs)
116
+ self.weight_bit_width = weight_bit_width
117
+
118
+ shape = self.weight.shape
119
+ del self.weight
120
+
121
+ if weight_tensor is None:
122
+ self.weight = torch.empty(
123
+ shape[0], shape[1] * weight_bit_width // 8, dtype=torch.int8, device=kwargs["device"]
124
+ )
125
+ self.weight_scale = torch.empty(shape[0], dtype=kwargs["params_dtype"], device=kwargs["device"])
126
+ else:
127
+ self.weight_scale = (weight_tensor.abs().max(dim=-1).values / ((2 ** (weight_bit_width - 1)) - 1)).half()
128
+ self.weight = torch.round(weight_tensor / self.weight_scale[:, None]).to(torch.int8)
129
+ if weight_bit_width == 4:
130
+ self.weight = compress_int4_weight(self.weight)
131
+
132
+ self.weight = Parameter(self.weight.to(kwargs["device"]), requires_grad=False)
133
+ self.weight_scale = Parameter(self.weight_scale.to(kwargs["device"]), requires_grad=False)
134
+ self.bias = Parameter(bias_tensor.to(kwargs["device"]), requires_grad=False)
135
+
136
+ def forward(self, input):
137
+ output = W8A16Linear.apply(input, self.weight, self.weight_scale, self.weight_bit_width)
138
+ if self.bias is not None:
139
+ output = output + self.bias
140
+ return output
141
+
142
+
143
+ def quantize(model, weight_bit_width):
144
+ """Replace fp16 linear with quantized linear"""
145
+
146
+ for layer in model.layers:
147
+ layer.attention.query_key_value = QuantizedLinear(
148
+ weight_bit_width=weight_bit_width,
149
+ weight_tensor=layer.attention.query_key_value.weight.to(torch.cuda.current_device()),
150
+ bias_tensor=layer.attention.query_key_value.bias,
151
+ in_features=layer.attention.query_key_value.in_features,
152
+ out_features=layer.attention.query_key_value.out_features,
153
+ bias=True,
154
+ dtype=torch.half,
155
+ device=layer.attention.query_key_value.weight.device,
156
+ )
157
+ layer.attention.dense = QuantizedLinear(
158
+ weight_bit_width=weight_bit_width,
159
+ weight_tensor=layer.attention.dense.weight.to(torch.cuda.current_device()),
160
+ bias_tensor=layer.attention.dense.bias,
161
+ in_features=layer.attention.dense.in_features,
162
+ out_features=layer.attention.dense.out_features,
163
+ bias=True,
164
+ dtype=torch.half,
165
+ device=layer.attention.dense.weight.device,
166
+ )
167
+ layer.mlp.dense_h_to_4h = QuantizedLinear(
168
+ weight_bit_width=weight_bit_width,
169
+ weight_tensor=layer.mlp.dense_h_to_4h.weight.to(torch.cuda.current_device()),
170
+ bias_tensor=layer.mlp.dense_h_to_4h.bias,
171
+ in_features=layer.mlp.dense_h_to_4h.in_features,
172
+ out_features=layer.mlp.dense_h_to_4h.out_features,
173
+ bias=True,
174
+ dtype=torch.half,
175
+ device=layer.mlp.dense_h_to_4h.weight.device,
176
+ )
177
+ layer.mlp.dense_4h_to_h = QuantizedLinear(
178
+ weight_bit_width=weight_bit_width,
179
+ weight_tensor=layer.mlp.dense_4h_to_h.weight.to(torch.cuda.current_device()),
180
+ bias_tensor=layer.mlp.dense_4h_to_h.bias,
181
+ in_features=layer.mlp.dense_4h_to_h.in_features,
182
+ out_features=layer.mlp.dense_4h_to_h.out_features,
183
+ bias=True,
184
+ dtype=torch.half,
185
+ device=layer.mlp.dense_4h_to_h.weight.device,
186
+ )
187
+ return model
tokenization_chatglm.py ADDED
@@ -0,0 +1,347 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """Tokenization classes for ChatGLM."""
2
+ import sys
3
+ import unicodedata
4
+ from typing import List, Optional, Union
5
+ from functools import lru_cache
6
+ import os
7
+ import collections
8
+ import re
9
+
10
+ from transformers.tokenization_utils import PreTrainedTokenizer
11
+ from icetk.text_tokenizer import TextTokenizer
12
+ from icetk.utils import auto_create
13
+ import icetk.sentencepiece_model_pb2 as sp_model
14
+ from transformers.utils import logging
15
+
16
+ logger = logging.get_logger(__name__)
17
+
18
+ VOCAB_FILES_NAMES = {"vocab_file": "ice_text.model"}
19
+
20
+ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
21
+ "THUDM/chatglm-6b": 2048,
22
+ }
23
+
24
+
25
+ class SPTokenizer:
26
+ def __init__(
27
+ self,
28
+ vocab_file,
29
+ max_blank_length=80,
30
+ byte_fallback=True,
31
+ ):
32
+ assert vocab_file is not None
33
+ self.vocab_file = vocab_file
34
+ self.special_tokens = ["[MASK]", "[gMASK]", "[sMASK]", "<unused_0>", "<sop>", "<eop>", "<ENC>", "<dBLOCK>"]
35
+ self.max_blank_length = max_blank_length
36
+ self.byte_fallback = byte_fallback
37
+ self.text_tokenizer = self._build_text_tokenizer(encode_special_tokens=False)
38
+ self.special_text_tokenizer = self._build_text_tokenizer(encode_special_tokens=True)
39
+
40
+ @staticmethod
41
+ def _configure_tokenizer(
42
+ text_tokenizer: TextTokenizer,
43
+ special_tokens: List[str],
44
+ max_blank_length: int,
45
+ byte_fallback: bool,
46
+ encode_special_tokens=False,
47
+ ):
48
+ # special token
49
+ special_token_type = 4 if encode_special_tokens else 3 # 3 - CONTROL, 4 - USER_DEFINE
50
+ for token in special_tokens:
51
+ text_tokenizer.proto.pieces.append(
52
+ sp_model.ModelProto.SentencePiece(piece=token, score=0.0, type=special_token_type)
53
+ )
54
+ # whitespaces
55
+ for token in [SPTokenizer.get_tab_token()] + [
56
+ SPTokenizer.get_blank_token(i) for i in range(2, max_blank_length + 1)
57
+ ]:
58
+ text_tokenizer.proto.pieces.append(sp_model.ModelProto.SentencePiece(piece=token, score=0.0, type=4))
59
+ # byte fallback
60
+ if byte_fallback:
61
+ text_tokenizer.proto.trainer_spec.byte_fallback = True
62
+ for i in range(256):
63
+ text_tokenizer.proto.pieces.append(
64
+ sp_model.ModelProto.SentencePiece(piece="<0x{:02X}>".format(i), score=0.0, type=6)
65
+ )
66
+ text_tokenizer.refresh()
67
+
68
+ def _build_text_tokenizer(self, encode_special_tokens=False):
69
+ tokenizer = TextTokenizer(self.vocab_file)
70
+ self._configure_tokenizer(
71
+ tokenizer, self.special_tokens, self.max_blank_length, self.byte_fallback, encode_special_tokens
72
+ )
73
+ return tokenizer
74
+
75
+ def _get_text_tokenizer(self, encode_special_tokens=False):
76
+ if encode_special_tokens:
77
+ return self.special_text_tokenizer
78
+ else:
79
+ return self.text_tokenizer
80
+
81
+ @staticmethod
82
+ def get_blank_token(length: int):
83
+ assert length >= 2
84
+ return f"<|blank_{length}|>"
85
+
86
+ @staticmethod
87
+ def get_tab_token():
88
+ return f"<|tab|>"
89
+
90
+ @property
91
+ def num_image_tokens(self):
92
+ return 20000
93
+
94
+ @property
95
+ def num_text_tokens(self):
96
+ return self.text_tokenizer.num_tokens
97
+
98
+ @property
99
+ def num_tokens(self):
100
+ return self.num_image_tokens + self.num_text_tokens
101
+
102
+ @staticmethod
103
+ def _encode_whitespaces(text: str, max_len: int = 80):
104
+ text = text.replace("\t", SPTokenizer.get_tab_token())
105
+ for i in range(max_len, 1, -1):
106
+ text = text.replace(" " * i, SPTokenizer.get_blank_token(i))
107
+ return text
108
+
109
+ def _preprocess(self, text: str, linebreak=True, whitespaces=True):
110
+ if linebreak:
111
+ text = text.replace("\n", "<n>")
112
+ if whitespaces:
113
+ text = self._encode_whitespaces(text, max_len=self.max_blank_length)
114
+ return text
115
+
116
+ def encode(
117
+ self, text: str, linebreak=True, whitespaces=True, special_tokens=False, add_dummy_prefix=True
118
+ ) -> List[int]:
119
+ """
120
+ @param text: Text to encode.
121
+ @param linebreak: Whether to encode newline (\n) in text.
122
+ @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
123
+ @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
124
+ @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
125
+ """
126
+ text = self._preprocess(text, linebreak, whitespaces)
127
+ if not add_dummy_prefix:
128
+ text = "<n>" + text
129
+ tmp = self._get_text_tokenizer(encode_special_tokens=special_tokens).encode(text)
130
+ tokens = [x + self.num_image_tokens for x in tmp]
131
+ return tokens if add_dummy_prefix else tokens[2:]
132
+
133
+ def decode(self, text_ids: List[int], special_tokens=False) -> str:
134
+ ids = [int(_id) - self.num_image_tokens for _id in text_ids]
135
+ text = self._get_text_tokenizer(encode_special_tokens=special_tokens).decode(ids)
136
+ text = text.replace("<n>", "\n")
137
+ text = text.replace(SPTokenizer.get_tab_token(), "\t")
138
+ for i in range(2, self.max_blank_length + 1):
139
+ text = text.replace(self.get_blank_token(i), " " * i)
140
+ return text
141
+
142
+ def tokenize(
143
+ self, text: str, linebreak=True, whitespaces=True, special_tokens=False, add_dummy_prefix=True
144
+ ) -> List[str]:
145
+ """
146
+ @param text: Text to encode.
147
+ @param linebreak: Whether to encode newline (\n) in text.
148
+ @param whitespaces: Whether to encode multiple whitespaces or tab in text, useful for source code encoding.
149
+ @param special_tokens: Whether to encode special token ([MASK], [gMASK], etc.) in text.
150
+ @param add_dummy_prefix: Whether to add dummy blank space in the beginning.
151
+ """
152
+ text = self._preprocess(text, linebreak, whitespaces)
153
+ if not add_dummy_prefix:
154
+ text = "<n>" + text
155
+ tokens = self._get_text_tokenizer(encode_special_tokens=special_tokens).tokenize(text)
156
+ return tokens if add_dummy_prefix else tokens[2:]
157
+
158
+ def __getitem__(self, x: Union[int, str]):
159
+ if isinstance(x, int):
160
+ if x < self.num_image_tokens:
161
+ return "<image_{}>".format(x)
162
+ else:
163
+ return self.text_tokenizer.convert_id_to_token(x - self.num_image_tokens)
164
+ elif isinstance(x, str):
165
+ if x.startswith("<image_") and x.endswith(">") and x[7:-1].isdigit():
166
+ return int(x[7:-1])
167
+ else:
168
+ return self.text_tokenizer.convert_token_to_id(x) + self.num_image_tokens
169
+ else:
170
+ raise ValueError("The key should be str or int.")
171
+
172
+
173
+ class ChatGLMTokenizer(PreTrainedTokenizer):
174
+ """
175
+ Construct a ChatGLM tokenizer. Based on byte-level Byte-Pair-Encoding.
176
+
177
+ Args:
178
+ vocab_file (`str`):
179
+ Path to the vocabulary file.
180
+ """
181
+
182
+ vocab_files_names = VOCAB_FILES_NAMES
183
+ max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
184
+ model_input_names = ["input_ids"]
185
+
186
+ def __init__(
187
+ self,
188
+ vocab_file,
189
+ do_lower_case=False,
190
+ remove_space=False,
191
+ bos_token='sop',
192
+ eos_token='eos',
193
+ eop_token='eop',
194
+ mask_token='[MASK]',
195
+ gmask_token='[gMASK]',
196
+ padding_side="left",
197
+ **kwargs
198
+ ) -> None:
199
+ super().__init__(
200
+ do_lower_case=do_lower_case,
201
+ remove_space=remove_space,
202
+ padding_side=padding_side,
203
+ **kwargs
204
+ )
205
+
206
+ self.do_lower_case = do_lower_case
207
+ self.remove_space = remove_space
208
+ self.vocab_file = vocab_file
209
+
210
+ self.bos_token = bos_token
211
+ self.eos_token = eos_token
212
+ self.eop_token = eop_token
213
+ self.mask_token = mask_token
214
+ self.gMASK_token = gmask_token
215
+
216
+ self.sp_tokenizer = SPTokenizer(vocab_file)
217
+
218
+ """ Initialisation """
219
+
220
+ @property
221
+ def eop_token_id(self) -> Optional[int]:
222
+ """
223
+ `Optional[int]`: Id of the end of sentence token in the vocabulary. Returns `None` if the token has not been
224
+ set.
225
+ """
226
+ if self.eop_token is None:
227
+ return None
228
+ return self.convert_tokens_to_ids(self.eop_token)
229
+
230
+ @property
231
+ def vocab_size(self):
232
+ """ Returns vocab size """
233
+ return self.sp_tokenizer.num_tokens
234
+
235
+ def get_vocab(self):
236
+ """ Returns vocab as a dict """
237
+ vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
238
+ vocab.update(self.added_tokens_encoder)
239
+ return vocab
240
+
241
+ def preprocess_text(self, inputs):
242
+ if self.remove_space:
243
+ outputs = " ".join(inputs.strip().split())
244
+ else:
245
+ outputs = inputs
246
+
247
+ if self.do_lower_case:
248
+ outputs = outputs.lower()
249
+
250
+ return outputs
251
+
252
+ def _tokenize(self, text, **kwargs):
253
+ """ Returns a tokenized string. """
254
+ text = self.preprocess_text(text)
255
+
256
+ seq = self.sp_tokenizer.tokenize(text)
257
+
258
+ return seq
259
+
260
+ def decode(
261
+ self,
262
+ token_ids: Union[List[int], List[List[int]]],
263
+ skip_special_tokens: bool = False,
264
+ clean_up_tokenization_spaces: bool = True,
265
+ spaces_between_special_tokens: bool = True,
266
+ **kwargs
267
+ ) -> str:
268
+ if isinstance(token_ids[0], list):
269
+ tokens = []
270
+ for single_token_ids in token_ids:
271
+ if self.pad_token_id in single_token_ids: # remove pad
272
+ single_token_ids = list(filter((self.pad_token_id).__ne__, single_token_ids))
273
+ tokens.append(self.sp_tokenizer.decode(single_token_ids))
274
+ return (tokens)
275
+ else:
276
+ if self.pad_token_id in token_ids: # remove pad
277
+ token_ids = list(filter((self.pad_token_id).__ne__, token_ids))
278
+ return self.sp_tokenizer.decode(token_ids)
279
+
280
+ def _convert_token_to_id(self, token):
281
+ """ Converts a token (str) in an id using the vocab. """
282
+ return self.sp_tokenizer[token]
283
+
284
+ def _convert_id_to_token(self, index):
285
+ """Converts an index (integer) in a token (str) using the vocab."""
286
+ return self.sp_tokenizer[index]
287
+
288
+ def save_vocabulary(self, save_directory, filename_prefix=None):
289
+ """
290
+ Save the vocabulary and special tokens file to a directory.
291
+
292
+ Args:
293
+ save_directory (`str`):
294
+ The directory in which to save the vocabulary.
295
+ filename_prefix (`str`, *optional*):
296
+ An optional prefix to add to the named of the saved files.
297
+
298
+ Returns:
299
+ `Tuple(str)`: Paths to the files saved.
300
+ """
301
+ if os.path.isdir(save_directory):
302
+ vocab_file = os.path.join(
303
+ save_directory, VOCAB_FILES_NAMES["vocab_file"]
304
+ )
305
+ else:
306
+ vocab_file = save_directory
307
+
308
+ with open(self.vocab_file, 'rb') as fin:
309
+ proto_str = fin.read()
310
+
311
+ with open(vocab_file, "wb") as writer:
312
+ writer.write(proto_str)
313
+
314
+ return (vocab_file,)
315
+
316
+ def build_inputs_with_special_tokens(
317
+ self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
318
+ ) -> List[int]:
319
+ """
320
+ Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
321
+ adding special tokens. A BERT sequence has the following format:
322
+
323
+ - single sequence: `[CLS] X [SEP]`
324
+ - pair of sequences: `[CLS] A [SEP] B [SEP]`
325
+
326
+ Args:
327
+ token_ids_0 (`List[int]`):
328
+ List of IDs to which the special tokens will be added.
329
+ token_ids_1 (`List[int]`, *optional*):
330
+ Optional second list of IDs for sequence pairs.
331
+
332
+ Returns:
333
+ `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
334
+ """
335
+ if token_ids_1 is not None:
336
+ token_ids_0 += token_ids_1
337
+ mask_ids = self.sp_tokenizer[self.mask_token]
338
+ gmask_ids = self.sp_tokenizer[self.gMASK_token]
339
+ if mask_ids not in token_ids_0 and gmask_ids not in token_ids_0:
340
+ token_ids_0 += [gmask_ids]
341
+
342
+ if token_ids_0[-1] != mask_ids and token_ids_0[-1] != gmask_ids:
343
+ token_ids_0 += [self.sp_tokenizer[self.eos_token]]
344
+
345
+ token_ids_0 += [self.sp_tokenizer[self.bos_token]]
346
+
347
+ return token_ids_0
tokenizer_config.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name_or_path": "THUDM/chatglm-6b",
3
+ "bos_token": "<sop>",
4
+ "eop_token": "<eop>",
5
+ "eos_token": "</s>",
6
+ "gmask_token": "[gMASK]",
7
+ "mask_token": "[MASK]",
8
+ "pad_token": "<pad>",
9
+ "unk_token": "<unk>",
10
+ "remove_space": false,
11
+ "do_lower_case": false,
12
+ "tokenizer_class": "ChatGLMTokenizer",
13
+ "auto_map": {
14
+ "AutoTokenizer": [
15
+ "tokenization_chatglm.ChatGLMTokenizer",
16
+ null
17
+ ]
18
+ }
19
+ }