Обновил структуру проекта

llm/credentials.py

@@ -10,12 +10,12 @@ from pydantic_ai.providers.openai_codex import (
 
 class FileCodexCredentionalSource(OpenAICodexCredentialSource):
     def __init__(self, 
-                 path: Path,
+                 path: str| Path,
                  encoding: str="utf-8",
                  ensure_ascii: bool=False,
                  indent: int=2):
         super().__init__()
-        self._path = path
+        self._path = Path(path)
         self._encoding=encoding
         self._ensure_ascii=ensure_ascii
         self._indent=indent
@@ -32,6 +32,8 @@ class FileCodexCredentionalSource(OpenAICodexCredentialSource):
             )
 
     async def save(self, credentials: OpenAICodexCredentials) -> None:
+        self.path.parent.mkdir(parents=True , exist_ok=True)
+
         self._path.write_text(
             json.dumps(
                 asdict(credentials),

llm/factory.py

@@ -1,5 +1,6 @@
 import os
 import webbrowser
+import pandas as pd
 from pathlib import Path
 
 from pydantic_ai.models.openai import OpenAIChatModel
@@ -12,6 +13,7 @@ from pydantic_ai.providers.openai_codex import (
 
 from .config import CodexConfig, LLMConfig, OpenAiConfig
 from .credentials import FileCodexCredentionalSource
+from .schemas import BaseDatasetRecord, ChatRecord, ChatMessage
 
 
 class LLMModelProvider:
@@ -81,4 +83,69 @@ class LLMModelProvider:
                 base_url=base_url
             )
 
-        raise ValueError(f"Unsupported LLM config: {type(config)}")
+        raise ValueError(f"Unsupported LLM config: {type(config)}")
+
+class DatasetWriter:
+    def __init__(self,
+                 dataset_record: BaseDatasetRecord,
+                 base_path: str | Path):
+
+
+        self._dataset_record = dataset_record    
+        self._path = Path(base_path)
+        self._path.mkdir(parents=True, exist_ok=True)
+
+    def __check_file_exists(self, filename :str)-> bool:
+        file_path = self._path / filename
+        return file_path.exists()
+
+    def to_csv(self,
+               filename: str) -> None:
+
+        exists=self.__check_file_exists(filename)
+        
+        df = pd.json_normalize(
+            self._dataset_record.model_dump()
+        )
+
+        df.to_csv(self._path / filename,
+                  mode="a" if exists else "w",
+                  header=not exists,
+                  index=False,
+                  encoding="utf-8")
+
+    def to_chat_record(self) -> ChatRecord:
+        response = self._dataset_record.response
+        prompts = self._dataset_record.prompts
+
+        assistant_content = (
+            response.assistant_content
+            or response.model_response
+        )
+
+        return ChatRecord(
+            messages=[
+                ChatMessage(
+                    role="system",
+                    content=prompts.system_prompt,
+                ),
+                ChatMessage(
+                    role="user",
+                    content=prompts.user_prompt,
+                ),
+                ChatMessage(
+                    role="assistant",
+                    content=assistant_content,
+                ),
+            ]
+        )
+
+    def to_jsonl(self,
+                 filename: str) -> None:
+        
+        chat_record = self.to_chat_record()
+        path = self._path/filename
+
+        with path.open("a", encoding='utf-8') as file:
+            file.write(chat_record.model_dump_json())
+            file.write("\n")

llm/schemas.py

@@ -1,4 +1,4 @@
-from typing import Literal
+from typing import Literal, Optional
 
 from pydantic import BaseModel, Field
 
@@ -10,7 +10,7 @@ class RobotPlan(BaseModel):
     )
 
 
-class Message(BaseModel):
+class ChatMessage(BaseModel):
     role: Literal["system",
                   "user",
                   "assistant"]
@@ -18,22 +18,32 @@ class Message(BaseModel):
     content: str
 
 
-class SFTFormat(BaseModel):
-    message: list[Message]
-
-
-class DatasetRecord(BaseModel):
-    task: str
+class ChatRecord(BaseModel):
+    messages: list[ChatMessage]
 
+class Prompts(BaseModel):
     system_prompt: str
     instructions: str
     user_prompt: str
 
-    model_response: str
-    assistant_content: str
 
+class ModelResponse(BaseModel):
+    model_response: str
+    assistant_content: Optional[str] = None
+
+class ModelInfo(BaseModel):
     provider: str
     model: str
 
+class GenerationSettings(BaseModel):
     temperature: float
-    max_tokens: int
+    max_tokens: int
+
+class BaseDatasetRecord(BaseModel):
+    prompts: Prompts
+    response: ModelResponse
+    model_info: ModelInfo
+    generation: GenerationSettings
+
+class RobotDatasetRecord(BaseDatasetRecord):
+    task: str