kaggle-environments 1.22.6__py3-none-any.whl → 1.24.3__py3-none-any.whl

kaggle_environments/envs/werewolf/harness/litellm_models.yaml

@@ -0,0 +1,51 @@
+openrouter/deepseek/deepseek-chat-v3.1:
+  input_cost_per_token: 2e-7
+  output_cost_per_token: 8e-7
+openrouter/openai/gpt-4o-mini:
+  input_cost_per_token: 1.5e-7
+  output_cost_per_token: 6e-7
+openrouter/qwen/qwen3-235b-a22b-2507:
+  input_cost_per_token: 7.8e-8
+  output_cost_per_token: 3.12e-7
+openrouter/z-ai/glm-4.5:
+  input_cost_per_token: 2e-7
+  output_cost_per_token: 8e-7
+openrouter/openai/gpt-oss-120b:
+  input_cost_per_token: 7.2e-8
+  output_cost_per_token: 2.8e-7
+openrouter/openai/gpt-oss-20b:
+  input_cost_per_token: 4e-8
+  output_cost_per_token: 1.5e-7
+openrouter/qwen/qwen3-30b-a3b:
+  input_cost_per_token: 1e-7
+  output_cost_per_token: 3e-7
+openrouter/openai/gpt-5:
+  input_cost_per_token: 1.25e-6
+  output_cost_per_token: 1e-5
+openrouter/openai/gpt-4.1:
+  input_cost_per_token: 2e-6
+  output_cost_per_token: 8e-6
+openrouter/anthropic/claude-sonnet-4:
+  input_cost_per_token: 3e-6
+  output_cost_per_token: 1.5e-5
+openrouter/x-ai/grok-4:
+  input_cost_per_token: 3e-6
+  output_cost_per_token: 1.5e-5
+openrouter/google/gemini-2.5-flash-lite:
+  input_cost_per_token: 1e-7
+  output_cost_per_token: 4e-7
+openrouter/google/gemini-2.5-pro:
+  input_cost_per_token: 1.25e-6
+  output_cost_per_token: 1e-5
+openrouter/google/gemini-2.5-flash:
+  input_cost_per_token: 3e-7
+  output_cost_per_token: 2.5e-6
+vertex_ai/gemini-2.5-pro:
+  input_cost_per_token: 1.25e-6
+  output_cost_per_token: 1e-5
+vertex_ai/gemini-2.5-flash:
+  input_cost_per_token: 3e-7
+  output_cost_per_token: 2.5e-6
+vertex_ai/gemini-2.5-flash-lite:
+  input_cost_per_token: 1e-7
+  output_cost_per_token: 4e-7

kaggle_environments/envs/werewolf/harness/main.py

@@ -0,0 +1,54 @@
+"""Main file for the Game Arena submission."""
+
+import os
+import sys
+
+_AGENT_OBJECT = None
+_SETUP_COMPLETE = False
+
+
+def agent(observation, configuration):
+    """Kaggle agent for Game Arena."""
+    global _AGENT_OBJECT, _SETUP_COMPLETE
+
+    if not _SETUP_COMPLETE:
+        print("--- Performing one-time agent setup... ---")
+
+        # 1. Add the vendored 'lib' directory to Python's search path.
+        print("Updating system path with vendored libraries...")
+        script_dir = os.path.dirname(configuration["__raw_path__"])
+        lib_dir = os.path.join(script_dir, "lib")
+        if lib_dir not in sys.path:
+            sys.path.insert(0, lib_dir)
+        print(f"System path updated. First entry is now: {sys.path[0]}")
+
+        # 2. Now that the path is set, we can import our libraries.
+        # pylint: disable=g-import-not-at-top
+
+        from kaggle_environments.envs.werewolf.werewolf import AgentFactoryWrapper, LLM_SYSTEM_PROMPT
+        from kaggle_environments.envs.werewolf.harness.base import LLMWerewolfAgent
+
+        if "MODEL_NAME" not in os.environ:
+            raise ValueError("MODEL_NAME was not specified as an environment variable. Agent cannot be configured.")
+
+        if "MODEL_PROXY_KEY" not in os.environ:
+            raise ValueError("MODEL_PROXY_KEY was not specified as an environment variable. "
+                             "Model proxy cannot function correctly.")
+
+        if "MODEL_PROXY_URL" not in os.environ:
+            raise ValueError("MODEL_PROXY_URL was not injected. Agent cannot run.")
+
+        _AGENT_OBJECT = AgentFactoryWrapper(
+            agent_class=LLMWerewolfAgent,
+            model_name=os.environ["MODEL_NAME"],
+            system_prompt=LLM_SYSTEM_PROMPT,
+            litellm_model_proxy_kwargs={
+                "api_base": os.environ["MODEL_PROXY_URL"],
+                "api_key": os.environ["MODEL_PROXY_KEY"]
+            }
+        )
+
+        _SETUP_COMPLETE = True
+        print("--- Agent setup complete. ---")
+
+    return _AGENT_OBJECT(observation, configuration)

kaggle_environments/envs/werewolf/harness/test_base.py

@@ -0,0 +1,35 @@
+import json
+import os
+
+import litellm
+import pytest
+from dotenv import load_dotenv
+
+load_dotenv()
+
+
+@pytest.mark.skip("Require the key to run test.")
+def test_vertex_ai():
+    model = "vertex_ai/deepseek-ai/deepseek-r1-0528-maas"
+    file_path = os.environ["GOOGLE_APPLICATION_CREDENTIALS"]
+    with open(file_path, "r") as file:
+        vertex_credentials = json.load(file)
+
+    vertex_credentials_json = json.dumps(vertex_credentials)
+
+    response = litellm.completion(
+        model=model,
+        messages=[{"role": "user", "content": "hi"}],
+        temperature=0.7,
+        vertex_ai_project=os.environ["VERTEXAI_PROJECT"],
+        vertex_ai_location=os.environ["VERTEXAI_LOCATION"],
+        vertex_credentials=vertex_credentials_json,
+    )
+    print(response)
+
+
+@pytest.mark.skip("Require the key to run test.")
+def test_together():
+    model = "together_ai/deepseek-ai/DeepSeek-R1"
+    response = litellm.completion(model=model, messages=[{"role": "user", "content": "hi"}])
+    print(response)

kaggle_environments/envs/werewolf/runner.py

@@ -0,0 +1,146 @@
+import logging
+import os
+import random
+import subprocess
+import time
+from datetime import datetime
+
+from kaggle_environments import PROJECT_ROOT, make
+
+logger = logging.getLogger(__name__)
+
+
+class LogExecutionTime:
+    """
+    A context manager to log the execution time of a code block.
+    The elapsed time is stored in the `elapsed_time` attribute.
+
+    Example:
+        logger = logging.getLogger(__name__)
+        with LogExecutionTime(logger, "My Task") as timer:
+            # Code to be timed
+            time.sleep(1)
+        print(f"Task took {timer.elapsed_time:.2f} seconds.")
+        print(f"Formatted time: {timer.elapsed_time_formatted()}")
+    """
+
+    def __init__(self, logger_obj: logging.Logger, task_str: str):
+        """
+        Initializes the context manager.
+
+        Args:
+            logger_obj: The logger instance to use for output.
+            task_str: A descriptive string for the task being timed.
+        """
+        self.logger = logger_obj
+        self.task_str = task_str
+        self.start_time = None
+        self.elapsed_time = 0.0
+
+    def __enter__(self):
+        """Records the start time when entering the context."""
+        self.start_time = time.time()
+        self.logger.info(f"Starting: {self.task_str}")
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """Calculates and logs the elapsed time upon exiting the context."""
+        end_time = time.time()
+        self.elapsed_time = end_time - self.start_time
+        self.logger.info(f"Finished: {self.task_str} in {self.elapsed_time_formatted()}.")
+
+    def elapsed_time_formatted(self) -> str:
+        """Returns the elapsed time as a formatted string (HH:MM:SS)."""
+        return time.strftime("%H:%M:%S", time.gmtime(self.elapsed_time))
+
+
+def append_timestamp_to_dir(dir_path, append=True):
+    if not append:
+        return dir_path
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    out = dir_path + f"_{timestamp}"
+    return out
+
+
+def shuffle_roles_inplace(config):
+    agents = config["agents"]
+    roles = [agent["role"] for agent in agents]
+    random.shuffle(roles)
+    for new_role, agent in zip(roles, agents):
+        agent["role"] = new_role
+
+
+def run_werewolf(output_dir, base_name, config, agents, debug):
+    """
+    Runs a game of Werewolf, saves the replay, and logs the execution time.
+
+    Args:
+        output_dir (str): The directory where the output files will be saved.
+        base_name (str): The base name for the output files (HTML, JSON).
+        config (dict): The configuration for the Werewolf environment.
+        agents (list): A list of agents to participate in the game.
+        debug (bool): A flag to enable or disable debug mode.
+    """
+    start_time = time.time()
+    logger.info(f"Results saved to {output_dir}.")
+    os.makedirs(output_dir, exist_ok=True)
+    html_file = os.path.join(output_dir, f"{base_name}.html")
+    json_file = os.path.join(output_dir, f"{base_name}.json")
+
+    with LogExecutionTime(logger_obj=logger, task_str="env run") as timer:
+        env = make("werewolf", debug=debug, configuration=config)
+        env.run(agents)
+
+    env.info["total_run_time"] = timer.elapsed_time
+    env.info["total_run_time_formatted"] = timer.elapsed_time_formatted()
+
+    logger.info("Game finished")
+    env_out = env.render(mode="html")
+    with open(html_file, "w") as out:
+        out.write(env_out)
+    logger.info(f"HTML replay written to {html_file}")
+    env_out = env.render(mode="json")
+    with open(json_file, "w") as out:
+        out.write(env_out)
+    logger.info(f"JSON replay written to {json_file}")
+    end_time = time.time()
+    elapsed_time = end_time - start_time
+    formatted_time = time.strftime("%H:%M:%S", time.gmtime(elapsed_time))
+    logger.info(f"Script finished in {formatted_time}.")
+    return env
+
+
+def setup_logger(output_dir, base_name):
+    """
+    Sets up a logger to output to both the console and a log file.
+
+    Args:
+        output_dir (str): The directory where the log file will be saved.
+        base_name (str): The base name for the log file.
+    """
+    log_file = os.path.join(output_dir, f"{base_name}.log")
+    os.makedirs(output_dir, exist_ok=True)
+    handlers = [logging.StreamHandler(), logging.FileHandler(log_file, mode="w")]
+    logging.basicConfig(
+        level=logging.INFO,
+        format="%(asctime)s - %(levelname)s - %(message)s",
+        handlers=handlers,
+    )
+
+
+def log_git_hash():
+    try:
+        result = subprocess.run(
+            ["git", "rev-parse", "HEAD"],
+            cwd=PROJECT_ROOT,
+            capture_output=True,
+            text=True,
+            check=False,  # Don't raise exception on non-zero exit code
+        )
+        if result.returncode == 0:
+            git_hash = result.stdout.strip()
+            logger.info(f"Running from git commit: {git_hash}")
+        else:
+            logger.info("Not a git repository or git command failed.")
+    except FileNotFoundError:
+        logger.info("Git command not found.")