ci: add Ruff config, pre-commit hooks, and GitHub Actions lint workflow

.github/workflows/lint.yml

@@ -0,0 +1,17 @@
+name: Lint
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: astral-sh/setup-uv@v5
+      - run: uv python install 3.10
+      - run: uv pip install ruff
+      - run: uv run ruff check .
+      - run: uv run ruff format --check .

.pre-commit-config.yaml

@@ -0,0 +1,7 @@
+repos:
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.12.2
+    hooks:
+      - id: ruff
+        args: [--fix]
+      - id: ruff-format

data_preprocess/aime2024_rstar2_agent_loop.py

@@ -9,10 +9,8 @@
 import os
 
 import datasets
-
 from verl.utils.hdfs_io import copy, makedirs
 
-
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--local_dir", default="~/data/rstar2-agent/aime2024")

data_preprocess/aime2025_rstar2_agent_loop.py

@@ -9,10 +9,8 @@
 import os
 
 import datasets
-
 from verl.utils.hdfs_io import copy, makedirs
 
-
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--local_dir", default="~/data/rstar2-agent/aime2025")

data_preprocess/dapo_rstar2_agent_loop.py

@@ -9,10 +9,8 @@
 import os
 
 import datasets
-
 from verl.utils.hdfs_io import copy, makedirs
 
-
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--local_dir", default="~/data/rstar2-agent/dapo-math-17k-en")

data_preprocess/math500_rstar2_agent_loop.py

@@ -9,10 +9,8 @@
 import os
 
 import datasets
-
 from verl.utils.hdfs_io import copy, makedirs
 
-
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--local_dir", default="~/data/rstar2-agent/math500")

examples/chat_with_tool_call.py

@@ -14,22 +14,21 @@
 
 curl http://localhost:8000/v1/models
 """
-import aiohttp
 import argparse
 import asyncio
 import json
-import requests
-import yaml
-
 from pathlib import Path
 
+import aiohttp
+import requests
+import yaml
 from transformers import AutoTokenizer, PreTrainedTokenizer
 from verl.tools.schemas import ToolResponse
 
 from rstar2_agent.tools.code_judge_utils import (
-    run_tool_calls_on_server_async,
     generate_tool_call_code,
     generate_tool_call_input,
+    run_tool_calls_on_server_async,
 )
 from rstar2_agent.tools.tool_parser import (
     RStar2AgentHermesToolParser,

fused_compute_score/__init__.py

@@ -1,5 +1,6 @@
-from .prime_math import compute_score as prime_compute_score
 from .math_verify import compute_score as math_verify_compute_score
+from .prime_math import compute_score as prime_compute_score
+
 
 def compute_score(model_output: str, ground_truth: str) -> bool:
     try:

fused_compute_score/prime_math/grader.py

@@ -381,12 +381,12 @@ def format_intervals(prediction):
     return prediction
 
 
+import multiprocessing
 import os
-import signal
 import queue
-import multiprocessing
+import signal
 from functools import wraps
-from typing import Callable, Any
+from typing import Any, Callable
 
 
 def _mp_target_wrapper(target_func: Callable, mp_queue: multiprocessing.Queue, args: tuple, kwargs: dict[str, Any]):

pyproject.toml

@@ -15,3 +15,16 @@ dependencies = [
 
 [tool.setuptools]
 packages = ["rstar2_agent", "fused_compute_score"]
+
+[tool.ruff]
+line-length = 120
+
+[tool.ruff.lint]
+select = ["E", "F", "I"]
+ignore = ["E501", "E402", "E731", "E722", "E741", "E721", "E701", "F841", "F403", "F405"]
+
+[tool.ruff.lint.per-file-ignores]
+"__init__.py" = ["F401"]
+
+[tool.uv]
+# Install with: uv sync

rstar2_agent/__init__.py

@@ -1,6 +1,6 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
+from .reward import CodeJudgeRewardManager
 from .rollout.rstar2_agent_loop import RStar2AgentLoop
 from .tools import RStar2AgentHermesToolParser
-from .reward import CodeJudgeRewardManager

rstar2_agent/down_sample/reject_sampling.py

@@ -3,8 +3,8 @@
 
 import numpy as np
 import torch
-
 from verl.protocol import DataProto
+
 from .utils import filter_by_mask
 
 

rstar2_agent/down_sample/roc.py

@@ -2,14 +2,15 @@
 # Licensed under the MIT license.
 
 import re
-import numpy as np
-import torch
 from pprint import pprint
 from typing import List
-from transformers import PreTrainedTokenizerFast
 
+import numpy as np
+import torch
+from transformers import PreTrainedTokenizerFast
 from verl.protocol import DataProto
-from .utils import filter_by_mask, decode_prompt_response_str
+
+from .utils import decode_prompt_response_str, filter_by_mask
 
 
 def resample_of_correct(batch: DataProto, tokenizer: PreTrainedTokenizerFast, config: dict, do_sample=True, world_size=None):

rstar2_agent/down_sample/utils.py

@@ -2,7 +2,7 @@
 # Licensed under the MIT license.
 
 import torch
-from verl.protocol import DataProto, DataProtoItem
+from verl.protocol import DataProto
 
 
 def filter_by_mask(batch: DataProto, mask: torch.Tensor, num_trainer_replicas: int) -> DataProto:

rstar2_agent/main_rstar2_agent.py

@@ -11,7 +11,6 @@
 import hydra
 import ray
 from omegaconf import OmegaConf
-
 from verl.trainer.constants_ppo import get_ppo_ray_runtime_env
 from verl.trainer.main_ppo import create_rl_dataset, create_rl_sampler
 from verl.trainer.ppo.reward import load_reward_manager
@@ -199,7 +198,6 @@ def run(self, config):
         from pprint import pprint
 
         from omegaconf import OmegaConf
-
         from verl.utils.fs import copy_to_local
 
         print(f"TaskRunner hostname: {socket.gethostname()}, PID: {os.getpid()}")
@@ -244,7 +242,7 @@ def run(self, config):
         tool_config_path = config.actor_rollout_ref.rollout.multi_turn.tool_config_path
         tool_list = []
         if tool_config_path is not None:
-            from verl.tools.utils.tool_registry import ToolType, get_tool_class, OpenAIFunctionToolSchema
+            from verl.tools.utils.tool_registry import OpenAIFunctionToolSchema, ToolType
             tools_config = OmegaConf.load(tool_config_path)
             for tool_config in tools_config.tools:
                 tool_type = ToolType(tool_config.config.type)

rstar2_agent/reward/compute_score.py

@@ -1,8 +1,9 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
-from verl.utils.reward_score.prime_math import compute_score as prime_compute_score
 from verl.utils.reward_score.math_verify import compute_score as math_verify_compute_score
+from verl.utils.reward_score.prime_math import compute_score as prime_compute_score
+
 
 def compute_score(model_output: str, ground_truth: str) -> bool:
     try:

rstar2_agent/reward/server.py

@@ -1,20 +1,19 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
+import asyncio
+import base64
+import re
 from collections import defaultdict
 from typing import Any
 
 import aiohttp
-import asyncio
-import base64
-import re
 import torch
-
-from verl import DataProto
 from verl.workers.reward_manager import register
 from verl.workers.reward_manager.abstract import AbstractRewardManager
 
 from rstar2_agent.tools.code_judge_utils import run_tool_calls_on_server_async
+from verl import DataProto
 
 verify_math_prefix = """
 from fused_compute_score import compute_score

rstar2_agent/rstar2_agent_ray_trainer.py

@@ -9,8 +9,6 @@
 import ray
 import torch
 from tqdm import tqdm
-
-from verl import DataProto
 from verl.experimental.dataset.sampler import AbstractCurriculumSampler
 from verl.trainer.ppo.core_algos import AdvantageEstimator, agg_loss
 from verl.trainer.ppo.metric_utils import (
@@ -30,6 +28,8 @@
 from verl.utils.metric import reduce_metrics
 from verl.utils.rollout_skip import RolloutSkip
 
+from verl import DataProto
+
 from .down_sample import reject_equal_reward, resample_of_correct
 
 
@@ -79,7 +79,6 @@ def fit(self):
         Most logic is same with RayPPOTrainer, mainly add down sample related.
         """
         from omegaconf import OmegaConf
-
         from verl.utils.tracking import Tracking
 
         logger = Tracking(

rstar2_agent/tools/__init__.py

@@ -1,5 +1,5 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
-from .code_judge_tool import CodeJudgeTool, SimJupyterTool, PythonTool
+from .code_judge_tool import CodeJudgeTool, PythonTool, SimJupyterTool
 from .tool_parser import RStar2AgentHermesToolParser

rstar2_agent/tools/code_judge_tool.py

@@ -1,19 +1,18 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
+import json
 from functools import partial
 from typing import Any, Optional
 from uuid import uuid4
 
 import aiohttp
-import json
-
-from verl.utils.rollout_trace import rollout_trace_op
 from verl.tools.base_tool import BaseTool
 from verl.tools.schemas import OpenAIFunctionToolSchema, ToolResponse
+from verl.utils.rollout_trace import rollout_trace_op
 
+from .code_judge_utils import generate_tool_call_code, generate_tool_call_input, run_tool_calls_on_server_async
 from .request_processor import RequestProcessor
-from .code_judge_utils import run_tool_calls_on_server_async, generate_tool_call_code, generate_tool_call_input
 
 
 class CodeJudgeTool(BaseTool):

rstar2_agent/tools/code_judge_utils.py

@@ -1,14 +1,14 @@
 # Copyright (c) Microsoft Corporation.
 # Licensed under the MIT license.
 
-import json
-import aiohttp
 import asyncio
-import traceback
-import os
 import datetime
+import json
+import os
+import traceback
+from typing import Callable, Dict, List, Literal, Optional
 
-from typing import Dict, List, Literal, Callable, Optional
+import aiohttp
 
 # Global variable to store the path for failed submissions
 _failed_submissions_path = os.path.expanduser("~")

rstar2_agent/tools/request_processor.py

@@ -2,12 +2,13 @@
 # Licensed under the MIT license.
 
 import asyncio
+import collections
 import time
+import traceback
 import uuid
-import collections
+from typing import Any, Awaitable, Callable, Dict, List
+
 import aiohttp
-import traceback
-from typing import List, Dict, Any, Callable, Awaitable
 
 # Define the expected signature for the batch submission function
 # It should be an async callable that takes:

rstar2_agent/tools/tool_parser.py

@@ -7,8 +7,7 @@
 import os
 
 import regex as re
-
-from verl.experimental.agent_loop.tool_parser import ToolParser, FunctionCall
+from verl.experimental.agent_loop.tool_parser import FunctionCall, ToolParser
 from verl.utils.rollout_trace import rollout_trace_op
 
 logger = logging.getLogger(__file__)