Merge main
Browse files- LICENSE +1 -1
- README.md +7 -7
- modeling_chatglm.py +3 -3
- tokenization_chatglm.py +77 -91
LICENSE
CHANGED
@@ -45,7 +45,7 @@ The glm-4-9b License
|
|
45 |
|
46 |
2. License
|
47 |
|
48 |
-
|
49 |
This license allows you to use all open source models in this repository for free for academic research. For users who wish to use the models for commercial purposes, please do so [here](https://open.bigmodel.cn/mla/form)
|
50 |
Complete registration. Registered users are free to use this model for commercial activities, but must comply with all terms and conditions of this license.
|
51 |
The copyright notice and this license notice shall be included in all copies or substantial portions of the Software.
|
|
|
45 |
|
46 |
2. License
|
47 |
|
48 |
+
Under the terms and conditions of this license, the Licensor hereby grants you a non-exclusive, worldwide, non-transferable, non-sublicensable, revocable, royalty-free copyright license.
|
49 |
This license allows you to use all open source models in this repository for free for academic research. For users who wish to use the models for commercial purposes, please do so [here](https://open.bigmodel.cn/mla/form)
|
50 |
Complete registration. Registered users are free to use this model for commercial activities, but must comply with all terms and conditions of this license.
|
51 |
The copyright notice and this license notice shall be included in all copies or substantial portions of the Software.
|
README.md
CHANGED
@@ -2,15 +2,15 @@
|
|
2 |
license: other
|
3 |
license_name: glm-4
|
4 |
license_link: https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/LICENSE
|
5 |
-
|
6 |
language:
|
7 |
-
|
8 |
-
|
9 |
tags:
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
inference: false
|
|
|
14 |
---
|
15 |
|
16 |
# GLM-4-9B-Chat
|
@@ -168,4 +168,4 @@ GLM-4 模型的权重的使用则需要遵循 [LICENSE](LICENSE)。
|
|
168 |
pages={320--335},
|
169 |
year={2022}
|
170 |
}
|
171 |
-
```
|
|
|
2 |
license: other
|
3 |
license_name: glm-4
|
4 |
license_link: https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/LICENSE
|
|
|
5 |
language:
|
6 |
+
- zh
|
7 |
+
- en
|
8 |
tags:
|
9 |
+
- glm
|
10 |
+
- chatglm
|
11 |
+
- thudm
|
12 |
inference: false
|
13 |
+
pipeline_tag: text-generation
|
14 |
---
|
15 |
|
16 |
# GLM-4-9B-Chat
|
|
|
168 |
pages={320--335},
|
169 |
year={2022}
|
170 |
}
|
171 |
+
```
|
modeling_chatglm.py
CHANGED
@@ -21,7 +21,7 @@ from transformers.modeling_outputs import (
|
|
21 |
SequenceClassifierOutputWithPast,
|
22 |
)
|
23 |
from transformers.modeling_utils import PreTrainedModel
|
24 |
-
from transformers.utils import logging
|
25 |
from transformers.generation.logits_process import LogitsProcessor
|
26 |
from transformers.generation.utils import LogitsProcessorList, StoppingCriteriaList, GenerationConfig, ModelOutput
|
27 |
|
@@ -29,7 +29,7 @@ from .configuration_chatglm import ChatGLMConfig
|
|
29 |
|
30 |
# flags required to enable jit fusion kernels
|
31 |
|
32 |
-
if sys.platform != 'darwin':
|
33 |
torch._C._jit_set_profiling_mode(False)
|
34 |
torch._C._jit_set_profiling_executor(False)
|
35 |
torch._C._jit_override_can_fuse_on_cpu(True)
|
@@ -1139,7 +1139,7 @@ class ChatGLMForSequenceClassification(ChatGLMPreTrainedModel):
|
|
1139 |
self.num_labels = config.num_labels
|
1140 |
self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
|
1141 |
|
1142 |
-
self.classifier_head = nn.Linear(config.hidden_size, config.num_labels, bias=True, dtype=
|
1143 |
if config.classifier_dropout is not None:
|
1144 |
self.dropout = nn.Dropout(config.classifier_dropout)
|
1145 |
else:
|
|
|
21 |
SequenceClassifierOutputWithPast,
|
22 |
)
|
23 |
from transformers.modeling_utils import PreTrainedModel
|
24 |
+
from transformers.utils import logging, is_torch_npu_available
|
25 |
from transformers.generation.logits_process import LogitsProcessor
|
26 |
from transformers.generation.utils import LogitsProcessorList, StoppingCriteriaList, GenerationConfig, ModelOutput
|
27 |
|
|
|
29 |
|
30 |
# flags required to enable jit fusion kernels
|
31 |
|
32 |
+
if sys.platform != 'darwin' and not is_torch_npu_available():
|
33 |
torch._C._jit_set_profiling_mode(False)
|
34 |
torch._C._jit_set_profiling_executor(False)
|
35 |
torch._C._jit_override_can_fuse_on_cpu(True)
|
|
|
1139 |
self.num_labels = config.num_labels
|
1140 |
self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
|
1141 |
|
1142 |
+
self.classifier_head = nn.Linear(config.hidden_size, config.num_labels, bias=True, dtype=config.torch_dtype)
|
1143 |
if config.classifier_dropout is not None:
|
1144 |
self.dropout = nn.Dropout(config.classifier_dropout)
|
1145 |
else:
|
tokenization_chatglm.py
CHANGED
@@ -130,6 +130,8 @@ class ChatGLM4Tokenizer(PreTrainedTokenizer):
|
|
130 |
prefix_tokens = [self.convert_tokens_to_ids("[gMASK]"), self.convert_tokens_to_ids("<sop>")]
|
131 |
return prefix_tokens
|
132 |
|
|
|
|
|
133 |
def build_single_message(self, role, metadata, message, tokenize=True):
|
134 |
assert role in ["system", "user", "assistant", "observation"], role
|
135 |
if tokenize:
|
@@ -142,97 +144,81 @@ class ChatGLM4Tokenizer(PreTrainedTokenizer):
|
|
142 |
return str(f"<|{role}|>{metadata}\n{message}")
|
143 |
|
144 |
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
# if tokenize:
|
221 |
-
# output = self.batch_encode_plus(
|
222 |
-
# [result] if isinstance(result[0], int) else result,
|
223 |
-
# padding=padding,
|
224 |
-
# truncation=truncation,
|
225 |
-
# max_length=max_length,
|
226 |
-
# return_tensors=return_tensors,
|
227 |
-
# is_split_into_words=True,
|
228 |
-
# add_special_tokens=False
|
229 |
-
# )
|
230 |
-
# if return_dict:
|
231 |
-
# return output
|
232 |
-
# else:
|
233 |
-
# return output["input_ids"]
|
234 |
-
# else:
|
235 |
-
# return result
|
236 |
|
237 |
|
238 |
def build_inputs_with_special_tokens(
|
|
|
130 |
prefix_tokens = [self.convert_tokens_to_ids("[gMASK]"), self.convert_tokens_to_ids("<sop>")]
|
131 |
return prefix_tokens
|
132 |
|
133 |
+
""" use chat_template, no need apply_chat_template
|
134 |
+
|
135 |
def build_single_message(self, role, metadata, message, tokenize=True):
|
136 |
assert role in ["system", "user", "assistant", "observation"], role
|
137 |
if tokenize:
|
|
|
144 |
return str(f"<|{role}|>{metadata}\n{message}")
|
145 |
|
146 |
|
147 |
+
def apply_chat_template(
|
148 |
+
self,
|
149 |
+
conversation: Union[List[Dict[str, str]], List[List[Dict[str, str]]], "Conversation"],
|
150 |
+
add_generation_prompt: bool = False,
|
151 |
+
tokenize: bool = True,
|
152 |
+
padding: bool = False,
|
153 |
+
truncation: bool = False,
|
154 |
+
max_length: Optional[int] = None,
|
155 |
+
return_tensors: Optional[Union[str, TensorType]] = None,
|
156 |
+
return_dict: bool = False,
|
157 |
+
tokenizer_kwargs: Optional[Dict[str, Any]] = None,
|
158 |
+
add_special_tokens: bool = True,
|
159 |
+
**kwargs,
|
160 |
+
) -> Union[str, List[int], List[str], List[List[int]], BatchEncoding]:
|
161 |
+
|
162 |
+
if return_dict and not tokenize:
|
163 |
+
raise ValueError(
|
164 |
+
"`return_dict=True` is incompatible with `tokenize=False`, because there is no dict "
|
165 |
+
"of tokenizer outputs to return."
|
166 |
+
)
|
167 |
+
|
168 |
+
def handle_single_conversation(conversation):
|
169 |
+
input_ids = self.get_prefix_tokens() if add_special_tokens else []
|
170 |
+
input_message = "[gMASK]<sop>" if add_special_tokens else ""
|
171 |
+
for item in conversation:
|
172 |
+
if item.get("tools"):
|
173 |
+
tools = item["tools"]
|
174 |
+
content = "你是一个名为 GhatGLM 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的,你的任务是针对用户的问题和要求提供适当的答复和支持。"
|
175 |
+
content += "\n\n# 可用工具"
|
176 |
+
for tool in tools:
|
177 |
+
if tool["type"] == "function":
|
178 |
+
function = tool["function"]
|
179 |
+
content += f"\n\n## {function['name']}\n\n{json.dumps(function, ensure_ascii=False, indent=4)}"
|
180 |
+
content += "\n在调用上述函数时,请使用 Json 格式表示调用的参数。"
|
181 |
+
elif tool["type"] == "python":
|
182 |
+
content += "\n\n## python\n\n当你向 `python` 发送包含 Python 代码的消息时,该代码将会在一个有状态的 Jupyter notebook 环境中执行。\n`python` 返回代码执行的输出,或在执行 60 秒后返回超时。\n`/mnt/data` 将会持久化存储你的文件。在此会话中,`python` 无法访问互联网。不要使用 `python` 进行任何网络请求或者在线 API 调用,这些在线内容的访问将不会成功。"
|
183 |
+
elif tool["type"] == "simple_browser":
|
184 |
+
content += "\n\n## simple_browser\n\n你可以使用 `simple_browser` 工具。该工具支持以下函数:\n`search(query: str, recency_days: int)`:使用搜索引擎进行查询并显示结果,可以使用 `recency_days` 参数控制搜索内容的时效性。\n`mclick(ids: list[int])`:获取一系列指定 id 的页面内容。每次调用时,须选择3-10个页面。选择多个角度的页面,同时尽可能选择可信任的信息来源。考虑到部分页面是无法加载的,你也可以多打开一些可能有用的页面而不用担心内容过多。\n`open_url(url: str)`:打开指定的 URL。\n\n使用 `【{引用 id}†{引用文本}】` 来引用内容。\n\n操作步骤:1. 使用 `search` 来获得信息列表; 2. 使用 `mclick` 来获取指定 ID 页面的内容; 3. 根据获得的内容进行回复。在回复中应当引用信息来源。\n 如果用户提供了 URL,也可以用 `open_url` 直接打开页面。\n如果初次搜索结果没有找到合适的信息,也可以再次使用 `search` 进行搜索。"
|
185 |
+
elif tool["type"] == "cogview":
|
186 |
+
content += "\n\n## cogview\n\n如果用户的请求中包含了对图像的描述,你可以使用 `cogview` 来生成图像并展示给用户。你需要向 `cogview` 发送图像描述,规则:\n- 发送给 `cogview` 的消息必须使用英语。用户的中文描述必须完全翻译为英语。\n- 应当尽可能详细地描述图像生成的需求,需求描述约 100 英文单词。\n- 保持用户原始描述的意图。不要虚构内容或者没见过的人物。\n- 如无特殊说明,所在地为中国,持有中国立场并遵循中国社会主义价值观。"
|
187 |
+
else:
|
188 |
+
raise NotImplementedError(f"Unknown tool type {tool['type']}")
|
189 |
+
input = self.build_single_message("system", "", content, tokenize=tokenize)
|
190 |
+
if tokenize:
|
191 |
+
input_ids.extend(input)
|
192 |
+
else:
|
193 |
+
input_message += input
|
194 |
+
if item["content"]:
|
195 |
+
input = self.build_single_message(
|
196 |
+
item["role"],
|
197 |
+
item.get("metadata", ""),
|
198 |
+
item["content"],
|
199 |
+
tokenize=tokenize
|
200 |
+
)
|
201 |
+
if tokenize:
|
202 |
+
input_ids.extend(input)
|
203 |
+
else:
|
204 |
+
input_message += input
|
205 |
+
if add_generation_prompt:
|
206 |
+
if tokenize:
|
207 |
+
input_ids.extend([self.convert_tokens_to_ids("<|assistant|>")])
|
208 |
+
else:
|
209 |
+
input_message += "<|assistant|>"
|
210 |
+
return input_ids if tokenize else input_message
|
211 |
+
|
212 |
+
# Main logic to handle different conversation formats
|
213 |
+
if isinstance(conversation, list) and all(isinstance(i, dict) for i in conversation):
|
214 |
+
result = handle_single_conversation(conversation)
|
215 |
+
elif isinstance(conversation, list) and all(isinstance(i, list) for i in conversation):
|
216 |
+
result = [handle_single_conversation(c) for c in conversation]
|
217 |
+
elif hasattr(conversation, "messages"):
|
218 |
+
result = handle_single_conversation(conversation.messages)
|
219 |
+
else:
|
220 |
+
raise ValueError("Invalid conversation format")
|
221 |
+
"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
222 |
|
223 |
|
224 |
def build_inputs_with_special_tokens(
|