feat: change to use dataclass for type safety

mllm/chat.py

@@ -236,16 +236,16 @@ def complete(self, model=None, cache=False, expensive=False, parse=None, retry=T
         """
         if options is None:
             options = {}
-        options = {**default_options, **options}
+        options = {**default_options.get_dict(), **options}
 
         contains_image = self.contains_image()
         if model is None:
             if not expensive:
-                model = default_models["normal"]
+                model = default_models.normal
             else:
-                model = default_models["expensive"]
+                model = default_models.expensive
             if contains_image:
-                model = default_models["vision"]
+                model = default_models.vision
 
         if get_llm_provider(model)[1] in ["openai"]:
             if parse == "dict":

mllm/config.py

@@ -1,22 +1,46 @@
+from dataclasses import dataclass
 
-default_models = {
-    "normal": "gpt-3.5-turbo",
-    "expensive": "gpt-4-turbo-preview",
-    "vision": "gpt-4-vision-preview",
-    "embedding": "text-embedding-3-large"
-}
 
+@dataclass
+class DefaultModels:
+    normal: str = "gpt-4o-mini"
+    expensive: str = "gpt-4-turbo-preview"
+    vision: str = "gpt-4o"
+    embedding: str = "text-embedding-3-large"
 
-default_options = {
-    "temperature": 0.2,
-    "max_tokens": None,
-    "frequency_penalty": None,
-    "timeout": 600,
-    "seed": None,
-}
+    def __getitem__(self, item):
+        return getattr(self, item)
 
+    def __setitem__(self, key, value):
+        setattr(self, key, value)
 
-class CacheOptions:
-    def __init__(self):
-        self.max_embedding_vecs = 1000
-        self.max_chat = 1000
+    def get_dict(self):
+        return {k: v for k, v in self.__dict__.items() if v is not None}
+
+
+@dataclass
+class DefaultOptions:
+    temperature: float = 0.2
+    max_tokens: int = None
+    frequency_penalty: float = None
+    timeout: int = 600
+    seed: int = None
+
+    def __getitem__(self, item):
+        return getattr(self, item)
+
+    def __setitem__(self, key, value):
+        setattr(self, key, value)
+
+    def get_dict(self):
+        return {k: v for k, v in self.__dict__.items() if v is not None}
+
+
+default_models = DefaultModels()
+
+default_options = DefaultOptions()
+
+from mllm.utils.parser import parse_options
+
+if __name__ == '__main__':
+    print(default_options.get_dict())

mllm/utils/parser.py

@@ -1,14 +1,19 @@
 import ast
 import json
+from dataclasses import dataclass
 
 """
 # Parse
 """
 
-parse_options = {
-    "cheap_model": "gpt-4o-mini",
-    "correct_json_by_model": False,
-}
+
+@dataclass
+class ParseOptions:
+    cheap_model: str = "gpt-4o-mini"
+    correct_json_by_model: bool = False
+
+
+parse_options = ParseOptions()
 
 
 class Parse:
@@ -50,7 +55,7 @@ def dict(src: str):
         except:
             pass
 
-        if parse_options["correct_json_by_model"]:
+        if parse_options.correct_json_by_model:
             try:
                 res = parse_json_by_cheap_model(json_src)
                 return res
@@ -87,7 +92,7 @@ def quotes(src: str):
                 break
         else:
             raise ValueError("No closing quotes")
-        for end in range(start+1, len(split_lines)):
+        for end in range(start + 1, len(split_lines)):
             if split_lines[end].startswith("```"):
                 break
         else:
@@ -125,5 +130,5 @@ def parse_json_by_cheap_model(json_src):
 You should directly output the corrected JSON dict with a minimal modification.
 """
     chat = Chat(prompt)
-    res = chat.complete(parse="dict", model=parse_options["cheap_model"])
+    res = chat.complete(parse="dict", model=parse_options.cheap_model)
     return res

test/test_parse.py

@@ -37,7 +37,11 @@ def test_dict_gen():
     res = chat.complete(parse="dict", cache=False)
     assert res == {"a": 1, "b": 2}
 
+
 def test_model_correct():
+    # this import is necessary to test whether it's imported into mllm.config
+    from mllm.config import parse_options
+
     raw_json = """
 {
 no_quote : "