chore: update dependencies and configuration

- Updated Python version requirement in `pyproject.toml` to 3.12.
- Adjusted dependency specifications for `llamafactory`, `torch`, `torchvision`, and `torchaudio` to include version constraints.

.cursor/rules/weclone-rules.mdc

@@ -14,4 +14,3 @@ alwaysApply: true
 - The project uses uv as the package manager and pyproject.toml as the project configuration file.
 - You should write as few code comments as possible.
 - Prefer using the encapsulated logger `from weclone.utils.log import logger` for printing.
-- Comments should prioritize the use of English.

.gitignore

@@ -196,3 +196,4 @@ settings-bot8006.jsonc
 models_final/*
 /data/*
 /llamaboard_cache
+eval_Result/*

pyproject.toml

@@ -4,7 +4,7 @@ version = "0.3.03"
 description = "One-stop solution for creating your digital avatar from chat history"
 authors = [{ name = "xming521" }]
 readme = "README.md"
-requires-python = ">=3.10,<3.11"
+requires-python = ">=3.12,<3.13"
 
 dependencies = [
   "pandas",
@@ -15,7 +15,6 @@ dependencies = [
   "pydantic==2.10.6",
   "setuptools>=78.1.0",
   "loguru>=0.7.3",
-  "tomli; python_version < '3.11'",
   "langchain",
   "openai==1.87.0",
   "pip"
@@ -33,10 +32,12 @@ config_changelog = """
 
 [dependency-groups]
 main = [
-  "llamafactory @ git+https://github.com/hiyouga/LLaMA-Factory.git",
+  "llamafactory==0.9.4",
   "vllm==0.10.0; platform_system == 'Linux'",
-  "torch==2.7.1",
-  "torchvision==0.22.1",
+  "torch==2.7.1+cu126; platform_system == 'Linux' or platform_system == 'Windows'",
+  "torchvision==0.22.1+cu126; platform_system == 'Linux' or platform_system == 'Windows'",
+  "torchaudio==2.7.1+cu126; platform_system == 'Linux' or platform_system == 'Windows'",
+  "torchdata>=0.10.0; platform_system == 'Linux' or platform_system == 'Windows'",
   "transformers==4.53.2",
   "accelerate==1.7.0",
   "triton==3.3.1; platform_system == 'Linux'",
@@ -83,7 +84,10 @@ triton = [
   { index = "pytorch-cu126", marker = "platform_system == 'Windows'" },
   { index = "pytorch-cu126", marker = "platform_system == 'Linux'" },
 ]
-
+torchdata = [
+  { index = "pytorch-cu126", marker = "platform_system == 'Windows'" },
+  { index = "pytorch-cu126", marker = "platform_system == 'Linux'" },
+]
 
 [[tool.uv.index]]
 name = "pytorch-cu126"
@@ -105,7 +109,7 @@ ignore = ["**/archive"]
 reportMissingImports = "error"
 reportMissingTypeStubs = false
 
-pythonVersion = "3.10"
+pythonVersion = "3.12"
 pythonPlatform = "Linux"
 
 [tool.ruff]
@@ -126,7 +130,7 @@ lint.select = [
   "C4",    # flake8-comprehensions
   "Q",     # flake8-quotes
 ]
-target-version = "py310"
+target-version = "py312"
 
 [tool.pytest.ini_options]
 addopts = "-x -v -s --tb=short"

tests/test_full_pipe.py

@@ -260,13 +260,11 @@ def run_webchat_demo_test(config_file: str):
     """执行 webchat-demo 测试"""
     print_test_header("webchat-demo", config_file)
     
-    with mock.patch("weclone.eval.web_demo.main") as mock_main:
-        mock_main.return_value = None
-        try:
-            result = run_cli_command(["webchat-demo"], config_file, timeout=20)
-            assert result.returncode == 0, f"webchat-demo command execution failed for config {config_file}"
-        except subprocess.TimeoutExpired:
-            pass
+    try:
+        result = run_cli_command(["webchat-demo"], config_file, timeout=20)
+        assert result.returncode == 0, f"webchat-demo command execution failed for config {config_file}"
+    except subprocess.TimeoutExpired:
+        pass
 
 def run_server_test(config_file: str) -> subprocess.Popen:
     """执行 server 测试，返回进程对象"""

weclone/core/inference/online_infer.py

@@ -22,6 +22,8 @@ class OnlineLLM:
         model_name: str,
         default_system: Optional[str] = None,
         max_workers: int = 10,
+        prompt_with_system: bool = False,
+        response_format: str = "json_object",
     ):
         self.api_key = api_key
         self.base_url = base_url
@@ -30,6 +32,8 @@ class OnlineLLM:
         self.max_workers = max_workers
         self.client = OpenAI(api_key=self.api_key, base_url=self.base_url, max_retries=0)
         self.executor = ThreadPoolExecutor(max_workers=max_workers)
+        self.prompt_with_system = prompt_with_system
+        self.response_format = response_format
 
     @retry_openai_api(max_retries=200, base_delay=30.0, max_delay=180.0)
     def chat(
@@ -40,20 +44,29 @@ class OnlineLLM:
         top_p: float = 0.95,
         stream: bool = False,
     ):
-        messages: List[ChatCompletionMessageParam] = [
-            # {"role": "system", "content": self.default_system},
-            {"role": "user", "content": prompt_text},
-        ]
-        response = self.client.chat.completions.create(
-            model=self.model_name,
-            messages=messages,
-            stream=stream,
-            temperature=temperature,
-            max_tokens=max_tokens,
-            top_p=top_p,
-            response_format={"type": "json_object"},
+        messages: List[ChatCompletionMessageParam] = []
+        if self.prompt_with_system:
+            messages = prompt_text
+        else:
+            messages = [
+                # {"role": "system", "content": self.default_system},
+                {"role": "user", "content": prompt_text},
+            ]
+
+        params = {
+            "model": self.model_name,
+            "messages": messages,
+            "stream": stream,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+            "top_p": top_p,
             # extra_body={"chat_template_kwargs": {"enable_thinking": False}}
-        )
+        }
+
+        if self.response_format:
+            params["response_format"] = {"type": self.response_format}
+
+        response = self.client.chat.completions.create(**params)
 
         return response