fix alpaca (#2771)

Author: Jintao-Huang

swift/llm/__init__.py

@@ -18,7 +18,8 @@
     from .model import (register_model, MODEL_MAPPING, ModelType, get_model_tokenizer, safe_snapshot_download,
                         HfConfigFactory, ModelInfo, ModelMeta, ModelKeys, register_model_arch, MultiModelKeys,
                         ModelArch, get_model_arch, MODEL_ARCH_MAPPING, get_model_info_meta, get_model_name, ModelGroup,
-                        Model, get_model_tokenizer_with_flash_attn, get_model_tokenizer_multimodal, load_by_unsloth)
+                        Model, get_model_tokenizer_with_flash_attn, get_model_tokenizer_multimodal, load_by_unsloth,
+                        git_clone_github)
     from .dataset import (AlpacaPreprocessor, ResponsePreprocessor, MessagesPreprocessor, AutoPreprocessor,
                           DATASET_MAPPING, MediaResource, register_dataset, register_dataset_info, EncodePreprocessor,
                           LazyLLMDataset, ConstantLengthDataset, standard_keys, load_dataset, DATASET_TYPE,
@@ -51,7 +52,7 @@
             'ModelInfo', 'ModelMeta', 'ModelKeys', 'register_model_arch', 'MultiModelKeys', 'ModelArch',
             'MODEL_ARCH_MAPPING', 'get_model_arch', 'get_model_info_meta', 'get_model_name', 'register_model',
             'ModelGroup', 'Model', 'get_model_tokenizer_with_flash_attn', 'get_model_tokenizer_multimodal',
-            'load_by_unsloth'
+            'load_by_unsloth', 'git_clone_github'
         ],
         'dataset': [
             'AlpacaPreprocessor', 'ClsPreprocessor', 'ComposePreprocessor', 'MessagesPreprocessor', 'DATASET_MAPPING',

swift/llm/dataset/dataset/llm.py

@@ -9,17 +9,20 @@
 from ..register import DatasetMeta, SubsetDataset, register_dataset
 
 
-def _concat_inst_inp_alpaca_zh(inst: str, inp: str) -> str:
-    if inp.startswith('输入：'):
-        inp = inp[3:]
-    return f'{inst}\n{inp}'
+class AlpacaZhPreprocessor(AlpacaPreprocessor):
+
+    @classmethod
+    def concat_inst_input(cls, instruction, input_):
+        if input_ and input_.startswith('输入：'):
+            input_ = input_[3:]
+        return super().concat_inst_input(instruction, input_)
 
 
 register_dataset(
     DatasetMeta(
         ms_dataset_id='AI-ModelScope/alpaca-gpt4-data-zh',
         hf_dataset_id='llm-wizard/alpaca-gpt4-data-zh',
-        preprocess_func=AlpacaPreprocessor(concat_inst_input=_concat_inst_inp_alpaca_zh),
+        preprocess_func=AlpacaZhPreprocessor(),
         tags=['chat', 'general', '🔥'],
     ))
 

swift/llm/dataset/preprocessor/core.py

@@ -312,34 +312,22 @@ def preprocess(self, row: Dict[str, Any]) -> Optional[Dict[str, Any]]:
 
 class AlpacaPreprocessor(ResponsePreprocessor):
 
-    def __init__(self,
-                 *,
-                 concat_inst_input: Union[Callable[[str, str], str]] = '\n',
-                 columns_mapping: Optional[Dict[str, str]] = None,
-                 **kwargs) -> None:
-        """Alpaca format preprocessor
-
-        Args:
-            concat_inst_input: The concat sep between instruction and input
-        """
-        super().__init__(columns_mapping=columns_mapping, **kwargs)
-        self.concat_inst_input = concat_inst_input
+    @classmethod
+    def concat_inst_input(cls, instruction, input_):
+        if instruction and input_:
+            query = f'{instruction}\n{input_}'
+        else:
+            query = instruction or input_
+        assert isinstance(query, str), f'query: {query}'
+        return query
 
     def preprocess(self, row: Dict[str, Any]) -> Optional[Dict[str, Any]]:
         instruction = row.pop('instruction', None)
         input_ = row.pop('input', None)
         output = row.pop('output', None)
         if output is not None:
             row['response'] = output
-
-        if instruction is not None or input_ is not None:
-            instruction = instruction or ''
-            input_ = input_ or ''
-            if isinstance(self.concat_inst_input, str):
-                query = instruction + self.concat_inst_input + input_
-            else:
-                query = self.concat_inst_input(instruction, input_)
-            row['query'] = query
+        row['query'] = self.concat_inst_input(instruction, input_)
         return super().preprocess(row)
 
 

swift/llm/model/__init__.py

@@ -6,4 +6,4 @@
                        get_default_torch_dtype, get_model_info_meta, get_model_name, get_model_tokenizer,
                        get_model_tokenizer_multimodal, get_model_tokenizer_with_flash_attn, get_model_with_value_head,
                        load_by_unsloth, register_model)
-from .utils import HfConfigFactory, ModelInfo, safe_snapshot_download
+from .utils import HfConfigFactory, ModelInfo, git_clone_github, safe_snapshot_download

swift/llm/model/utils.py

@@ -274,6 +274,8 @@ def git_clone_github(github_url: str,
                      local_repo_name: Optional[str] = None,
                      branch: Optional[str] = None,
                      commit_hash: Optional[str] = None) -> str:
+    if github_url.endswith('.git'):
+        github_url = github_url[:-4]
     git_cache_dir = os.path.join(get_cache_dir(), '_github')
     os.makedirs(git_cache_dir, exist_ok=True)
     if local_repo_name is None:
@@ -282,8 +284,7 @@ def git_clone_github(github_url: str,
     local_repo_path = os.path.join(git_cache_dir, local_repo_name)
     with safe_ddp_context(hash_id=local_repo_path):
         if not os.path.exists(local_repo_path):
-            if not github_url.endswith('.git'):
-                github_url = f'{github_url}.git'
+            github_url = f'{github_url}.git'
             command = ['git', '-C', git_cache_dir, 'clone', github_url, local_repo_name]
             command_str = f"git -C '{git_cache_dir}' clone '{github_url}' {local_repo_name}"
             if branch is not None:

tests/general/test_dataset.py

@@ -15,8 +15,11 @@ def test_sft():
     # _test_dataset(['AI-ModelScope/Duet-v0.5'])
     # _test_dataset(['swift/SlimOrca', 'swift/cosmopedia-100k'])
     # _test_dataset(['OmniData/Zhihu-KOL-More-Than-100-Upvotes'])
-    _test_dataset(['OmniData/Zhihu-KOL'])
-    # _test_dataset(['AI-ModelScope/alpaca-gpt4-data-zh#1000', 'AI-ModelScope/alpaca-gpt4-data-en#200'])
+    # _test_dataset(['OmniData/Zhihu-KOL'])
+    _test_dataset([
+        'AI-ModelScope/alpaca-gpt4-data-zh#1000', 'AI-ModelScope/alpaca-gpt4-data-en#1000',
+        'AI-ModelScope/LongAlpaca-12k#1000'
+    ])
     # _test_dataset(['swift/Infinity-Instruct:all'])
     # _test_dataset(['swift/sharegpt:all'])
     # _test_dataset(['AI-ModelScope/sharegpt_gpt4:all'])

tests/general/test_stream.py

@@ -3,11 +3,10 @@
 
 def test_local_dataset():
     # please use git clone
-    local_dataset = '/mnt/nas2/huangjintao.hjt/work/datasets/swift-sft-mixture:firefly#100'
-    dataset = load_dataset(datasets=[local_dataset], streaming=True)[0]
-    for i, x in enumerate(dataset):
-        pass
-    print(i, x)
+    from swift.llm import git_clone_github
+    model_dir = git_clone_github('https://www.modelscope.cn/datasets/swift/swift-sft-mixture.git')
+    dataset = load_dataset(datasets=[f'{model_dir}:firefly'], streaming=True)[0]
+    print(next(iter(dataset)))
 
 
 def test_hub_dataset():