add code for reading from multi files in one directory

data/dataset_info.json

@@ -103,5 +103,14 @@
       "response": "",
       "history": ""
     }
+  },
+  "pretrain_data": {
+    "file_name": "pretrain_data",
+    "columns": {
+      "prompt": "content",
+      "query": "",
+      "response": "",
+      "history": ""
+    }
   }
 }

data/pretrain_data/blog.json

@@ -0,0 +1,7 @@
+[
+  {
+    "id": 0,
+    "title": "拥有自己的航空器",
+    "content": "想自己驾驶飞机或从事通用航空事业的人，大都想拥有自己的航空器。\"自己的\"意思包括自己购买、自己制造、可供自己使用（租用或借用）等等。\n花自己的钱买一架飞机来开一开，国内有些人或企业已实现了这个愿望。现在一架国产超轻型的“蜜蜂”飞机售价在l0万元以下，进口的一些单发的双座飞机售价在100万元之内。据估计，全国大约有几十万人具有这种购买能力。\n自己造一架飞机来开也是一个好创意。美国的通航飞机中有l／5是自制的。有的自制飞机甚至还创造了世界飞行纪录。今天自己造飞机比当年莱特兄弟容易多了。飞机的基本构造已无秘密可言，各种飞机部件和材料都不难买到。尤其主要的是，技术进步大大改进了配件的性能，与此同时，配件的重量也下降了很多。莱特兄弟当年使用的12马力汽油发动机比现在30马力的同类产品还重。如果有人有志于此而且具备造飞机的种种条件，应该说这个目标也是可以实现的。有两点值得注意，一是在莱特兄弟造飞机时没有前人经验，全靠自己摸索。现在不同了，航空制造已有了上百年的知识和经验可供后人学习和利用。现在如果谁想自己造飞机就不用闭门造车了。制造者本人首先应该去学习和掌握一些必要知识和经验才行。其次，在莱特兄弟时代，没有国家民航当局，他们的航空活动不受法规约束。今天就不一样了，所有要升空的航空器必须先接受民航当局的鉴定，以保证飞行安全。绝不允许以生命为赌注的任何冒险行为。\n租用飞机也是实现自驾飞机的方式之一。国内也还有另一种形式，即参加飞行驾驶学校接受培训，当然所交的学费价格是不菲的。预计未来在我国必将出现出各类飞行俱乐部。到那时，飞行爱好者可以租用飞机去上天过一把瘾了。"
+  }
+]

data/pretrain_data/wiki.json

@@ -0,0 +1,7 @@
+[
+  {
+    "id": 0,
+    "title": "大卫·亨利",
+    "content": "大卫·亨利\n\n大卫·克莱顿·亨利（David Clayton Henrie，），美国演员。近来在迪士尼频道原创电视影集《少年魔法师》（Wizards of Waverly Place）当中演出贾斯汀·鲁索（Justin Russo）一角。\n\n大卫·亨利出生在加州Mission Viejo，在凤凰城长大。他的胞弟劳伦斯·亨利（Lorenzo Henrie）也是演员。大卫·亨利就读夏安传统学校。家中是信奉罗马天主教。 \n\n大卫在2007年拍摄少年魔法师期间认识女演员露西·海尔（Lucy Hale），之后与其交往，于2009年分手。\n\n10岁时，大卫·亨利和SAG在凤凰城签订了合约，并开始走出去试镜。 9岁的时候，在沙加缅度进行商业拍摄，SAG董事建议大卫·亨利搬到洛杉矶。在10岁那年夏天，他和他的家人搬到了好莱坞。他预定他的前2支商业试镜，扮演主要角色为汉堡王和桂格燕麦。他初演电视节目为Providence。 \n\n到了13岁，大卫有了他的第一次重大突破，在福克斯公司的喜剧The Pitts饰演 Petey Pitt一角。大卫下出作品为的Hallmark movie为Monster Maker，和琳达布莱儿、乔治甘迺迪共同演出，并要求回来Hallmark movie公司。 \n\n在18岁时，大卫得到了迪士尼频道原创系列演出机会，该节目2007年10月12日首播。大卫2008年参加了迪士尼频道的游戏节目。他是绿色团队的队长，隔年，为旋风队队长。他在迪士尼原创电影《少年魔法师》之后在《酷爸的疯狂假期》中有饰演一角。\n"
+  }
+]

src/utils/common.py

@@ -56,7 +56,6 @@ require_version("accelerate>=0.19.0", "To fix: pip install accelerate>=0.19.0")
 require_version("peft>=0.3.0", "To fix: pip install peft>=0.3.0")
 require_version("trl>=0.4.1", "To fix: pip install trl>=0.4.1")
 
-
 logger = get_logger(__name__)
 
 
@@ -92,10 +91,12 @@ def _init_adapter(
 
     if model_args.checkpoint_dir is not None:
         if finetuning_args.finetuning_type != "lora":
-            assert is_mergeable and len(model_args.checkpoint_dir) == 1, "Only LoRA tuning accepts multiple checkpoints."
+            assert is_mergeable and len(
+                model_args.checkpoint_dir) == 1, "Only LoRA tuning accepts multiple checkpoints."
             load_trainable_params(model, model_args.checkpoint_dir[0])  # load model checkpoints for non-peft methods
         else:
-            assert is_mergeable or len(model_args.checkpoint_dir) == 1, "Quantized model only accepts a single checkpoint."
+            assert is_mergeable or len(
+                model_args.checkpoint_dir) == 1, "Quantized model only accepts a single checkpoint."
 
     if finetuning_args.finetuning_type == "lora":
         logger.info("Fine-tuning method: LoRA")
@@ -105,7 +106,8 @@ def _init_adapter(
             assert os.path.exists(os.path.join(model_args.checkpoint_dir[0], CONFIG_NAME)), \
                 "The given checkpoint is not a LoRA checkpoint, please specify `--finetuning_type full/freeze` instead."
 
-            if (is_trainable and model_args.resume_lora_training) or (not is_mergeable): # continually train on the lora weights
+            if (is_trainable and model_args.resume_lora_training) or (
+                    not is_mergeable):  # continually train on the lora weights
                 checkpoints_to_merge, lastest_checkpoint = model_args.checkpoint_dir[:-1], model_args.checkpoint_dir[-1]
             else:
                 checkpoints_to_merge = model_args.checkpoint_dir
@@ -184,9 +186,11 @@ def load_pretrained(
             )
         elif model_args.quantization_bit == 4:
             require_version("bitsandbytes>=0.39.0", "To fix: pip install bitsandbytes>=0.39.0")
-            require_version("transformers>=4.30.0.dev0", "To fix: pip install git+https://github.com/huggingface/transformers.git")
+            require_version("transformers>=4.30.0.dev0",
+                            "To fix: pip install git+https://github.com/huggingface/transformers.git")
             require_version("peft>=0.4.0.dev0", "To fix: pip install git+https://github.com/huggingface/peft.git")
-            require_version("accelerate>=0.20.0.dev0", "To fix: pip install git+https://github.com/huggingface/accelerate.git")
+            require_version("accelerate>=0.20.0.dev0",
+                            "To fix: pip install git+https://github.com/huggingface/accelerate.git")
             config_kwargs["load_in_4bit"] = True
             config_kwargs["quantization_config"] = BitsAndBytesConfig(
                 load_in_4bit=True,
@@ -241,11 +245,11 @@ def load_pretrained(
 def prepare_args(
         stage: Literal["pt", "sft", "rm", "ppo"]
 ) -> Tuple[ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments, FinetuningArguments]:
-
     parser = HfArgumentParser((ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments, FinetuningArguments))
 
     if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):  # Provide arguments with a json file.
-        model_args, data_args, training_args, finetuning_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+        model_args, data_args, training_args, finetuning_args = parser.parse_json_file(
+            json_file=os.path.abspath(sys.argv[1]))
     else:
         model_args, data_args, training_args, finetuning_args = parser.parse_args_into_dataclasses()
 
@@ -310,7 +314,6 @@ def prepare_args(
 
 
 def prepare_infer_args() -> Tuple[ModelArguments, DataTrainingArguments, FinetuningArguments]:
-
     parser = HfArgumentParser((ModelArguments, DataTrainingArguments, FinetuningArguments))
 
     if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):  # Provide arguments with a json file.
@@ -331,7 +334,6 @@ def prepare_data(
         model_args: ModelArguments,
         data_args: DataTrainingArguments
 ) -> Dataset:
-
     def checksum(file_path, hash):
         with open(file_path, "rb") as datafile:
             binary_data = datafile.read()
@@ -361,7 +363,7 @@ def prepare_data(
                 checksum(data_file, dataset_attr.file_sha1)
             else:
                 logger.warning("Checksum failed: missing SHA-1 hash value in dataset_info.json.")
-
+            print(extension)
             raw_datasets = load_dataset(
                 extension if extension in ["csv", "json"] else "text",
                 data_files=data_file,
@@ -406,7 +408,6 @@ def preprocess_data(
         training_args: Seq2SeqTrainingArguments,
         stage: Literal["pt", "sft", "rm", "ppo"]
 ) -> Dataset:
-
     column_names = list(dataset.column_names)
     prefix = data_args.source_prefix if data_args.source_prefix is not None else ""
     prompt_template = Template(data_args.prompt_template)
@@ -429,7 +430,8 @@ def preprocess_data(
         # we drop the small remainder, and if the total_length < block_size, we exclude this batch
         total_length = (total_length // data_args.max_source_length) * data_args.max_source_length
         # split by chunks of max_source_length
-        result = [concatenated_ids[i: i+data_args.max_source_length] for i in range(0, total_length, data_args.max_source_length)]
+        result = [concatenated_ids[i: i + data_args.max_source_length] for i in
+                  range(0, total_length, data_args.max_source_length)]
         return {
             "input_ids": result,
             "labels": result.copy()

src/utils/config.py

@@ -7,7 +7,6 @@ from dataclasses import asdict, dataclass, field
 
 @dataclass
 class DatasetAttr:
-
     load_from: str
     dataset_name: Optional[str] = None
     file_name: Optional[str] = None
@@ -68,7 +67,8 @@ class ModelArguments:
     )
     checkpoint_dir: Optional[str] = field(
         default=None,
-        metadata={"help": "Path to the directory(s) containing the delta model checkpoints as well as the configurations."}
+        metadata={
+            "help": "Path to the directory(s) containing the delta model checkpoints as well as the configurations."}
     )
     reward_model: Optional[str] = field(
         default=None,
@@ -76,7 +76,8 @@ class ModelArguments:
     )
     resume_lora_training: Optional[bool] = field(
         default=True,
-        metadata={"help": "Whether to resume training from the last LoRA weights or create new weights after merging them."}
+        metadata={
+            "help": "Whether to resume training from the last LoRA weights or create new weights after merging them."}
     )
     plot_loss: Optional[bool] = field(
         default=False,
@@ -155,24 +156,41 @@ class DataTrainingArguments:
         for name in dataset_names:
             if name not in dataset_info:
                 raise ValueError("Undefined dataset {} in dataset_info.json.".format(name))
-
+            dataset_attrs = []
+            dataset_attr = None
             if "hf_hub_url" in dataset_info[name]:
                 dataset_attr = DatasetAttr("hf_hub", dataset_name=dataset_info[name]["hf_hub_url"])
             elif "script_url" in dataset_info[name]:
                 dataset_attr = DatasetAttr("script", dataset_name=dataset_info[name]["script_url"])
-            else:
+            elif os.path.isfile(os.path.join(self.dataset_dir, dataset_info[name]["file_name"])):
                 dataset_attr = DatasetAttr(
                     "file",
                     file_name=dataset_info[name]["file_name"],
                     file_sha1=dataset_info[name]["file_sha1"] if "file_sha1" in dataset_info[name] else None
                 )
-
+            else:
+                # Support Directory
+                for file_name in os.listdir(os.path.join(self.dataset_dir, dataset_info[name]["file_name"])):
+                    path = os.path.join(dataset_info[name]["file_name"], file_name)
+                    dataset_attrs.append(DatasetAttr(
+                        "file",
+                        file_name=path,
+                        file_sha1=dataset_info[name]["file_sha1"] if "file_sha1" in dataset_info[name] else None
+                    ))
+            if dataset_attr is not None:
+                if "columns" in dataset_info[name]:
+                    dataset_attr.prompt_column = dataset_info[name]["columns"].get("prompt", None)
+                    dataset_attr.query_column = dataset_info[name]["columns"].get("query", None)
+                    dataset_attr.response_column = dataset_info[name]["columns"].get("response", None)
+                    dataset_attr.history_column = dataset_info[name]["columns"].get("history", None)
+                self.dataset_list.append(dataset_attr)
+            else:
+                for i, dataset_attr in enumerate(dataset_attrs):
                     if "columns" in dataset_info[name]:
                         dataset_attr.prompt_column = dataset_info[name]["columns"].get("prompt", None)
                         dataset_attr.query_column = dataset_info[name]["columns"].get("query", None)
                         dataset_attr.response_column = dataset_info[name]["columns"].get("response", None)
                         dataset_attr.history_column = dataset_info[name]["columns"].get("history", None)
-
                     self.dataset_list.append(dataset_attr)
 
 
@@ -216,14 +234,16 @@ class FinetuningArguments:
 
     def __post_init__(self):
         if isinstance(self.lora_target, str):
-            self.lora_target = [target.strip() for target in self.lora_target.split(",")] # support custom target modules of LoRA
+            self.lora_target = [target.strip() for target in
+                                self.lora_target.split(",")]  # support custom target modules of LoRA
 
         if self.num_layer_trainable > 0:  # fine-tuning the last n layers if num_layer_trainable > 0
-            trainable_layer_ids = [27-k for k in range(self.num_layer_trainable)]
+            trainable_layer_ids = [27 - k for k in range(self.num_layer_trainable)]
         else:  # fine-tuning the first n layers if num_layer_trainable < 0
             trainable_layer_ids = [k for k in range(-self.num_layer_trainable)]
 
-        self.trainable_layers = ["layers.{:d}.{}".format(idx, self.name_module_trainable) for idx in trainable_layer_ids]
+        self.trainable_layers = ["layers.{:d}.{}".format(idx, self.name_module_trainable) for idx in
+                                 trainable_layer_ids]
 
         assert self.finetuning_type in ["none", "freeze", "lora", "full"], "Invalid fine-tuning method."