wip

modelscope · Dec 4, 2024 · c4539fb · c4539fb
1 parent 738e9fd
commit c4539fb
Show file tree

Hide file tree

Showing 6 changed files with 17 additions and 25 deletions.
diff --git a/docs/source/GetStarted/ReleaseNote.md → docs/source/Instruction/ReleaseNote3.0.md b/docs/source/GetStarted/ReleaseNote.md → docs/source/Instruction/ReleaseNote3.0.md
diff --git a/docs/source/Instruction/智能体.md → docs/source/Instruction/智能体的支持.md b/docs/source/Instruction/智能体.md → docs/source/Instruction/智能体的支持.md
diff --git a/docs/source/index.rst b/docs/source/index.rst
@@ -10,26 +10,24 @@ Swift DOCUMENTATION
    :caption: Get Started
 
    GetStarted/SWIFT安装.md
-   GetStarted/界面训练推理.md
-   GetStarted/推送模型.md
-   GetStarted/使用tuners.md
-   GetStarted/ResTuning.md
-   GetStarted/SCEdit.md
-   GetStarted/在SWIFT内使用PEFT.md
+   GetStarted/快速开始.md
+   GetStarted/界面使用.md
 
 .. toctree::
    :maxdepth: 2
    :caption: Instruction
 
-   Instruction/index.md
-   Instruction/LLM微调文档.md
-   Instruction/LLM推理文档.md
-   Instruction/LLM评测文档.md
-   Instruction/LLM量化与导出文档.md
-   Instruction/LLM实验文档.md
+   Instruction/预训练及微调.md
+   Instruction/人类对齐.md
+   Instruction/推理和部署.md
+   Instruction/评测.md
+   Instruction/导出.md
    Instruction/命令行参数.md
+   Instruction/NPU支持.md
+   Instruction/使用tuners.md
    Instruction/支持的模型和数据集.md
-   Instruction/自定义与拓展.md
+   Instruction/推送模型.md
+   Instruction/ReleaseNote3.0.md
    Instruction/常见问题整理.md
 
 

diff --git a/swift/llm/argument/base_args/model_args.py b/swift/llm/argument/base_args/model_args.py
@@ -102,12 +102,12 @@ def _init_model_info(self) -> torch.dtype:
         self.model_dir = self.model_info.model_dir
         self.model_type = self.model_info.model_type
         if self.rope_scaling is not None and isinstance(self.rope_scaling, str):
-            assert self.max_length is not None
-            max_model_len_no_scaling = self.model_info.max_model_len_no_scaling
+            assert self.max_length is not None, 'Use max_model_len together with rope_scaling'
             rope_scaling = self.model_info.rope_scaling or {}
+            max_model_len = self.model_info.max_model_len
             rope_scaling_factor = 1.0
-            if max_model_len_no_scaling:
-                rope_scaling_factor = max(float(math.ceil(self.max_length / max_model_len_no_scaling)), 1.0)
+            if max_model_len:
+                rope_scaling_factor = max(float(math.ceil(self.max_length / max_model_len)), 1.0)
             if rope_scaling:
                 rope_scaling_factor = max(rope_scaling.get('factor', -1), rope_scaling_factor)
                 rope_scaling['type'] = self.rope_scaling

diff --git a/swift/llm/model/register.py b/swift/llm/model/register.py
@@ -371,7 +371,6 @@ def _get_model_info(model_dir: str, model_type: Optional[str], quantization_conf
     quant_info = HfConfigFactory.get_quant_info(config_dict) or {}
     torch_dtype = HfConfigFactory.get_torch_dtype(config_dict, quant_info)
     max_model_len = HfConfigFactory.get_max_model_len(config_dict)
-    max_model_len_no_scaling = HfConfigFactory.get_max_model_len(config_dict, ignore_rope_scaling=True)
     rope_scaling = HfConfigFactory.get_config_attr(config_dict, 'rope_scaling')
 
     if model_type is None:
@@ -382,7 +381,7 @@ def _get_model_info(model_dir: str, model_type: Optional[str], quantization_conf
         elif len(model_types) == 1:
             model_type = model_types[0]
 
-    res = ModelInfo(model_type, model_dir, torch_dtype, max_model_len, max_model_len_no_scaling,
+    res = ModelInfo(model_type, model_dir, torch_dtype, max_model_len,
                     quant_info.get('quant_method'), quant_info.get('quant_bits'), rope_scaling)
     return res
 

diff --git a/swift/llm/model/utils.py b/swift/llm/model/utils.py
@@ -56,7 +56,6 @@ class ModelInfo:
     model_dir: str
     torch_dtype: torch.dtype
     max_model_len: int
-    max_model_len_no_scaling: int
     quant_method: Literal['gptq', 'awq', 'bnb', 'aqlm', 'hqq', None]
     quant_bits: int
     rope_scaling: Dict[str, Any]
@@ -123,7 +122,7 @@ def set_config_attr(config: Union[PretrainedConfig, Dict[str, Any]], attr_name:
                 setattr(config, attr_name, value)
 
     @staticmethod
-    def get_max_model_len(config: Union[PretrainedConfig, Dict[str, Any]], ignore_rope_scaling=False) -> Optional[int]:
+    def get_max_model_len(config: Union[PretrainedConfig, Dict[str, Any]]) -> Optional[int]:
         """Get the max length supported by the model"""
         INF = int(1e9)
         max_model_len = INF
@@ -145,10 +144,6 @@ def get_max_model_len(config: Union[PretrainedConfig, Dict[str, Any]], ignore_ro
                 max_model_len = min(max_model_len, max_len_key)
         if max_model_len == INF:
             max_model_len = None
-
-        if (not ignore_rope_scaling and max_model_len and getattr(config, 'rope_scaling', None)
-                and config.rope_scaling.get('factor')):
-            max_model_len = max(int(max_model_len * config.rope_scaling.get('factor')), max_model_len)
         return max_model_len
 
     @staticmethod