diff --git a/Dockerfile b/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..b20e3729a43ceea9a1afdc12c365d6e70d29988a --- /dev/null +++ b/Dockerfile @@ -0,0 +1,31 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +# Use the standard openenv base image +# Built from: docker build -t openenv-base:latest -f src/core/containers/images/Dockerfile . +# In GitHub Actions, this is overridden to use the GHCR base image +ARG BASE_IMAGE=ghcr.io/meta-pytorch/openenv-base:latest +FROM ghcr.io/meta-pytorch/openenv-base:latest + +# Install dependencies +COPY src/envs/coding_env/server/requirements.txt /tmp/requirements.txt +RUN pip install --no-cache-dir -r /tmp/requirements.txt && rm /tmp/requirements.txt + +# Copy only what's needed for this environment +COPY src/core/ /app/src/core/ +COPY src/envs/coding_env/ /app/src/envs/coding_env/ + +# Copy README for web interface documentation +COPY src/envs/coding_env/README.md /app/README.md + +# Health check +HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \ + CMD curl -f http://localhost:8000/health || exit 1 + +# Run the FastAPI server +ENV ENABLE_WEB_INTERFACE=true + +CMD ["uvicorn", "envs.coding_env.server.app:app", "--host", "0.0.0.0", "--port", "8000"] diff --git a/README.md b/README.md index 79c3c746db259887b1783d47d3402db260e06d19..5390a529e3932b88d15505d9c711a5342cff461e 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,160 @@ --- -title: Coding Env -emoji: šŸ¢ -colorFrom: pink -colorTo: pink +title: Coding_env Environment Server +emoji: šŸ’» +colorFrom: blue +colorTo: gray sdk: docker pinned: false +app_port: 8000 +base_path: /web +tags: + - openenv --- -Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference +# Coding_env Environment Server + +FastAPI server for coding_env environment powered by Meta's OpenEnv. + +## About + +This Space provides a containerized environment for coding_env interactions. +Built with FastAPI and OpenEnv framework. + +## Web Interface + +This deployment includes an interactive web interface for exploring the environment: +- **HumanAgent Interface**: Interact with the environment using a web form +- **State Observer**: Real-time view of environment state and action history +- **Live Updates**: WebSocket-based real-time updates + +Access the web interface at: `/web` + +## API Documentation + +Visit `/docs` for interactive API documentation. + +## Health Check + +The environment provides a health check endpoint at `/health`. + + +# Coding Environment + +A Python code execution environment that runs arbitrary Python code and returns results. Perfect for testing code execution infrastructure and demonstrating environment usage patterns. + +## Quick Start + +The simplest way to use the Coding environment is through the `CodingEnv` class: + +```python +from envs.coding_env import CodeAction, CodingEnv + +try: + # Create environment from Docker image + coding_env = CodingEnv.from_docker_image("coding-env:latest") + + # Reset + result = coding_env.reset() + print(f"Reset complete: exit_code={result.observation.exit_code}") + + # Execute Python code + code_samples = [ + "print('Hello, World!')", + "x = 5 + 3\nprint(f'Result: {x}')", + "import math\nprint(math.pi)" + ] + + for code in code_samples: + result = coding_env.step(CodeAction(code=code)) + print(f"Code: {code}") + print(f" ā†’ stdout: {result.observation.stdout.strip()}") + print(f" ā†’ exit_code: {result.observation.exit_code}") + +finally: + # Always clean up + coding_env.close() +``` + +That's it! The `CodingEnv.from_docker_image()` method handles: +- Starting the Docker container +- Waiting for the server to be ready +- Connecting to the environment +- Container cleanup when you call `close()` + +## Building the Docker Image + +Before using the environment, you need to build the Docker image: + +```bash +# From project root +docker build -t coding-env:latest -f src/envs/coding_env/server/Dockerfile . +``` + +## Environment Details + +### Action +**CodeAction**: Contains a single field +- `code` (str) - The Python code to execute + +### Observation +**CodeObservation**: Contains the execution results +- `stdout` (str) - Standard output from code execution +- `stderr` (str) - Standard error from code execution +- `exit_code` (int) - Exit code (0 for success, non-zero for errors) + +### State +**CodeState**: Tracks execution state +- `episode_id` (str) - Unique identifier for the episode +- `step_count` (int) - Number of steps taken +- `last_exit_code` (int) - Exit code from the last execution + +## Advanced Usage + +### Connecting to an Existing Server + +If you already have a Coding environment server running, you can connect directly: + +```python +from envs.coding_env import CodingEnv + +# Connect to existing server +coding_env = CodingEnv(base_url="") + +# Use as normal +result = coding_env.reset() +result = coding_env.step(CodeAction(code="print('Hello!')")) +``` + +Note: When connecting to an existing server, `coding_env.close()` will NOT stop the server. + +## Development & Testing + +### Running the Full Example + +Run the complete example that demonstrates the full workflow: + +```bash +python3 src/envs/coding_env/client/example_usage.py +``` + +This example shows: +- Creating an environment from a Docker image +- Resetting and executing code through the environment +- Automatic cleanup with `close()` + +## Project Structure + +``` +coding_env/ +ā”œā”€ā”€ README.md # This file +ā”œā”€ā”€ models.py # Action, Observation, and State models +ā”œā”€ā”€ client/ +ā”‚ ā”œā”€ā”€ coding_env_client.py # CodingEnv client implementation +ā”‚ ā””ā”€ā”€ example_usage.py # Usage examples +ā””ā”€ā”€ server/ + ā”œā”€ā”€ python_codeact_env.py # Core environment logic + ā”œā”€ā”€ app.py # FastAPI application + ā”œā”€ā”€ transforms.py # Observation transforms + ā”œā”€ā”€ Dockerfile # Container image definition + ā””ā”€ā”€ README.md # Server-specific documentation +``` diff --git a/src/core/README.md b/src/core/README.md new file mode 100644 index 0000000000000000000000000000000000000000..bb8b4746e1955a4546a30e9d29a6bd5ff1866511 --- /dev/null +++ b/src/core/README.md @@ -0,0 +1,180 @@ +# image OpenEnv: Agentic Execution Environments + +An e2e framework for creating, deploying and using isolated execution environments for agentic RL training, built using Gymnasium style simple APIs. OpenEnv provides a standard for interacting with agentic execution environments via simple Gymnasium style APIs - step(), reset(), state(). Users of agentic execution environments can interact with the environment during RL training loops using these simple APIs. + +In addition to making it easier for researchers and RL framework writers, we also provide tools for environment creators making it easier for them to create richer environments and make them available over familar protocols like HTTP and packaged using canonical technologies like docker. Environment creators can use the OpenEnv framework to create environments that are isolated, secure, and easy to deploy and use. + + +## Overview +`openenv-core` provides the foundational building blocks for creating and interacting with containerized environments over HTTP. It enables you to build agent environments that can be deployed as Docker containers and accessed via a simple HTTP API. + +> āš ļø **Early Development Warning** OpenEnv is currently in an experimental +> stage. You should expect bugs, incomplete features, and APIs that may change +> in future versions. The project welcomes bugfixes, but to make sure things are +> well coordinated you should discuss any significant change before starting the +> work. It's recommended that you signal your intention to contribute in the +> issue tracker, either by filing a new issue or by claiming an existing one. + + +# OpenEnv Core + +Core components for OpenEnv - a framework for building HTTP-based agentic environments. + +## Features + +- **HTTPEnvClient**: Generic HTTP client for interacting with remote environments +- **HTTPEnvServer**: FastAPI-based server wrapper for exposing environments over HTTP +- **Container Providers**: Pluggable architecture for running containers (Docker, Kubernetes, etc.) +- **Type System**: Strongly-typed Action/Observation/State interfaces +- **Web Interface**: Optional web UI for interacting with environments + +## Installation + +```bash +pip install openenv-core +``` + +For development: +```bash +pip install openenv-core[dev] +``` + +## Quick Start + +### Creating an Environment Client + +```python +from openenv_core import HTTPEnvClient, StepResult +from dataclasses import dataclass + +@dataclass +class MyAction: + text: str + +@dataclass +class MyObservation: + response: str + +class MyEnvClient(HTTPEnvClient[MyAction, MyObservation]): + def _step_payload(self, action: MyAction) -> dict: + return {"text": action.text} + + def _parse_result(self, payload: dict) -> StepResult[MyObservation]: + obs_data = payload["observation"] + return StepResult( + observation=MyObservation(**obs_data), + reward=payload.get("reward"), + done=payload.get("done", False) + ) + + def _parse_state(self, payload: dict) -> Any: + return payload + +# Use with Docker +env = MyEnvClient.from_docker_image("my-env:latest") +result = env.reset() +step_result = env.step(MyAction(text="hello")) +env.close() +``` + +### Creating an Environment Server + +```python +from openenv_core.env_server import Environment, HTTPEnvServer, create_app +from dataclasses import dataclass + +@dataclass +class MyAction: + text: str + +@dataclass +class MyObservation: + response: str + reward: float = 0.0 + done: bool = False + +class MyEnvironment(Environment): + def reset(self) -> MyObservation: + return MyObservation(response="Ready") + + def step(self, action: MyAction) -> MyObservation: + return MyObservation( + response=f"Echo: {action.text}", + reward=1.0, + done=False + ) + +# Create FastAPI app +env = MyEnvironment() +app = create_app(env, MyAction, MyObservation) + +# Run with: uvicorn module:app --host 0.0.0.0 --port 8000 +``` + +## Container Providers + +OpenEnv Core supports multiple container providers: + +### Local Docker Provider + +```python +from openenv_core.containers.runtime import LocalDockerProvider + +provider = LocalDockerProvider() +base_url = provider.start_container("my-env:latest") +provider.wait_for_ready(base_url) +# Use environment... +provider.stop_container() +``` + +### Kubernetes Provider (Coming Soon) + +```python +from openenv_core.containers.runtime import KubernetesProvider + +provider = KubernetesProvider(namespace="envs") +base_url = provider.start_container("my-env:latest") +# Use environment... +provider.stop_container() +``` + + +## API Reference + +### HTTPEnvClient + +Base class for environment clients with these abstract methods: + +- `_step_payload(action)`: Convert action to JSON +- `_parse_result(payload)`: Parse response to StepResult +- `_parse_state(payload)`: Parse state response + +### HTTPEnvServer + +Server wrapper with these methods: + +- `register_routes(app)`: Register endpoints on FastAPI app +- `_deserialize_action(data)`: Convert JSON to Action +- `_serialize_observation(obs)`: Convert Observation to JSON + +### Environment Interface + +Base interface for environment implementations: + +- `reset()`: Reset environment and return initial observation +- `step(action)`: Execute action and return observation +- `state`: Property returning current environment state + +## License + +This project is licensed under the BSD-3-Clause License - see the LICENSE file for details. + +## Contributing + +Contributions are welcome! Please see the main OpenEnv repository for contribution guidelines. + +## Links + +- **Homepage**: https://github.com/facebookresearch/OpenEnv +- **Documentation**: https://github.com/facebookresearch/OpenEnv/blob/main/README.md +- **Bug Tracker**: https://github.com/facebookresearch/OpenEnv/issues diff --git a/src/core/__init__.py b/src/core/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..99507ab55a82b406c8bdef55a77b79855484686e --- /dev/null +++ b/src/core/__init__.py @@ -0,0 +1,19 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Core components for agentic environments.""" + +# Re-export main components from submodules for convenience +from .env_server import * +from .client_types import StepResult +from .http_env_client import HTTPEnvClient + +# Note: MCP module doesn't export anything yet + +__all__ = [ + "HTTPEnvClient", + "StepResult", +] diff --git a/src/core/__pycache__/__init__.cpython-311.pyc b/src/core/__pycache__/__init__.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ba8ebcb7f8b8749f8a2b8704d4595ecf0761d681 Binary files /dev/null and b/src/core/__pycache__/__init__.cpython-311.pyc differ diff --git a/src/core/__pycache__/__init__.cpython-313.pyc b/src/core/__pycache__/__init__.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..88c44b6dc9955eee654fecbc3691949d1fa85be8 Binary files /dev/null and b/src/core/__pycache__/__init__.cpython-313.pyc differ diff --git a/src/core/__pycache__/http_env_client.cpython-311.pyc b/src/core/__pycache__/http_env_client.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..6b6b005f00b0ffcaf273052532ad5435a3173d62 Binary files /dev/null and b/src/core/__pycache__/http_env_client.cpython-311.pyc differ diff --git a/src/core/__pycache__/http_env_client.cpython-313.pyc b/src/core/__pycache__/http_env_client.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d07b11d4de7cfa2393fd44ef242709601b8e3c8a Binary files /dev/null and b/src/core/__pycache__/http_env_client.cpython-313.pyc differ diff --git a/src/core/__pycache__/types.cpython-311.pyc b/src/core/__pycache__/types.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..ec35f9f8707f7797eda9cc4c318cf022bf6ebcaf Binary files /dev/null and b/src/core/__pycache__/types.cpython-311.pyc differ diff --git a/src/core/__pycache__/types.cpython-313.pyc b/src/core/__pycache__/types.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..238fc7dfc99847eac022e74f4fc2efcceb7ecae3 Binary files /dev/null and b/src/core/__pycache__/types.cpython-313.pyc differ diff --git a/src/core/client_types.py b/src/core/client_types.py new file mode 100644 index 0000000000000000000000000000000000000000..8808e96bf713e95f94d9bc7f2e743f3fee616306 --- /dev/null +++ b/src/core/client_types.py @@ -0,0 +1,22 @@ +# Type definitions for EnvTorch +from dataclasses import dataclass +from typing import Any, Generic, Optional, TypeVar + +# Generic type for observations +ObsT = TypeVar("ObsT") # TypeVar for typehinting in IDEs + + +@dataclass +class StepResult(Generic[ObsT]): + """ + Represents the result of one environment step. + + Attributes: + observation: The environment's observation after the action. + reward: Scalar reward for this step (optional). + done: Whether the episode is finished. + """ + + observation: ObsT + reward: Optional[float] = None + done: bool = False diff --git a/src/core/containers/__init__.py b/src/core/containers/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..59ce71cdfb16733e5bb941be99d412ec05a7ba7a --- /dev/null +++ b/src/core/containers/__init__.py @@ -0,0 +1,7 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Container management for environment servers.""" \ No newline at end of file diff --git a/src/core/containers/__pycache__/__init__.cpython-311.pyc b/src/core/containers/__pycache__/__init__.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..746a81f3bf915ce548b3da69ade840c757702066 Binary files /dev/null and b/src/core/containers/__pycache__/__init__.cpython-311.pyc differ diff --git a/src/core/containers/__pycache__/__init__.cpython-313.pyc b/src/core/containers/__pycache__/__init__.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..beb9d5d8f9d1e8368112395c9de1aa15dbb4a6a5 Binary files /dev/null and b/src/core/containers/__pycache__/__init__.cpython-313.pyc differ diff --git a/src/core/containers/images/Dockerfile b/src/core/containers/images/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..9dd7c42a37a4285f2e5e1d43b497e1fcf04624cf --- /dev/null +++ b/src/core/containers/images/Dockerfile @@ -0,0 +1,46 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +# +# OpenEnv Base Image +# +# This is the standard base image for all OpenEnv environment servers. +# It includes the minimal dependencies needed to run HTTP environment servers. +# +# Build: docker build -t openenv-base:latest -f src/core/containers/images/Dockerfile . +# Tag: docker tag openenv-base:latest openenv-base:0.1.0 +# + +FROM python:3.11-slim + +# Set metadata +LABEL maintainer="OpenEnv Team" +LABEL description="Base image for OpenEnv based environment servers" +LABEL version="0.1.0" + +# Install system dependencies +RUN apt-get update && apt-get install -y --no-install-recommends \ + curl \ + && rm -rf /var/lib/apt/lists/* + +# Install Python dependencies that all environments need +RUN pip install --no-cache-dir \ + fastapi>=0.104.0 \ + "uvicorn[standard]>=0.24.0" \ + requests>=2.25.0 \ + wsproto>=1.0.0 + +# Set working directory +WORKDIR /app + +# Default environment variables +ENV PYTHONPATH=/app/src +ENV PYTHONUNBUFFERED=1 + +# Default expose port (can be overridden) +EXPOSE 8000 + +# Note: CMD should be specified in child Dockerfiles diff --git a/src/core/containers/images/README.md b/src/core/containers/images/README.md new file mode 100644 index 0000000000000000000000000000000000000000..bc286446673987614fa5d4f15b5d1de4060b3642 --- /dev/null +++ b/src/core/containers/images/README.md @@ -0,0 +1,92 @@ +# OpenEnv Base Image + +Standard base image for all OpenEnv environment servers. + +## What's Included + +| Layer | Size | Contents | +|-------|------|----------| +| python:3.11-slim | 200 MB | Base Python runtime | +| + Dependencies | 100 MB | FastAPI, uvicorn, requests | +| **Total** | **~300 MB** | Ready for environment servers | + +## Image Sizes + +``` +openenv-base:latest 300 MB (python + fastapi + uvicorn) +``` +echo-env:latest 500 MB (python + fastapi + uvicorn + app) +coding-env:latest 520 MB (python + fastapi + uvicorn + app + tools) +another-env:latest 510 MB (python + fastapi + uvicorn + app) +--- +Total: 1.5 GB (with lots of duplication) +``` + +### With Base Images (āœ… Solution) +``` +openenv-base:latest 300 MB (python + fastapi + uvicorn) +echo-env:latest 50 MB (app only, uses base) +coding-env:latest 70 MB (app + tools, uses base) +another-env:latest 45 MB (app only, uses base) +--- +Total: 465 MB (base shared, minimal duplication) +``` + +## Building the Base Image + +```bash +# From project root +docker build -t openenv-base:latest -f src/core/containers/images/Dockerfile . +``` + +## Usage in Environment Dockerfiles + +Each environment Dockerfile should start with: + +```dockerfile +FROM openenv-base:latest + +# Copy only environment-specific files +COPY src/core/ /app/src/core/ +COPY src/envs/my_env/ /app/src/envs/my_env/ + +# Run the server +CMD ["uvicorn", "envs.my_env.server.app:app", "--host", "0.0.0.0", "--port", "8000"] +``` + +## Base Image Contents + +- Python 3.11-slim +- FastAPI >= 0.104.0 +- Uvicorn >= 0.24.0 +- Requests >= 2.25.0 +- curl (for health checks) + +## Example: Building Echo Environment + +```bash +# Step 1: Build base image (do this once) +docker build -t openenv-base:latest -f src/core/containers/images/Dockerfile . + +# Step 2: Build echo environment (uses base) +docker build -t echo-env:latest -f src/envs/echo_env/server/Dockerfile . + +# Step 3: Run echo environment +docker run -p 8000:8000 echo-env:latest +``` + +## Updating the Base + +When dependencies need updating: + +1. Update `src/core/containers/images/Dockerfile` +2. Rebuild base image +3. Rebuild all environment images (they'll use new base) + +```bash +# Update base +docker build -t openenv-base:latest -f src/core/containers/images/Dockerfile . + +# Rebuild environments (they automatically use new base) +docker build -t echo-env:latest -f src/envs/echo_env/server/Dockerfile . +``` diff --git a/src/core/containers/runtime/__init__.py b/src/core/containers/runtime/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..a72b5301017f33e13586dbf86d109c18ef008970 --- /dev/null +++ b/src/core/containers/runtime/__init__.py @@ -0,0 +1,15 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Container runtime providers.""" + +from .providers import ContainerProvider, KubernetesProvider, LocalDockerProvider + +__all__ = [ + "ContainerProvider", + "LocalDockerProvider", + "KubernetesProvider", +] \ No newline at end of file diff --git a/src/core/containers/runtime/__pycache__/__init__.cpython-311.pyc b/src/core/containers/runtime/__pycache__/__init__.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..079aaa94902a3c79800183743ce17bff27f898bf Binary files /dev/null and b/src/core/containers/runtime/__pycache__/__init__.cpython-311.pyc differ diff --git a/src/core/containers/runtime/__pycache__/__init__.cpython-313.pyc b/src/core/containers/runtime/__pycache__/__init__.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..09b66b143d5673d754bb5a43c0a396c9ac864e97 Binary files /dev/null and b/src/core/containers/runtime/__pycache__/__init__.cpython-313.pyc differ diff --git a/src/core/containers/runtime/__pycache__/providers.cpython-311.pyc b/src/core/containers/runtime/__pycache__/providers.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..e5f923e32dc09b421d57329c92536da89e597910 Binary files /dev/null and b/src/core/containers/runtime/__pycache__/providers.cpython-311.pyc differ diff --git a/src/core/containers/runtime/__pycache__/providers.cpython-313.pyc b/src/core/containers/runtime/__pycache__/providers.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..8fce811a6c7a7146e9c1f800a26db7ceb48feed7 Binary files /dev/null and b/src/core/containers/runtime/__pycache__/providers.cpython-313.pyc differ diff --git a/src/core/containers/runtime/providers.py b/src/core/containers/runtime/providers.py new file mode 100644 index 0000000000000000000000000000000000000000..a8022ddcac04592044997393fb4d4427c5c43e66 --- /dev/null +++ b/src/core/containers/runtime/providers.py @@ -0,0 +1,293 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Container provider abstractions for running environment servers. + +This module provides a pluggable architecture for different container providers +(local Docker, Kubernetes, cloud providers, etc.) to be used with HTTPEnvClient. +""" + +from __future__ import annotations + +from abc import ABC, abstractmethod +from typing import Any, Dict, Optional + + +class ContainerProvider(ABC): + """ + Abstract base class for container providers. + + Providers implement this interface to support different container platforms: + - LocalDockerProvider: Runs containers on local Docker daemon + - KubernetesProvider: Runs containers in Kubernetes cluster + - FargateProvider: Runs containers on AWS Fargate + - CloudRunProvider: Runs containers on Google Cloud Run + + The provider manages a single container lifecycle and provides the base URL + for connecting to it. + + Example: + >>> provider = LocalDockerProvider() + >>> base_url = provider.start_container("echo-env:latest") + >>> print(base_url) # http://localhost:8000 + >>> # Use the environment via base_url + >>> provider.stop_container() + """ + + @abstractmethod + def start_container( + self, + image: str, + port: Optional[int] = None, + env_vars: Optional[Dict[str, str]] = None, + **kwargs: Any, + ) -> str: + """ + Start a container from the specified image. + + Args: + image: Container image name (e.g., "echo-env:latest") + port: Port to expose (if None, provider chooses) + env_vars: Environment variables to pass to container + **kwargs: Provider-specific options + + Returns: + Base URL to connect to the container (e.g., "http://localhost:8000") + + Raises: + RuntimeError: If container fails to start + """ + pass + + @abstractmethod + def stop_container(self) -> None: + """ + Stop and remove the running container. + + This cleans up the container that was started by start_container(). + """ + pass + + @abstractmethod + def wait_for_ready(self, base_url: str, timeout_s: float = 30.0) -> None: + """ + Wait for the container to be ready to accept requests. + + This typically polls the /health endpoint until it returns 200. + + Args: + base_url: Base URL of the container + timeout_s: Maximum time to wait + + Raises: + TimeoutError: If container doesn't become ready in time + """ + pass + + +class LocalDockerProvider(ContainerProvider): + """ + Container provider for local Docker daemon. + + This provider runs containers on the local machine using Docker. + Useful for development and testing. + + Example: + >>> provider = LocalDockerProvider() + >>> base_url = provider.start_container("echo-env:latest") + >>> # Container running on http://localhost: + >>> provider.stop_container() + """ + + def __init__(self): + """Initialize the local Docker provider.""" + self._container_id: Optional[str] = None + self._container_name: Optional[str] = None + + # Check if Docker is available + import subprocess + + try: + subprocess.run( + ["docker", "version"], + check=True, + capture_output=True, + timeout=5, + ) + except (subprocess.CalledProcessError, FileNotFoundError, subprocess.TimeoutExpired): + raise RuntimeError( + "Docker is not available. Please install Docker Desktop or Docker Engine." + ) + + def start_container( + self, + image: str, + port: Optional[int] = None, + env_vars: Optional[Dict[str, str]] = None, + **kwargs: Any, + ) -> str: + """ + Start a Docker container locally. + + Args: + image: Docker image name + port: Port to expose (if None, finds available port) + env_vars: Environment variables for the container + **kwargs: Additional Docker run options + + Returns: + Base URL to connect to the container + """ + import subprocess + import time + + # Find available port if not specified + if port is None: + port = self._find_available_port() + + # Generate container name + self._container_name = self._generate_container_name(image) + + # Build docker run command + cmd = [ + "docker", "run", + "-d", # Detached + "--name", self._container_name, + "-p", f"{port}:8000", # Map port + ] + + # Add environment variables + if env_vars: + for key, value in env_vars.items(): + cmd.extend(["-e", f"{key}={value}"]) + + # Add image + cmd.append(image) + + # Run container + try: + result = subprocess.run(cmd, capture_output=True, text=True, check=True) + self._container_id = result.stdout.strip() + except subprocess.CalledProcessError as e: + error_msg = f"Failed to start Docker container.\nCommand: {' '.join(cmd)}\nExit code: {e.returncode}\nStderr: {e.stderr}\nStdout: {e.stdout}" + raise RuntimeError(error_msg) from e + + # Wait a moment for container to start + time.sleep(1) + + base_url = f"http://localhost:{port}" + return base_url + + def stop_container(self) -> None: + """ + Stop and remove the Docker container. + """ + if self._container_id is None: + return + + import subprocess + + try: + # Stop container + subprocess.run( + ["docker", "stop", self._container_id], + capture_output=True, + check=True, + timeout=10, + ) + + # Remove container + subprocess.run( + ["docker", "rm", self._container_id], + capture_output=True, + check=True, + timeout=10, + ) + except subprocess.CalledProcessError: + # Container might already be stopped/removed + pass + finally: + self._container_id = None + self._container_name = None + + def wait_for_ready(self, base_url: str, timeout_s: float = 30.0) -> None: + """ + Wait for container to be ready by polling /health endpoint. + + Args: + base_url: Base URL of the container + timeout_s: Maximum time to wait + + Raises: + TimeoutError: If container doesn't become ready + """ + import time + import requests + + start_time = time.time() + health_url = f"{base_url}/health" + + while time.time() - start_time < timeout_s: + try: + response = requests.get(health_url, timeout=2.0) + if response.status_code == 200: + return + except requests.RequestException: + pass + + time.sleep(0.5) + + raise TimeoutError( + f"Container at {base_url} did not become ready within {timeout_s}s" + ) + + def _find_available_port(self) -> int: + """ + Find an available port on localhost. + + Returns: + An available port number + """ + import socket + + with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: + s.bind(("", 0)) + s.listen(1) + port = s.getsockname()[1] + return port + + def _generate_container_name(self, image: str) -> str: + """ + Generate a unique container name based on image name and timestamp. + + Args: + image: Docker image name + + Returns: + A unique container name + """ + import time + + clean_image = image.split("/")[-1].split(":")[0] + timestamp = int(time.time() * 1000) + return f"{clean_image}-{timestamp}" + + +class KubernetesProvider(ContainerProvider): + """ + Container provider for Kubernetes clusters. + + This provider creates pods in a Kubernetes cluster and exposes them + via services or port-forwarding. + + Example: + >>> provider = KubernetesProvider(namespace="envtorch-dev") + >>> base_url = provider.start_container("echo-env:latest") + >>> # Pod running in k8s, accessible via service or port-forward + >>> provider.stop_container() + """ + pass diff --git a/src/core/containers/test_local_docker_provider.py b/src/core/containers/test_local_docker_provider.py new file mode 100644 index 0000000000000000000000000000000000000000..e435ff6dc56f90ecf454c1b4ebe368348e3dac1f --- /dev/null +++ b/src/core/containers/test_local_docker_provider.py @@ -0,0 +1,258 @@ +#!/usr/bin/env python3 +""" +End-to-end test for LocalDockerProvider. + +This script tests the complete flow: +1. Start a container using LocalDockerProvider +2. Wait for it to be ready +3. Make HTTP requests to test the environment +4. Clean up the container +""" + +import sys +from pathlib import Path + +# Add src to path +sys.path.insert(0, str(Path(__file__).parent.parent.parent)) + +import requests + +from core.containers.runtime import LocalDockerProvider + +# TODO: Remove this test or make it a functional test sicne this will be tested in e2e test for echo env +def test_local_docker_provider(): + """Test LocalDockerProvider end-to-end.""" + print("=" * 60) + print("LocalDockerProvider End-to-End Test") + print("=" * 60) + print() + + provider = None + + try: + # Step 1: Create provider + print("Step 1: Creating LocalDockerProvider...") + provider = LocalDockerProvider() + print("āœ“ Provider created\n") + + # Step 2: Start container + print("Step 2: Starting echo-env container...") + base_url = provider.start_container("echo-env:latest") + print(f"āœ“ Container started at: {base_url}") + if provider._container_id: + print(f" Container ID: {provider._container_id[:12]}...") + if provider._container_name: + print(f" Container name: {provider._container_name}\n") + + # Step 3: Wait for ready + print("Step 3: Waiting for container to be ready...") + provider.wait_for_ready(base_url, timeout_s=30.0) + print("āœ“ Container is ready!\n") + + # Step 4: Test health endpoint + print("Step 4: Testing /health endpoint...") + response = requests.get(f"{base_url}/health") + print(f" Status: {response.status_code}") + print(f" Response: {response.json()}") + assert response.status_code == 200 + assert response.json()["status"] == "healthy" + print("āœ“ Health check passed\n") + + # Step 5: Test reset endpoint + print("Step 5: Testing /reset endpoint...") + response = requests.post( + f"{base_url}/reset", + json={}, + headers={"Content-Type": "application/json"}, + ) + print(f" Status: {response.status_code}") + data = response.json() + print(f" Message: {data['observation']['echoed_message']}") + print(f" Reward: {data['reward']}") + print(f" Done: {data['done']}") + assert response.status_code == 200 + assert data["observation"]["echoed_message"] == "Echo environment ready!" + print("āœ“ Reset test passed\n") + + # Step 6: Test step endpoint + print("Step 6: Testing /step endpoint...") + response = requests.post( + f"{base_url}/step", + json={"action": {"message": "Hello from LocalDockerProvider!"}}, + headers={"Content-Type": "application/json"}, + ) + print(f" Status: {response.status_code}") + data = response.json() + print(f" Echoed: {data['observation']['echoed_message']}") + print(f" Length: {data['observation']['message_length']}") + print(f" Reward: {data['reward']}") + assert response.status_code == 200 + assert data["observation"]["echoed_message"] == "Hello from LocalDockerProvider!" + assert data["observation"]["message_length"] == 31 + print("āœ“ Step test passed\n") + + # Step 7: Test state endpoint + print("Step 7: Testing /state endpoint...") + response = requests.get(f"{base_url}/state") + print(f" Status: {response.status_code}") + data = response.json() + print(f" Episode ID: {data['episode_id']}") + print(f" Step count: {data['step_count']}") + assert response.status_code == 200 + assert data["step_count"] == 1 # One step from above + print("āœ“ State test passed\n") + + # Step 8: Multiple steps + print("Step 8: Testing multiple steps...") + for i in range(3): + response = requests.post( + f"{base_url}/step", + json={"action": {"message": f"Message {i+1}"}}, + headers={"Content-Type": "application/json"}, + ) + assert response.status_code == 200 + print(f" Step {i+1}: āœ“") + + # Check state updated + response = requests.get(f"{base_url}/state") + data = response.json() + assert data["step_count"] == 4 # 1 + 3 more steps + print(f" Final step count: {data['step_count']}") + print("āœ“ Multiple steps test passed\n") + + print("=" * 60) + print("āœ“ All tests passed!") + print("=" * 60) + print() + + return True + + except Exception as e: + print(f"\nāŒ Test failed: {e}") + import traceback + traceback.print_exc() + return False + + finally: + # Step 9: Cleanup + if provider is not None: + print("\nStep 9: Cleaning up container...") + try: + provider.stop_container() + print("āœ“ Container stopped and removed\n") + except Exception as e: + print(f"āš ļø Cleanup warning: {e}\n") + + +def test_provider_with_custom_port(): + """Test provider with custom port.""" + print("=" * 60) + print("LocalDockerProvider with Custom Port Test") + print("=" * 60) + print() + + provider = None + + try: + provider = LocalDockerProvider() + + print("Starting container on custom port 8123...") + base_url = provider.start_container("echo-env:latest", port=8123) + print(f"āœ“ Started at: {base_url}") + assert ":8123" in base_url + + print("Waiting for ready...") + provider.wait_for_ready(base_url) + print("āœ“ Ready!") + + print("Testing health...") + response = requests.get(f"{base_url}/health") + assert response.status_code == 200 + print("āœ“ Health check passed") + + print("\nāœ“ Custom port test passed!\n") + return True + + except Exception as e: + print(f"\nāŒ Test failed: {e}") + return False + + finally: + if provider is not None: + provider.stop_container() + print("āœ“ Cleaned up\n") + + +def test_provider_with_env_vars(): + """Test provider with environment variables.""" + print("=" * 60) + print("LocalDockerProvider with Environment Variables Test") + print("=" * 60) + print() + + provider = None + + try: + provider = LocalDockerProvider() + + print("Starting container with environment variables...") + base_url = provider.start_container( + "echo-env:latest", + env_vars={"DEBUG": "true", "LOG_LEVEL": "info"} + ) + print(f"āœ“ Started at: {base_url}") + + print("Waiting for ready...") + provider.wait_for_ready(base_url) + print("āœ“ Ready!") + + print("Testing health...") + response = requests.get(f"{base_url}/health") + assert response.status_code == 200 + print("āœ“ Health check passed") + + print("\nāœ“ Environment variables test passed!\n") + return True + + except Exception as e: + print(f"\nāŒ Test failed: {e}") + return False + + finally: + if provider is not None: + provider.stop_container() + print("āœ“ Cleaned up\n") + + +if __name__ == "__main__": + print() + print("šŸ³ LocalDockerProvider Test Suite") + print() + + results = [] + + # Run basic test + results.append(("Basic End-to-End", test_local_docker_provider())) + + # Run custom port test + results.append(("Custom Port", test_provider_with_custom_port())) + + # Run environment variables test + results.append(("Environment Variables", test_provider_with_env_vars())) + + # Summary + print("=" * 60) + print("Test Summary") + print("=" * 60) + for name, passed in results: + status = "āœ“ PASSED" if passed else "āœ— FAILED" + print(f"{name:25} {status}") + print("=" * 60) + + all_passed = all(result for _, result in results) + if all_passed: + print("\nšŸŽ‰ All tests passed!") + exit(0) + else: + print("\nāŒ Some tests failed") + exit(1) diff --git a/src/core/env_server/__init__.py b/src/core/env_server/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..79e66535f0e74ae18d84181a234365fca8f3ffc1 --- /dev/null +++ b/src/core/env_server/__init__.py @@ -0,0 +1,35 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Core environment interfaces and types.""" + +from .base_transforms import CompositeTransform, NullTransform +from .http_server import HTTPEnvServer, create_app, create_fastapi_app +from .interfaces import Environment, Message, ModelTokenizer, Transform +from .types import Action, Observation, State +from .web_interface import create_web_interface_app, WebInterfaceManager + +__all__ = [ + # Core interfaces + "Environment", + "Transform", + "Message", + "ModelTokenizer", + # Types + "Action", + "Observation", + "State", + # Base transforms + "CompositeTransform", + "NullTransform", + # HTTP Server + "HTTPEnvServer", + "create_app", + "create_fastapi_app", + # Web Interface + "create_web_interface_app", + "WebInterfaceManager", +] diff --git a/src/core/env_server/__pycache__/__init__.cpython-311.pyc b/src/core/env_server/__pycache__/__init__.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..a0a51ed3a35d5b3a28698f72357ebe03175e0350 Binary files /dev/null and b/src/core/env_server/__pycache__/__init__.cpython-311.pyc differ diff --git a/src/core/env_server/__pycache__/__init__.cpython-313.pyc b/src/core/env_server/__pycache__/__init__.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2b861c9150aead17bae1e82a7b440b3b8a769e03 Binary files /dev/null and b/src/core/env_server/__pycache__/__init__.cpython-313.pyc differ diff --git a/src/core/env_server/__pycache__/base_transforms.cpython-311.pyc b/src/core/env_server/__pycache__/base_transforms.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..606d0f23676904137940a658494d5bff04753a80 Binary files /dev/null and b/src/core/env_server/__pycache__/base_transforms.cpython-311.pyc differ diff --git a/src/core/env_server/__pycache__/base_transforms.cpython-313.pyc b/src/core/env_server/__pycache__/base_transforms.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..950bba6af0a51ac7a91477d6eb5cbe879e1f45f6 Binary files /dev/null and b/src/core/env_server/__pycache__/base_transforms.cpython-313.pyc differ diff --git a/src/core/env_server/__pycache__/http_server.cpython-311.pyc b/src/core/env_server/__pycache__/http_server.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..5fc86f58da2a3444881c55c406199620bcfeb54d Binary files /dev/null and b/src/core/env_server/__pycache__/http_server.cpython-311.pyc differ diff --git a/src/core/env_server/__pycache__/http_server.cpython-313.pyc b/src/core/env_server/__pycache__/http_server.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..725771dc396cdb029566c305d2e94bd4084efe31 Binary files /dev/null and b/src/core/env_server/__pycache__/http_server.cpython-313.pyc differ diff --git a/src/core/env_server/__pycache__/interfaces.cpython-311.pyc b/src/core/env_server/__pycache__/interfaces.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..87266363c5c0059b778a30357c3c0b0b9ae18fec Binary files /dev/null and b/src/core/env_server/__pycache__/interfaces.cpython-311.pyc differ diff --git a/src/core/env_server/__pycache__/interfaces.cpython-313.pyc b/src/core/env_server/__pycache__/interfaces.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..093efd15810f772043140b2d80d2b515ed94c244 Binary files /dev/null and b/src/core/env_server/__pycache__/interfaces.cpython-313.pyc differ diff --git a/src/core/env_server/__pycache__/types.cpython-311.pyc b/src/core/env_server/__pycache__/types.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..dc459b2849155a21845eb184dec1f993dc92f4ec Binary files /dev/null and b/src/core/env_server/__pycache__/types.cpython-311.pyc differ diff --git a/src/core/env_server/__pycache__/types.cpython-313.pyc b/src/core/env_server/__pycache__/types.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..2e6abb8f0a00ab35fc258820af8fb300dde8e2cc Binary files /dev/null and b/src/core/env_server/__pycache__/types.cpython-313.pyc differ diff --git a/src/core/env_server/__pycache__/web_interface.cpython-311.pyc b/src/core/env_server/__pycache__/web_interface.cpython-311.pyc new file mode 100644 index 0000000000000000000000000000000000000000..46b3abde314829fec78b6fb7e02426d136ff9aa6 Binary files /dev/null and b/src/core/env_server/__pycache__/web_interface.cpython-311.pyc differ diff --git a/src/core/env_server/__pycache__/web_interface.cpython-313.pyc b/src/core/env_server/__pycache__/web_interface.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..b6e72a4a8cdb0d8548d5838713f3fc49fc921d79 Binary files /dev/null and b/src/core/env_server/__pycache__/web_interface.cpython-313.pyc differ diff --git a/src/core/env_server/base_transforms.py b/src/core/env_server/base_transforms.py new file mode 100644 index 0000000000000000000000000000000000000000..d8165e3d77ba23bbd4c765e46cc38bc6c475ad4c --- /dev/null +++ b/src/core/env_server/base_transforms.py @@ -0,0 +1,29 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Base transform implementations for composing environment-specific transforms.""" + +from .interfaces import Transform +from .types import Observation + + +class CompositeTransform(Transform): + """Combines multiple transforms into a single transform.""" + + def __init__(self, transforms: list[Transform]): + self.transforms = transforms + + def __call__(self, observation: Observation) -> Observation: + for transform in self.transforms: + observation = transform(observation) + return observation + + +class NullTransform(Transform): + """Default transform that passes through unchanged.""" + + def __call__(self, observation: Observation) -> Observation: + return observation \ No newline at end of file diff --git a/src/core/env_server/http_server.py b/src/core/env_server/http_server.py new file mode 100644 index 0000000000000000000000000000000000000000..d18873f0126d8c9198ac24f7308c6a358ccd9435 --- /dev/null +++ b/src/core/env_server/http_server.py @@ -0,0 +1,233 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +HTTP server wrapper for Environment instances. + +This module provides utilities to wrap any Environment subclass and expose it +over HTTP endpoints that HTTPEnvClient can consume. +""" + +from __future__ import annotations + +import os +from dataclasses import asdict +from typing import Any, Dict, Type + +from .interfaces import Environment +from .types import Action, Observation +from fastapi import Body, FastAPI + +class HTTPEnvServer: + """ + HTTP server wrapper for Environment instances. + + This class wraps an Environment and exposes its reset(), step(), and state + methods as HTTP endpoints compatible with HTTPEnvClient. + + The server expects: + - Action deserialization: Converts JSON dict to Action subclass + - Observation serialization: Converts Observation subclass to JSON dict + + Example: + >>> from core.env_server import HTTPEnvServer + >>> from envs.coding_env.server import CodeExecutionEnvironment + >>> + >>> env = CodeExecutionEnvironment() + >>> server = HTTPEnvServer(env) + >>> + >>> # Register routes with FastAPI + >>> from fastapi import FastAPI + >>> app = FastAPI() + >>> server.register_routes(app) + """ + + def __init__( + self, + env: Environment, + action_cls: Type[Action], + observation_cls: Type[Observation], + ): + """ + Initialize HTTP server wrapper. + + Args: + env: The Environment instance to wrap + action_cls: The Action subclass this environment expects + observation_cls: The Observation subclass this environment returns + """ + self.env = env + self.action_cls = action_cls + self.observation_cls = observation_cls + + def register_routes(self, app: Any) -> None: + """ + Register HTTP routes on a FastAPI application. + + Args: + app: FastAPI application instance + """ + + if not isinstance(app, FastAPI): + raise TypeError("app must be a FastAPI instance") + + @app.post("/reset") + async def reset(request: Dict[str, Any] = Body(default={})) -> Dict[str, Any]: + """Reset endpoint - returns initial observation.""" + # TODO: Handle seed, episode_id from request if provided + observation = self.env.reset() + return self._serialize_observation(observation) + + @app.post("/step") + async def step(request: Dict[str, Any]) -> Dict[str, Any]: + """Step endpoint - executes action and returns observation.""" + action_data = request.get("action", {}) + # TODO: Handle timeout_s, request_id, episode_id from request if provided + + # Deserialize action + action = self._deserialize_action(action_data) + + # Execute step + observation = self.env.step(action) + + # Return serialized observation + return self._serialize_observation(observation) + + @app.get("/state") + async def get_state() -> Dict[str, Any]: + """State endpoint - returns current environment state.""" + state = self.env.state + return asdict(state) + + @app.get("/health") + async def health() -> Dict[str, str]: + """Health check endpoint.""" + return {"status": "healthy"} + + + def _deserialize_action(self, action_data: Dict[str, Any]) -> Action: + """ + Convert JSON dict to Action instance. + + Args: + action_data: Dictionary containing action data + + Returns: + Action instance + + Note: + This is a simple implementation. Subclasses may need to override + for more complex deserialization logic. + """ + # Remove metadata if present (it will be set via kw_only field) + metadata = action_data.pop("metadata", {}) + action = self.action_cls(**action_data) + action.metadata = metadata + return action + + def _serialize_observation(self, observation: Observation) -> Dict[str, Any]: + """ + Convert Observation instance to JSON-compatible dict. + + Args: + observation: Observation instance + + Returns: + Dictionary compatible with HTTPEnvClient._parse_result() + + The format matches what HTTPEnvClient expects: + { + "observation": {...}, # Observation fields + "reward": float | None, + "done": bool, + } + """ + obs_dict = asdict(observation) + + # Extract reward and done (these are part of StepResult on client side) + reward = obs_dict.pop("reward", None) + done = obs_dict.pop("done", False) + obs_dict.pop("metadata", None) # Remove metadata from observation + + # Return in HTTPEnvClient expected format + return { + "observation": obs_dict, + "reward": reward, + "done": done, + } + +def create_app( + env: Environment, + action_cls: Type[Action], + observation_cls: Type[Observation], + env_name: Optional[str] = None, +) -> Any: + """ + Create a FastAPI application with or without web interface. + + This function creates a FastAPI app with the web interface enabled by default, + including README integration for better user experience. + + Args: + env: The Environment instance to serve + action_cls: The Action subclass this environment expects + observation_cls: The Observation subclass this environment returns + env_name: Optional environment name for README loading + + Returns: + FastAPI application instance with or without web interface and README integration + """ + # Check if web interface should be enabled + # This can be controlled via environment variable or build argument + enable_web = ( + os.getenv("ENABLE_WEB_INTERFACE", "false").lower() in ("true", "1", "yes") + ) + + if enable_web: + # Import web interface only when needed + from .web_interface import create_web_interface_app + return create_web_interface_app(env, action_cls, observation_cls, env_name) + else: + # Use standard FastAPI app without web interface + return create_fastapi_app(env, action_cls, observation_cls) + + +def create_fastapi_app( + env: Environment, + action_cls: Type[Action], + observation_cls: Type[Observation], +) -> Any: + """ + Create a FastAPI application with routes for the given environment. + + Args: + env: The Environment instance to serve + action_cls: The Action subclass this environment expects + observation_cls: The Observation subclass this environment returns + + Returns: + FastAPI application instance with routes registered + + Example: + >>> from envs.coding_env.server import CodeExecutionEnvironment + >>> from envs.coding_env.models import CodeAction, CodeObservation + >>> + >>> env = CodeExecutionEnvironment() + >>> app = create_fastapi_app(env, CodeAction, CodeObservation) + >>> + >>> # Run with: uvicorn module:app --host 0.0.0.0 --port 8000 + """ + try: + from fastapi import FastAPI + except ImportError: + raise ImportError( + "FastAPI is required. Install with: pip install fastapi uvicorn" + ) + + app = FastAPI(title="Environment HTTP Server") + server = HTTPEnvServer(env, action_cls, observation_cls) + server.register_routes(app) + return app diff --git a/src/core/env_server/interfaces.py b/src/core/env_server/interfaces.py new file mode 100644 index 0000000000000000000000000000000000000000..caa2d76db1b079f0c15277d1dac2db44e4173ac9 --- /dev/null +++ b/src/core/env_server/interfaces.py @@ -0,0 +1,118 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from abc import ABC, abstractmethod +from typing import Any, Protocol, TypedDict + +from .types import Action, Observation, State + + +class Message(TypedDict): + """A message in a conversation. + + Compatible with Huggingface chat template format. + """ + + role: str + content: str + + +class ModelTokenizer(Protocol): + """Protocol for tokenizers that support chat templates. + + This protocol defines the interface that tokenizers must implement + to work with chat-based environments. It's compatible with + Huggingface transformers tokenizers. + """ + + def apply_chat_template( + self, + conversation: list[Message], + tokenize: bool = True, + return_tensors: str | None = None, + **kwargs: Any, + ) -> Any: + """Apply a chat template to format and optionally tokenize a conversation. + + Args: + conversation: List of message dictionaries with 'role' and 'content' + tokenize: Whether to tokenize the output + return_tensors: Format for returned tensors ('pt' for PyTorch) + **kwargs: Additional arguments + + Returns: + Formatted and optionally tokenized conversation + """ + ... + + def decode( + self, token_ids: Any, skip_special_tokens: bool = False, **kwargs: Any + ) -> str: + """Decode token IDs back to text. + + Args: + token_ids: Token IDs to decode + skip_special_tokens: Whether to skip special tokens in output + **kwargs: Additional arguments + + Returns: + Decoded text string + """ + ... + + +class Transform(ABC): + """Transform observations to add rewards, metrics, or other modifications. + + Transforms follow the TorchRL pattern where they take an observation + and return a (potentially modified) observation. This allows for + flexible reward computation and observation augmentation. + """ + + @abstractmethod + def __call__(self, observation: Observation) -> Observation: + """Transform an observation. + + Args: + observation: The input observation + + Returns: + The transformed observation + """ + pass + + +class Environment(ABC): + """Base class for all environment servers following Gym/Gymnasium API. + + Args: + transform: Optional transform to apply to observations + """ + + def __init__(self, transform: Transform | None = None): + self.transform = transform + + @abstractmethod + def reset(self) -> Observation: + """Reset the environment and return initial observation.""" + pass + + @abstractmethod + def step(self, action: Action) -> Observation: + """Take a step in the environment.""" + pass + + @property + @abstractmethod + def state(self) -> State: + """Get the current environment state.""" + pass + + def _apply_transform(self, observation: Observation) -> Observation: + """Apply transform if one is provided.""" + if self.transform is not None: + return self.transform(observation) + return observation diff --git a/src/core/env_server/types.py b/src/core/env_server/types.py new file mode 100644 index 0000000000000000000000000000000000000000..70da9f3ca2257ba6c27fb95a58db0a6ec37ccf3e --- /dev/null +++ b/src/core/env_server/types.py @@ -0,0 +1,57 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +from dataclasses import dataclass, field +from typing import Any, Dict, List, Optional, Union + + +# Type aliases +Scalar = Union[int, float, bool] + + +@dataclass(kw_only=True) +class Action: + """Base class for all environment actions.""" + + metadata: Dict[str, Any] = field(default_factory=dict) + + +@dataclass(kw_only=True) +class Observation: + """Base class for all environment observations.""" + + done: bool = False + reward: Union[bool, int, float, None] = None + metadata: Dict[str, Any] = field(default_factory=dict) + + +@dataclass +class State: + """Base class for environment state.""" + + episode_id: Optional[str] = None + step_count: int = 0 + + +@dataclass +class CodeExecResult: + """Result of code execution containing stdout, stderr, and exit code.""" + + stdout: str + stderr: str + exit_code: int + + +@dataclass +class EnvironmentMetadata: + """Metadata about an environment for documentation and UI purposes.""" + + name: str + description: str + readme_content: Optional[str] = None + version: Optional[str] = None + author: Optional[str] = None + documentation_url: Optional[str] = None diff --git a/src/core/env_server/web_interface.py b/src/core/env_server/web_interface.py new file mode 100644 index 0000000000000000000000000000000000000000..3c36aa1de336f86161429f8a382054ec89db1912 --- /dev/null +++ b/src/core/env_server/web_interface.py @@ -0,0 +1,1613 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Web interface for OpenEnv environments. + +This module provides a web-based interface for interacting with OpenEnv environments, +including a two-pane layout for HumanAgent interaction and state observation. +""" + +from __future__ import annotations + +import json +import time +from dataclasses import asdict, dataclass +from typing import Any, Dict, List, Optional, Type +from datetime import datetime + +from fastapi import FastAPI, WebSocket, WebSocketDisconnect, Request +from fastapi.responses import HTMLResponse, FileResponse +from fastapi.staticfiles import StaticFiles +from pydantic import BaseModel + +from .interfaces import Environment +from .types import Action, Observation, State, EnvironmentMetadata + + +def load_environment_metadata(env: Environment, env_name: Optional[str] = None) -> EnvironmentMetadata: + """ + Load environment metadata including README content. + + Args: + env: The environment instance + env_name: Optional environment name for README file lookup + + Returns: + EnvironmentMetadata with loaded information + """ + # Try to get metadata from environment if it has a method for it + if hasattr(env, 'get_metadata'): + return env.get_metadata() + + # Default metadata + metadata = EnvironmentMetadata( + name=env_name or env.__class__.__name__, + description=f"{env.__class__.__name__} environment", + version="1.0.0" + ) + + # Try to load README from file system + readme_content = _load_readme_from_filesystem(env_name) + if readme_content: + metadata.readme_content = readme_content + + return metadata + + +def _load_readme_from_filesystem(env_name: Optional[str]) -> Optional[str]: + """ + Load README content from the filesystem. + + Tries multiple locations: + 1. Container filesystem: /app/README.md + 2. Local development: src/envs/{env_name}/README.md + 3. Environment variable: ENV_README_PATH + """ + import os + from pathlib import Path + + # Try container filesystem first + container_readme = Path("/app/README.md") + if container_readme.exists(): + try: + return container_readme.read_text(encoding='utf-8') + except Exception: + pass + + # Try environment variable path + custom_path = os.environ.get("ENV_README_PATH") + if custom_path and Path(custom_path).exists(): + try: + return Path(custom_path).read_text(encoding='utf-8') + except Exception: + pass + + # Try local development path + if env_name: + local_readme = Path(f"src/envs/{env_name}/README.md") + if local_readme.exists(): + try: + return local_readme.read_text(encoding='utf-8') + except Exception: + pass + + return None + + +@dataclass +class ActionLog: + """Log entry for an action taken.""" + timestamp: str + action: Dict[str, Any] + observation: Dict[str, Any] + reward: Optional[float] + done: bool + step_count: int + + +@dataclass +class EpisodeState: + """Current episode state for the web interface.""" + episode_id: Optional[str] + step_count: int + current_observation: Optional[Dict[str, Any]] + action_logs: List[ActionLog] + is_reset: bool = True + + +class WebInterfaceManager: + """Manages the web interface for an environment.""" + + def __init__( + self, + env: Environment, + action_cls: Type[Action], + observation_cls: Type[Observation], + metadata: Optional[EnvironmentMetadata] = None, + ): + self.env = env + self.action_cls = action_cls + self.observation_cls = observation_cls + self.metadata = metadata or EnvironmentMetadata( + name=env.__class__.__name__, + description=f"{env.__class__.__name__} environment" + ) + self.episode_state = EpisodeState( + episode_id=None, + step_count=0, + current_observation=None, + action_logs=[] + ) + self.connected_clients: List[WebSocket] = [] + + async def connect_websocket(self, websocket: WebSocket): + """Connect a new WebSocket client.""" + await websocket.accept() + self.connected_clients.append(websocket) + + # Send current state to the new client + await self._send_state_update() + + async def disconnect_websocket(self, websocket: WebSocket): + """Disconnect a WebSocket client.""" + if websocket in self.connected_clients: + self.connected_clients.remove(websocket) + + async def _send_state_update(self): + """Send current state to all connected clients.""" + if not self.connected_clients: + return + + state_data = { + "type": "state_update", + "episode_state": asdict(self.episode_state) + } + + # Send to all connected clients + disconnected_clients = [] + for client in self.connected_clients: + try: + await client.send_text(json.dumps(state_data)) + except: + disconnected_clients.append(client) + + # Remove disconnected clients + for client in disconnected_clients: + self.connected_clients.remove(client) + + async def reset_environment(self) -> Dict[str, Any]: + """Reset the environment and update state.""" + observation = self.env.reset() + state = self.env.state + + # Update episode state + self.episode_state.episode_id = state.episode_id + self.episode_state.step_count = 0 + self.episode_state.current_observation = asdict(observation) + self.episode_state.action_logs = [] + self.episode_state.is_reset = True + + # Send state update + await self._send_state_update() + + return { + "observation": asdict(observation), + "reward": observation.reward, + "done": observation.done, + } + + async def step_environment(self, action_data: Dict[str, Any]) -> Dict[str, Any]: + """Execute a step in the environment and update state.""" + # Deserialize action + action = self._deserialize_action(action_data) + + # Execute step + observation = self.env.step(action) + state = self.env.state + + # Create action log + action_log = ActionLog( + timestamp=datetime.now().isoformat(), + action=asdict(action), + observation=asdict(observation), + reward=observation.reward, + done=observation.done, + step_count=state.step_count + ) + + # Update episode state + self.episode_state.episode_id = state.episode_id + self.episode_state.step_count = state.step_count + self.episode_state.current_observation = asdict(observation) + self.episode_state.action_logs.append(action_log) + self.episode_state.is_reset = False + + # Send state update + await self._send_state_update() + + return { + "observation": asdict(observation), + "reward": observation.reward, + "done": observation.done, + } + + def get_state(self) -> Dict[str, Any]: + """Get current environment state.""" + state = self.env.state + return asdict(state) + + def _deserialize_action(self, action_data: Dict[str, Any]) -> Action: + """Convert JSON dict to Action instance.""" + metadata = action_data.pop("metadata", {}) + + # Handle tensor fields that come from JSON as lists + processed_data = {} + for key, value in action_data.items(): + if key == "tokens" and isinstance(value, (list, str)): + # Convert list or string to tensor + if isinstance(value, str): + # If it's a string, try to parse it as a list of numbers + try: + import json + value = json.loads(value) + except: + # If parsing fails, treat as empty list + value = [] + if isinstance(value, list): + import torch + processed_data[key] = torch.tensor(value, dtype=torch.long) + else: + processed_data[key] = value + elif key == "action_id" and isinstance(value, str): + # Convert action_id from string to int + try: + processed_data[key] = int(value) + except ValueError: + # If conversion fails, keep original value + processed_data[key] = value + else: + processed_data[key] = value + + action = self.action_cls(**processed_data) + action.metadata = metadata + return action + + +def create_web_interface_app( + env: Environment, + action_cls: Type[Action], + observation_cls: Type[Observation], + env_name: Optional[str] = None, +) -> FastAPI: + """ + Create a FastAPI application with web interface for the given environment. + + Args: + env: The Environment instance to serve + action_cls: The Action subclass this environment expects + observation_cls: The Observation subclass this environment returns + env_name: Optional environment name for README loading + + Returns: + FastAPI application instance with web interface + """ + from .http_server import create_fastapi_app + + # Create the base environment app + app = create_fastapi_app(env, action_cls, observation_cls) + + # Load environment metadata + metadata = load_environment_metadata(env, env_name) + + # Create web interface manager + web_manager = WebInterfaceManager(env, action_cls, observation_cls, metadata) + + # Add web interface routes + @app.get("/web", response_class=HTMLResponse) + async def web_interface(): + """Serve the web interface.""" + return get_web_interface_html(action_cls, web_manager.metadata) + + @app.get("/web/metadata") + async def web_metadata(): + """Get environment metadata.""" + return asdict(web_manager.metadata) + + @app.websocket("/ws") + async def websocket_endpoint(websocket: WebSocket): + """WebSocket endpoint for real-time updates.""" + await web_manager.connect_websocket(websocket) + try: + while True: + # Keep connection alive + await websocket.receive_text() + except WebSocketDisconnect: + await web_manager.disconnect_websocket(websocket) + + @app.post("/web/reset") + async def web_reset(): + """Reset endpoint for web interface.""" + return await web_manager.reset_environment() + + @app.post("/web/step") + async def web_step(request: Dict[str, Any]): + """Step endpoint for web interface.""" + # Check if this is a message-based request (chat environment) + if "message" in request: + message = request["message"] + # Convert message to action using the environment's message_to_action method + action = web_manager.env.message_to_action(message) + action_data = {"tokens": action.tokens.tolist()} + else: + action_data = request.get("action", {}) + + return await web_manager.step_environment(action_data) + + @app.get("/web/state") + async def web_state(): + """State endpoint for web interface.""" + return web_manager.get_state() + + return app + + +def get_web_interface_html(action_cls: Type[Action], metadata: Optional[EnvironmentMetadata] = None) -> str: + """Generate the HTML for the web interface.""" + + # Check if this is a chat environment by looking for tokens field + is_chat_env = False + if hasattr(action_cls, '__dataclass_fields__'): + for field_name, field_info in action_cls.__dataclass_fields__.items(): + if field_name == 'tokens' and hasattr(field_info.type, '__name__') and 'Tensor' in field_info.type.__name__: + is_chat_env = True + break + + # Get action fields for dynamic form generation with enhanced metadata + action_fields = _extract_action_fields(action_cls) + + return f""" + + + + + + OpenEnv Web Interface + + + +
+ +
+
+ + HumanAgent Interface +
+
+ + {_generate_instructions_section(metadata)} + + + {_generate_action_interface(action_fields, is_chat_env)} + + +
+ + +
+ + +
+

Current State

+
+
+ Status: + Not initialized +
+
+ Episode ID: + - +
+
+ Step Count: + 0 +
+
+
+
+
+ + +
+
+ State Observer +
+
+ +
+

Current Observation

+
+ No observation yet +
+
+ + +
+

Action History

+
+ No actions taken yet +
+
+
+
+
+ + + + + """.replace('{_generate_action_form_fields(action_fields)}', _generate_action_form_fields(action_fields)) + + +def _generate_instructions_section(metadata: Optional[EnvironmentMetadata]) -> str: + """Generate the instructions section with environment documentation.""" + if not metadata or not metadata.readme_content: + return '' + + # Convert markdown to HTML (basic conversion) + import re + html_content = _markdown_to_html(metadata.readme_content) + + return f''' + +
+
+

{metadata.name}

+ +
+
+
+ {html_content} +
+
+
+ ''' + + +def _extract_action_fields(action_cls: Type[Action]) -> List[Dict[str, Any]]: + """Extract enhanced field metadata from Action class for form generation.""" + import typing + from typing import get_origin, get_args + + action_fields = [] + if not hasattr(action_cls, '__dataclass_fields__'): + return action_fields + + for field_name, field_info in action_cls.__dataclass_fields__.items(): + if field_name == 'metadata': + continue + + field_type = field_info.type + field_metadata = _extract_field_metadata(field_name, field_info) + + # Determine input type based on field type + input_type = _determine_input_type(field_type) + + # Check if field is required + is_required = field_info.default is field_info.default_factory + + action_fields.append({ + 'name': field_name, + 'type': input_type, + 'required': is_required, + 'description': field_metadata.get('description', ''), + 'default_value': field_metadata.get('default_value'), + 'choices': field_metadata.get('choices', []), + 'min_value': field_metadata.get('min_value'), + 'max_value': field_metadata.get('max_value'), + 'placeholder': field_metadata.get('placeholder', ''), + 'help_text': field_metadata.get('help_text', ''), + }) + + return action_fields + + +def _extract_field_metadata(field_name: str, field_info) -> Dict[str, Any]: + """Extract metadata from dataclass field including docstring and type hints.""" + import typing + from typing import get_origin, get_args, Literal, Union, Optional + + metadata = {} + + # Extract description from field docstring or annotation + if hasattr(field_info, 'metadata') and field_info.metadata: + # Check for custom metadata + for meta in field_info.metadata: + if isinstance(meta, dict): + metadata.update(meta) + + # Extract type information + field_type = field_info.type + origin = get_origin(field_type) + + # Handle Literal types for dropdown choices + if origin is Literal: + args = get_args(field_type) + metadata['choices'] = list(args) + + # Handle Optional types + if origin is Union: + args = get_args(field_type) + if len(args) == 2 and type(None) in args: + # This is Optional[SomeType] + non_none_type = args[0] if args[1] is type(None) else args[1] + metadata['optional'] = True + # Recursively check the non-None type for choices + if get_origin(non_none_type) is Literal: + metadata['choices'] = list(get_args(non_none_type)) + else: + # Regular Union type + metadata['choices'] = [str(arg) for arg in args if arg is not type(None)] + + # Handle numeric constraints + if field_type in (int, float): + # Check for common constraint patterns in field name + if 'count' in field_name.lower() or 'num' in field_name.lower(): + metadata['min_value'] = 0 + if 'id' in field_name.lower(): + metadata['min_value'] = 0 + + # Generate placeholder text + if 'message' in field_name.lower(): + metadata['placeholder'] = f'Enter {field_name.replace("_", " ")}...' + elif 'code' in field_name.lower(): + metadata['placeholder'] = 'Enter Python code here...' + elif 'tokens' in field_name.lower(): + metadata['placeholder'] = 'Enter comma-separated token IDs (e.g., 1,2,3,4,5)' + else: + metadata['placeholder'] = f'Enter {field_name.replace("_", " ")}...' + + # Generate help text based on field name and type + if 'action_id' in field_name.lower(): + metadata['help_text'] = 'The action ID to execute in the environment' + elif 'game_name' in field_name.lower(): + metadata['help_text'] = 'Name of the game or environment' + elif 'tokens' in field_name.lower(): + metadata['help_text'] = 'Token IDs as a comma-separated list of integers' + elif 'code' in field_name.lower(): + metadata['help_text'] = 'Python code to execute in the environment' + elif 'message' in field_name.lower(): + metadata['help_text'] = 'Text message to send' + + return metadata + + +def _determine_input_type(field_type) -> str: + """Determine the appropriate HTML input type for a field type.""" + import typing + from typing import get_origin, get_args, Literal, Union + + # Handle direct types + if field_type == str: + return "text" + elif field_type == int: + return "number" + elif field_type == float: + return "number" + elif field_type == bool: + return "checkbox" + + # Handle complex types + origin = get_origin(field_type) + + if origin is Literal: + return "select" + elif origin is Union: + args = get_args(field_type) + if len(args) == 2 and type(None) in args: + # Optional type - use the non-None type + non_none_type = args[0] if args[1] is type(None) else args[1] + return _determine_input_type(non_none_type) + elif all(isinstance(arg, str) for arg in args if arg is not type(None)): + return "select" + else: + return "text" + elif hasattr(field_type, '__name__') and 'Tensor' in field_type.__name__: + return "tensor" + else: + return "text" + + +def _markdown_to_html(markdown: str) -> str: + """Convert basic markdown to HTML for README display.""" + import html + import re + + # Escape HTML first + html_content = html.escape(markdown) + + # Convert headers + html_content = re.sub(r'^# (.*?)$', r'

\1

', html_content, flags=re.MULTILINE) + html_content = re.sub(r'^## (.*?)$', r'

\1

', html_content, flags=re.MULTILINE) + html_content = re.sub(r'^### (.*?)$', r'

\1

', html_content, flags=re.MULTILINE) + + # Convert code blocks + html_content = re.sub(r'```(.*?)\n(.*?)\n```', r'
\2
', html_content, flags=re.DOTALL) + html_content = re.sub(r'`([^`]+)`', r'\1', html_content) + + # Convert bold and italic + html_content = re.sub(r'\*\*(.*?)\*\*', r'\1', html_content) + html_content = re.sub(r'\*(.*?)\*', r'\1', html_content) + + # Convert lists + html_content = re.sub(r'^- (.*?)$', r'
  • \1
    • ', html_content, flags=re.MULTILINE) + html_content = re.sub(r'(
  • .*
    • )', r'', html_content, flags=re.DOTALL) + + # Convert line breaks + html_content = html_content.replace('\n', '
    ') + + return html_content + + +def _generate_action_interface(action_fields: List[Dict[str, Any]], is_chat_env: bool) -> str: + """Generate either a chat interface or action form based on environment type.""" + if is_chat_env: + return _generate_chat_interface() + else: + return _generate_action_form(action_fields) + +def _generate_chat_interface() -> str: + """Generate a chat-style interface for chat environments.""" + return ''' + +
    +

    Chat Interface

    +
    +
    +
    System
    +
    Chat environment ready. Send a message to start the conversation.
    +
    +
    +
    +
    + + +
    +
    + + +
    +
    +
    + ''' + +def _generate_action_form(action_fields: List[Dict[str, Any]]) -> str: + """Generate a traditional action form for non-chat environments.""" + return f''' + +
    +

    Take Action

    +
    + {_generate_action_form_fields(action_fields)} + +
    +
    + ''' + +def _generate_action_form_fields(action_fields: List[Dict[str, Any]]) -> str: + """Generate HTML form fields for action input with enhanced metadata.""" + if not action_fields: + return '

    No action fields available

    ' + + fields_html = [] + for field in action_fields: + field_html = _generate_single_field(field) + fields_html.append(field_html) + + return '\n'.join(fields_html) + + +def _generate_single_field(field: Dict[str, Any]) -> str: + """Generate HTML for a single form field with enhanced metadata.""" + field_name = field['name'] + field_type = field['type'] + required = field['required'] + placeholder = field.get('placeholder', '') + help_text = field.get('help_text', '') + choices = field.get('choices', []) + min_value = field.get('min_value') + max_value = field.get('max_value') + default_value = field.get('default_value') + + # Build label with required indicator + label_text = field_name.replace('_', ' ').title() + if required: + label_text += ' *' + + # Build input attributes + input_attrs = [] + if required: + input_attrs.append('required') + if placeholder: + input_attrs.append(f'placeholder="{placeholder}"') + if min_value is not None: + input_attrs.append(f'min="{min_value}"') + if max_value is not None: + input_attrs.append(f'max="{max_value}"') + if default_value is not None: + input_attrs.append(f'value="{default_value}"') + + attrs_str = ' '.join(input_attrs) + + if field_type == 'checkbox': + return f''' +
    + + {f'{help_text}' if help_text else ''} +
    + ''' + + elif field_type == 'select': + options_html = [] + if not required: + options_html.append(f'') + + for choice in choices: + selected = 'selected' if str(choice) == str(default_value) else '' + options_html.append(f'') + + return f''' +
    + + + {f'{help_text}' if help_text else ''} +
    + ''' + + elif field_type == 'tensor': + return f''' +
    + + + {help_text or 'Enter token IDs as comma-separated integers (e.g., 1,2,3,4,5)'} +
    + ''' + + elif field_type == 'text' and ('message' in field_name.lower() or 'code' in field_name.lower()): + return f''' +
    + + + {f'{help_text}' if help_text else ''} +
    + ''' + + else: + return f''' +
    + + + {f'{help_text}' if help_text else ''} +
    + ''' diff --git a/src/core/http_env_client.py b/src/core/http_env_client.py new file mode 100644 index 0000000000000000000000000000000000000000..b304e08899a8328d18fda7d5b61c2fc69bce3f39 --- /dev/null +++ b/src/core/http_env_client.py @@ -0,0 +1,185 @@ +""" +core/runner_env.py +Minimal HTTP-based environment client. +- Talks to a single env worker exposing: POST /reset, POST /step + +Future hooks (commented below) for: +- episode_id, seed on reset +- request_id on step +- custom headers (auth/trace) +""" + +from __future__ import annotations + +from abc import ABC, abstractmethod +from typing import Any, Dict, Generic, Optional, Type, TYPE_CHECKING, TypeVar + +import requests + +from .client_types import StepResult +from .containers.runtime import LocalDockerProvider + +if TYPE_CHECKING: + from .containers.runtime import ContainerProvider + +ActT = TypeVar("ActT") +ObsT = TypeVar("ObsT") +EnvClientT = TypeVar("EnvClientT", bound="HTTPEnvClient") + + +class HTTPEnvClient(ABC, Generic[ActT, ObsT]): + def __init__( + self, + base_url: str, + request_timeout_s: float = 15.0, + default_headers: Optional[Dict[str, str]] = None, + provider: Optional["ContainerProvider"] = None, + ): + self._base = base_url.rstrip("/") + self._timeout = float(request_timeout_s) + self._http = requests.Session() + self._headers = default_headers or {} + self._provider = provider + + @classmethod + def from_docker_image( + cls: Type[EnvClientT], + image: str, + provider: Optional["ContainerProvider"] = None, + **kwargs: Any, + ) -> EnvClientT: + """ + Create an environment client by spinning up a Docker container locally. + + This is a development utility that: + 1. Starts a Docker container from the specified image + 2. Waits for the server to be ready + 3. Creates and returns a client instance connected to the container + + Note: The container lifecycle management is left to the user or higher-level + orchestration. The container will keep running until manually stopped. + + Args: + image: Docker image name to run (e.g., "echo-env:latest") + provider: Container provider to use (defaults to LocalDockerProvider) + **kwargs: Additional arguments to pass to provider.start_container() + (e.g., env_vars, port) + + Returns: + An instance of the client class connected to the running container + + Example: + >>> from envs.coding_env.client import CodingEnv + >>> from envs.coding_env.models import CodeAction + >>> + >>> # Create environment from image + >>> env = CodingEnv.from_docker_image("coding-env:latest") + >>> + >>> # Create environment with custom env vars + >>> env = CodingEnv.from_docker_image( + ... "coding-env:latest", + ... env_vars={"MY_VAR": "value"} + ... ) + >>> + >>> # Use the environment + >>> result = env.reset() + >>> print(result.observation) + >>> + >>> step_result = env.step(CodeAction(code="print('hello')")) + >>> print(step_result.observation.stdout) + >>> + >>> # Cleanup (optional) + >>> env.close() + """ + + # Use default provider if none provided + if provider is None: + provider = LocalDockerProvider() + + # 1. Start container with optional kwargs (e.g., env_vars, port) + base_url = provider.start_container(image, **kwargs) + + # 2. Wait for server to be ready + provider.wait_for_ready(base_url) + + # 3. Create and return client instance with provider reference + return cls(base_url=base_url, provider=provider) + + @abstractmethod + def _step_payload(self, action: ActT) -> dict: + """Convert an Action object to the JSON body expected by the env server.""" + raise NotImplementedError + + @abstractmethod + def _parse_result(self, payload: dict) -> StepResult[ObsT]: + """Convert a JSON response from the env server to StepResult[ObsT].""" + raise NotImplementedError + + @abstractmethod + def _parse_state(self, payload: dict) -> Any: + """Convert a JSON response from the state endpoint to a State object.""" + raise NotImplementedError + + # ---------- Environment Server Interface Methods ---------- + def reset(self) -> StepResult[ObsT]: + body: Dict[str, Any] = {} + # TODO: later: + # body["seed"] = seed + # body["episode_id"] = episode_id + r = self._http.post( + f"{self._base}/reset", + json=body, + headers=self._headers, + timeout=self._timeout, + ) + r.raise_for_status() + return self._parse_result(r.json()) + + def step(self, action: ActT) -> StepResult[ObsT]: + body: Dict[str, Any] = { + "action": self._step_payload(action), + "timeout_s": int(self._timeout), + } + # TODO: later: + # body["request_id"] = str(uuid.uuid4()) + # body["episode_id"] = current_episode_id + r = self._http.post( + f"{self._base}/step", + json=body, + headers=self._headers, + timeout=self._timeout, + ) + r.raise_for_status() + return self._parse_result(r.json()) + + def state(self) -> Any: + """ + Get the current environment state from the server. + + Returns: + State object with environment state information (e.g., episode_id, step_count) + + Example: + >>> client = EchoEnv.from_docker_image("echo-env:latest") + >>> result = client.reset() + >>> state = client.state() + >>> print(state.episode_id) + >>> print(state.step_count) + """ + r = self._http.get( + f"{self._base}/state", + headers=self._headers, + timeout=self._timeout, + ) + r.raise_for_status() + return self._parse_state(r.json()) + + def close(self) -> None: + """ + Close the environment and clean up resources. + + If this client was created via from_docker_image(), this will stop + and remove the associated container. + """ + if self._provider is not None: + self._provider.stop_container() diff --git a/src/core/pyproject.toml b/src/core/pyproject.toml new file mode 100644 index 0000000000000000000000000000000000000000..32602f58cb6f066e6874f977f1838483a2ea931b --- /dev/null +++ b/src/core/pyproject.toml @@ -0,0 +1,46 @@ +[build-system] +requires = ["setuptools>=45", "wheel"] +build-backend = "setuptools.build_meta" + +[project] +name = "openenv-core" +version = "0.1.0" +description = "Core components for OpenEnv - HTTP-based agentic environments" +readme = "README.md" +requires-python = ">=3.8" +license = {text = "BSD-3-Clause"} +authors = [ + {name = "Meta Platforms, Inc.", email = "opensource@meta.com"} +] +keywords = ["environment", "agent", "http", "docker", "fastapi"] + +dependencies = [ + "requests>=2.25.0", + "fastapi>=0.104.0", + "uvicorn>=0.24.0", +] + +[project.optional-dependencies] +dev = [ + "pytest>=7.0.0", + "black>=23.0.0", + "ruff>=0.1.0", + "mypy>=1.0.0", +] + +[project.urls] +Homepage = "https://github.com/facebookresearch/OpenEnv" +Repository = "https://github.com/facebookresearch/OpenEnv" +Documentation = "https://github.com/facebookresearch/OpenEnv/blob/main/README.md" +"Bug Tracker" = "https://github.com/facebookresearch/OpenEnv/issues" + +[tool.setuptools] +py-modules = ["openenv_core.__init__", "openenv_core.http_env_client", "openenv_core.client_types"] +packages = [ + "openenv_core", + "openenv_core.containers", + "openenv_core.containers.runtime", + "openenv_core.env_server", + "openenv_core.tools" +] +package-dir = {"openenv_core" = "."} diff --git a/src/core/tools/__init__.py b/src/core/tools/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..034e7f0686b81bcdda4f30cf1e32b86cd4fa2bf5 --- /dev/null +++ b/src/core/tools/__init__.py @@ -0,0 +1,16 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Core tools for code execution and other utilities.""" + +from .git_server_client import GitServerClient, RepoInfo +from .local_python_executor import PyExecutor + +__all__ = [ + "PyExecutor", + "GitServerClient", + "RepoInfo", +] \ No newline at end of file diff --git a/src/core/tools/git_server_client.py b/src/core/tools/git_server_client.py new file mode 100644 index 0000000000000000000000000000000000000000..31b1ed4c1806287cffe43ad2f93d24f44329d575 --- /dev/null +++ b/src/core/tools/git_server_client.py @@ -0,0 +1,362 @@ +#!/usr/bin/env python3 +""" +Git Server Client for connecting to external Gitea instance. + +This module provides a lightweight client for interacting with a shared +Gitea service, optimized for task-based isolation where multiple environment +instances share the same Gitea server but have isolated workspaces. +""" + +import json +import os +import shutil +import subprocess +import time +from dataclasses import dataclass +from pathlib import Path +from urllib.parse import urlparse + + +@dataclass +class RepoInfo: + """Information about a repository.""" + + name: str + url: str + commit: str + clone_url: str + + +class GitServerClient: + """ + Client for connecting to an external Gitea server. + + This client is optimized for task-based isolation where: + - Multiple tasks share the same Gitea instance + - Each task has its own isolated workspace + - Fast reset() via git operations (no server restart) + - Repos are pre-migrated to Gitea once + + Args: + gitea_url: URL of the Gitea server (e.g., "http://gitea:3000") + username: Gitea username for authentication + password: Gitea password for authentication + workspace_dir: Local workspace directory for cloning repos + + Example: + >>> # Connect to shared Gitea (credentials from environment) + >>> import os + >>> client = GitServerClient( + ... gitea_url=os.getenv("GITEA_URL"), + ... username=os.getenv("GITEA_USERNAME"), + ... password=os.getenv("GITEA_PASSWORD") + ... ) + >>> client.wait_for_ready() + >>> # Clone repo to workspace + >>> path = client.clone_to_workspace("my-repo", commit="abc123") + >>> # Fast reset to base state + >>> client.reset_workspace("my-repo", commit="abc123") + """ + + def __init__( + self, + gitea_url: str, + username: str, + password: str, + workspace_dir: str = "/workspace", + ): + """Initialize Git Server Client.""" + self.gitea_url = gitea_url.rstrip("/") + self.username = username + self.password = password + self.workspace_dir = Path(workspace_dir) + self.is_ready = False + + # Parse Gitea URL + parsed = urlparse(self.gitea_url) + self.domain = parsed.hostname or "localhost" + self.port = parsed.port or 3000 + + # Ensure workspace exists + os.makedirs(self.workspace_dir, exist_ok=True) + + # Configure git credentials + self._configure_git() + + def _configure_git(self): + """Configure git credentials for automatic authentication.""" + home_dir = Path.home() + + # Git config + git_config = f"""[user] + name = {self.username} + email = {self.username}@local.env +[init] + defaultBranch = main +[credential] + helper = store +""" + gitconfig_path = home_dir / ".gitconfig" + gitconfig_path.write_text(git_config) + + # Git credentials + git_credentials = f"http://{self.username}:{self.password}@{self.domain}:{self.port}\n" + gitcreds_path = home_dir / ".git-credentials" + gitcreds_path.write_text(git_credentials) + gitcreds_path.chmod(0o600) + + def wait_for_ready(self, timeout: int = 30) -> bool: + """ + Wait for Gitea server to be ready. + + Args: + timeout: Maximum seconds to wait + + Returns: + True if server is ready, False otherwise + """ + start_time = time.time() + while time.time() - start_time < timeout: + try: + result = subprocess.run( + ["curl", "-sf", f"{self.gitea_url}/"], + capture_output=True, + timeout=5, + ) + if result.returncode == 0: + self.is_ready = True + return True + except subprocess.TimeoutExpired: + pass + except Exception: + pass + + time.sleep(1) + + return False + + def list_repositories(self) -> list[dict[str, str]]: + """ + List all repositories in Gitea. + + Returns: + List of repository information dictionaries + """ + if not self.is_ready: + raise RuntimeError("Gitea server is not ready") + + result = subprocess.run( + [ + "curl", + "-s", + f"{self.gitea_url}/api/v1/user/repos", + "-u", + f"{self.username}:{self.password}", + ], + capture_output=True, + text=True, + ) + + if result.returncode != 0: + return [] + + try: + repos = json.loads(result.stdout) + return [ + { + "name": repo["name"], + "full_name": repo["full_name"], + "clone_url": repo["clone_url"], + "description": repo.get("description", ""), + } + for repo in repos + ] + except (json.JSONDecodeError, KeyError): + return [] + + def clone_to_workspace( + self, repo_name: str, target_dir: str | None = None, commit: str = "main" + ) -> str: + """ + Clone a repository to the workspace at a specific commit. + + This creates a fresh clone optimized for task isolation. + + Args: + repo_name: Name of repository to clone + target_dir: Target directory name (defaults to repo_name) + commit: Commit hash or branch to checkout + + Returns: + Path to cloned repository + + Raises: + RuntimeError: If clone fails + """ + if not self.is_ready: + raise RuntimeError("Gitea server is not ready") + + target_dir = target_dir or repo_name + target_path = self.workspace_dir / target_dir + + # Remove existing directory if present + if target_path.exists(): + shutil.rmtree(target_path) + + clone_url = f"{self.gitea_url}/{self.username}/{repo_name}.git" + + # Clone repository + result = subprocess.run( + ["git", "clone", clone_url, str(target_path)], + capture_output=True, + text=True, + ) + + if result.returncode != 0: + raise RuntimeError(f"Clone failed: {result.stderr}") + + # Checkout specific commit + if commit != "main": + result = subprocess.run( + ["git", "checkout", commit], + cwd=str(target_path), + capture_output=True, + text=True, + ) + + if result.returncode != 0: + raise RuntimeError(f"Checkout failed: {result.stderr}") + + return str(target_path) + + def reset_workspace(self, repo_name: str, commit: str = "main") -> bool: + """ + Fast reset of workspace to base state (optimized for task resets). + + This is much faster than re-cloning. It: + 1. Checks out the target commit + 2. Resets to that commit (hard) + 3. Cleans untracked files + + Args: + repo_name: Name of repository (directory in workspace) + commit: Commit hash or branch to reset to + + Returns: + True if reset successful + + Raises: + RuntimeError: If reset fails + """ + repo_path = self.workspace_dir / repo_name + + if not repo_path.exists(): + raise RuntimeError(f"Repository not found in workspace: {repo_name}") + + # Fetch latest (in case commit is new) + subprocess.run( + ["git", "fetch", "--all"], + cwd=str(repo_path), + capture_output=True, + ) + + # Checkout and hard reset to commit + result = subprocess.run( + ["git", "checkout", commit], + cwd=str(repo_path), + capture_output=True, + text=True, + ) + + if result.returncode != 0: + raise RuntimeError(f"Checkout failed: {result.stderr}") + + result = subprocess.run( + ["git", "reset", "--hard", f"origin/{commit}" if commit != "main" else commit], + cwd=str(repo_path), + capture_output=True, + text=True, + ) + + if result.returncode != 0: + # Try without origin/ prefix + result = subprocess.run( + ["git", "reset", "--hard", commit], + cwd=str(repo_path), + capture_output=True, + text=True, + ) + if result.returncode != 0: + raise RuntimeError(f"Reset failed: {result.stderr}") + + # Clean untracked files and directories + subprocess.run( + ["git", "clean", "-fdx"], + cwd=str(repo_path), + capture_output=True, + ) + + return True + + def execute_git_command( + self, command: str, working_dir: str = "" + ) -> tuple[int, str, str]: + """ + Execute a git command in the workspace. + + Args: + command: Git command to execute (without 'git' prefix) + working_dir: Working directory relative to workspace + + Returns: + Tuple of (exit_code, stdout, stderr) + """ + work_path = ( + self.workspace_dir / working_dir if working_dir else self.workspace_dir + ) + + if not work_path.exists(): + return (1, "", f"Working directory does not exist: {work_path}") + + # Split command safely + cmd_parts = ["git"] + command.split() + + result = subprocess.run( + cmd_parts, + cwd=str(work_path), + capture_output=True, + text=True, + ) + + return (result.returncode, result.stdout, result.stderr) + + def get_current_commit(self, repo_name: str) -> str: + """ + Get current commit hash of a workspace repository. + + Args: + repo_name: Name of repository in workspace + + Returns: + Commit hash + """ + repo_path = self.workspace_dir / repo_name + + if not repo_path.exists(): + raise RuntimeError(f"Repository not found: {repo_name}") + + result = subprocess.run( + ["git", "rev-parse", "HEAD"], + cwd=str(repo_path), + capture_output=True, + text=True, + ) + + if result.returncode != 0: + raise RuntimeError(f"Failed to get commit: {result.stderr}") + + return result.stdout.strip() + + def workspace_exists(self, repo_name: str) -> bool: + """Check if a repository exists in workspace.""" + return (self.workspace_dir / repo_name).exists() diff --git a/src/core/tools/local_python_executor.py b/src/core/tools/local_python_executor.py new file mode 100644 index 0000000000000000000000000000000000000000..ba4477d52f3f0f295ec568dc21d807a711ea2cc5 --- /dev/null +++ b/src/core/tools/local_python_executor.py @@ -0,0 +1,105 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Local Python Executor. + +This module provides functionality for executing Python code locally by wrapping +the smolagents LocalPythonExecutor. +""" + +from smolagents import LocalPythonExecutor + +from core.env_server.types import CodeExecResult + + +class PyExecutor: + """ + Wrapper around smolagents LocalPythonExecutor for executing Python code. + + This class provides a simple interface to execute Python code in a subprocess + and capture the results including stdout, stderr, and exit code. + + Args: + additional_imports: List of additional module imports to authorize. + For example: ["numpy", "pandas", "matplotlib"] + These will be added to the base authorized imports. + + Example: + >>> # Basic usage with default imports + >>> executor = PyExecutor() + >>> result = executor.run("print('Hello, World!')") + >>> print(result.stdout) # "Hello, World!\n" + >>> print(result.exit_code) # 0 + >>> + >>> # Usage with additional imports + >>> executor = PyExecutor(additional_imports=["numpy", "pandas"]) + >>> result = executor.run("import numpy as np\\nprint(np.array([1, 2, 3]))") + >>> print(result.stdout) # "[1 2 3]\n" + """ + + def __init__(self, additional_imports: list[str] | None = None): + """ + Initialize the PyExecutor with a LocalPythonExecutor instance. + + Args: + additional_imports: List of additional module names to authorize for import. + Defaults to an empty list if not provided. + """ + if additional_imports is None: + additional_imports = [] + self._executor = LocalPythonExecutor( + additional_authorized_imports=additional_imports + ) + # Initialize tools to make BASE_PYTHON_TOOLS available (including print) + self._executor.send_tools({}) + + def run(self, code: str) -> CodeExecResult: + """ + Execute Python code and return the result. + + Args: + code: Python code string to execute + + Returns: + CodeExecResult containing stdout, stderr, and exit_code + + Example: + >>> executor = PyExecutor() + >>> result = executor.run("x = 5 + 3\\nprint(x)") + >>> print(result.stdout) # "8\n" + >>> print(result.exit_code) # 0 + >>> + >>> # Error handling + >>> result = executor.run("1 / 0") + >>> print(result.exit_code) # 1 + >>> print(result.stderr) # Contains error message + """ + try: + # Execute the code using LocalPythonExecutor + # LocalPythonExecutor returns a CodeOutput object with output, logs, is_final_answer + exec_result = self._executor(code) + + # Extract the logs (which contain print outputs) as stdout + # The output field contains the return value of the code + stdout = exec_result.logs + stderr = "" + exit_code = 0 # Success + + return CodeExecResult( + stdout=stdout, + stderr=stderr, + exit_code=exit_code, + ) + + except Exception as e: + # LocalPythonExecutor raises InterpreterError for various issues + # (syntax errors, forbidden operations, runtime errors, etc.) + return CodeExecResult( + stdout="", + stderr=str(e), + exit_code=1, # Non-zero indicates error + ) diff --git a/src/envs/coding_env/README.md b/src/envs/coding_env/README.md new file mode 100644 index 0000000000000000000000000000000000000000..ab500002750c4a07741a8539f1acbc779d101020 --- /dev/null +++ b/src/envs/coding_env/README.md @@ -0,0 +1,133 @@ +--- +title: Coding Environment Server +emoji: šŸ’» +colorFrom: '#007ACC' +colorTo: '#1E1E1E' +sdk: docker +pinned: false +app_port: 8000 +base_path: /web +tags: + - openenv +--- + +# Coding Environment + +A Python code execution environment that runs arbitrary Python code and returns results. Perfect for testing code execution infrastructure and demonstrating environment usage patterns. + +## Quick Start + +The simplest way to use the Coding environment is through the `CodingEnv` class: + +```python +from envs.coding_env import CodeAction, CodingEnv + +try: + # Create environment from Docker image + coding_env = CodingEnv.from_docker_image("coding-env:latest") + + # Reset + result = coding_env.reset() + print(f"Reset complete: exit_code={result.observation.exit_code}") + + # Execute Python code + code_samples = [ + "print('Hello, World!')", + "x = 5 + 3\nprint(f'Result: {x}')", + "import math\nprint(math.pi)" + ] + + for code in code_samples: + result = coding_env.step(CodeAction(code=code)) + print(f"Code: {code}") + print(f" ā†’ stdout: {result.observation.stdout.strip()}") + print(f" ā†’ exit_code: {result.observation.exit_code}") + +finally: + # Always clean up + coding_env.close() +``` + +That's it! The `CodingEnv.from_docker_image()` method handles: +- Starting the Docker container +- Waiting for the server to be ready +- Connecting to the environment +- Container cleanup when you call `close()` + +## Building the Docker Image + +Before using the environment, you need to build the Docker image: + +```bash +# From project root +docker build -t coding-env:latest -f src/envs/coding_env/server/Dockerfile . +``` + +## Environment Details + +### Action +**CodeAction**: Contains a single field +- `code` (str) - The Python code to execute + +### Observation +**CodeObservation**: Contains the execution results +- `stdout` (str) - Standard output from code execution +- `stderr` (str) - Standard error from code execution +- `exit_code` (int) - Exit code (0 for success, non-zero for errors) + +### State +**CodeState**: Tracks execution state +- `episode_id` (str) - Unique identifier for the episode +- `step_count` (int) - Number of steps taken +- `last_exit_code` (int) - Exit code from the last execution + +## Advanced Usage + +### Connecting to an Existing Server + +If you already have a Coding environment server running, you can connect directly: + +```python +from envs.coding_env import CodingEnv + +# Connect to existing server +coding_env = CodingEnv(base_url="") + +# Use as normal +result = coding_env.reset() +result = coding_env.step(CodeAction(code="print('Hello!')")) +``` + +Note: When connecting to an existing server, `coding_env.close()` will NOT stop the server. + +## Development & Testing + +### Running the Full Example + +Run the complete example that demonstrates the full workflow: + +```bash +python3 src/envs/coding_env/client/example_usage.py +``` + +This example shows: +- Creating an environment from a Docker image +- Resetting and executing code through the environment +- Automatic cleanup with `close()` + +## Project Structure + +``` +coding_env/ +ā”œā”€ā”€ README.md # This file +ā”œā”€ā”€ models.py # Action, Observation, and State models +ā”œā”€ā”€ client/ +ā”‚ ā”œā”€ā”€ coding_env_client.py # CodingEnv client implementation +ā”‚ ā””ā”€ā”€ example_usage.py # Usage examples +ā””ā”€ā”€ server/ + ā”œā”€ā”€ python_codeact_env.py # Core environment logic + ā”œā”€ā”€ app.py # FastAPI application + ā”œā”€ā”€ transforms.py # Observation transforms + ā”œā”€ā”€ Dockerfile # Container image definition + ā””ā”€ā”€ README.md # Server-specific documentation +``` diff --git a/src/envs/coding_env/__init__.py b/src/envs/coding_env/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..9fd16ead48c278750e4db3e0c772f398e749f84c --- /dev/null +++ b/src/envs/coding_env/__init__.py @@ -0,0 +1,12 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Coding Environment - A Python code execution environment.""" + +from .coding_env_client import CodingEnv +from .models import CodeAction, CodeObservation, CodeState + +__all__ = ["CodeAction", "CodeObservation", "CodeState", "CodingEnv"] diff --git a/src/envs/coding_env/__pycache__/__init__.cpython-313.pyc b/src/envs/coding_env/__pycache__/__init__.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..92beb0bd2bb321a77431f437ef614721ad4d3423 Binary files /dev/null and b/src/envs/coding_env/__pycache__/__init__.cpython-313.pyc differ diff --git a/src/envs/coding_env/__pycache__/coding_env_client.cpython-313.pyc b/src/envs/coding_env/__pycache__/coding_env_client.cpython-313.pyc new file mode 100644 index 0000000000000000000000000000000000000000..cbd0eea16f505e9b5b15cc079a38c2b95bb2b5ee Binary files /dev/null and b/src/envs/coding_env/__pycache__/coding_env_client.cpython-313.pyc differ diff --git a/src/envs/coding_env/coding_env_client.py b/src/envs/coding_env/coding_env_client.py new file mode 100644 index 0000000000000000000000000000000000000000..6d06e782ef4a80cfff05dcf89a2155ad2b359c26 --- /dev/null +++ b/src/envs/coding_env/coding_env_client.py @@ -0,0 +1,60 @@ +""" +CodingEnv +--------- +Client-side wrapper for the Coding environment server. +Talks HTTP to a single base_url exposing: /reset and /step. + +- users instantiate CodingEnv with a base_url provided by the higher-level + vector/orchestration layer. +- Environment authors ship the Docker image that serves the HTTP API. + +(Seeds, episode IDs, request IDs, capabilities can be added later in the payloads.) +""" + +from __future__ import annotations + +from typing import Optional, TYPE_CHECKING + +from core.client_types import StepResult + +from core.http_env_client import HTTPEnvClient + +from .models import CodeAction, CodeObservation, CodeState + +if TYPE_CHECKING: + from core.containers.runtime import ContainerProvider + + +class CodingEnv(HTTPEnvClient[CodeAction, CodeObservation]): + # --- HTTPEnvClient abstract hooks --- + + def _step_payload(self, action: CodeAction) -> dict: + # Shape expected by the server's /step endpoint under "action" + return { + "code": action.code, + } + + def _parse_result(self, payload: dict) -> StepResult[CodeObservation]: + # Expecting: { "observation": {...}, "reward": , "done": , "info": {...} } + obs = CodeObservation(**payload["observation"]) + return StepResult( + observation=obs, + reward=payload.get("reward"), + done=bool(payload.get("done", False)), + ) + + def _parse_state(self, payload: dict) -> CodeState: + """ + Parse server response into CodeState object. + + Args: + payload: JSON response from /state endpoint + + Returns: + CodeState object with episode_id, step_count, and last_exit_code + """ + return CodeState( + episode_id=payload.get("episode_id"), + step_count=payload.get("step_count", 0), + last_exit_code=payload.get("last_exit_code", 0), + ) diff --git a/src/envs/coding_env/models.py b/src/envs/coding_env/models.py new file mode 100644 index 0000000000000000000000000000000000000000..92436ba5ab7475f5b032bc8594624afb910d114b --- /dev/null +++ b/src/envs/coding_env/models.py @@ -0,0 +1,40 @@ +""" +envs/coding_env/models.py +-------------------------------- +Action/Observation types for the Coding environment. +""" + +from __future__ import annotations + +from dataclasses import dataclass +from typing import Any, Optional + +from core.env_server import Action, Observation, State + + +@dataclass +class CodeAction(Action): + """ + Represents a single code execution request. + """ + + code: str + # Optional: future fields like 'lint': bool, 'timeout_s': float, etc. + + +@dataclass +class CodeObservation(Observation): + """ + Result of executing code in the environment. + """ + + stdout: str = "" + stderr: str = "" + exit_code: int = 0 + + +@dataclass +class CodeState(State): + """State for CodeAct environment with persistent execution context.""" + + last_exit_code: int = 0 diff --git a/src/envs/coding_env/server/Dockerfile b/src/envs/coding_env/server/Dockerfile new file mode 100644 index 0000000000000000000000000000000000000000..7cf90d5ddf5dcefc191a84b2930f276a2b6e8c36 --- /dev/null +++ b/src/envs/coding_env/server/Dockerfile @@ -0,0 +1,29 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +# Use the standard openenv base image +# Built from: docker build -t openenv-base:latest -f src/core/containers/images/Dockerfile . +# In GitHub Actions, this is overridden to use the GHCR base image +ARG BASE_IMAGE=openenv-base:latest +FROM ${BASE_IMAGE} + +# Install dependencies +COPY src/envs/coding_env/server/requirements.txt /tmp/requirements.txt +RUN pip install --no-cache-dir -r /tmp/requirements.txt && rm /tmp/requirements.txt + +# Copy only what's needed for this environment +COPY src/core/ /app/src/core/ +COPY src/envs/coding_env/ /app/src/envs/coding_env/ + +# Copy README for web interface documentation +COPY src/envs/coding_env/README.md /app/README.md + +# Health check +HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \ + CMD curl -f http://localhost:8000/health || exit 1 + +# Run the FastAPI server +CMD ["uvicorn", "envs.coding_env.server.app:app", "--host", "0.0.0.0", "--port", "8000"] diff --git a/src/envs/coding_env/server/README.md b/src/envs/coding_env/server/README.md new file mode 100644 index 0000000000000000000000000000000000000000..a4ffa757004e3cccab861d0d71b21b1112b5f2a2 --- /dev/null +++ b/src/envs/coding_env/server/README.md @@ -0,0 +1,51 @@ +# CodingEnv HTTP Server + +This directory contains the HTTP server implementation for the CodingEnvironment. + +## Running Locally + +### Prerequisites +```bash +pip install fastapi uvicorn +``` + +### Start the server +```bash +# From the project root (/Users/pankit/git/envtorch) +cd src +uvicorn envs.coding_env.server.app:app --reload --host 0.0.0.0 --port 8000 +``` + +The server will be available at `http://localhost:8000` + +### API Endpoints + +- `POST /reset` - Reset the environment +- `POST /step` - Execute a code action +- `GET /state` - Get current environment state +- `GET /health` - Health check + +### Test with curl + +```bash +# Health check +curl http://localhost:8000/health + +# Reset +curl -X POST http://localhost:8000/reset \ + -H "Content-Type: application/json" \ + -d '{}' + +# Execute code +curl -X POST http://localhost:8000/step \ + -H "Content-Type: application/json" \ + -d '{ + "action": { + "code": "print(\"Hello from HTTP!\")" + }, + "timeout_s": 15 + }' + +# Get state +curl http://localhost:8000/state +``` diff --git a/src/envs/coding_env/server/app.py b/src/envs/coding_env/server/app.py new file mode 100644 index 0000000000000000000000000000000000000000..3a895474771c21396aef9797ee7b7e748406c14e --- /dev/null +++ b/src/envs/coding_env/server/app.py @@ -0,0 +1,39 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +FastAPI application for the Coding Environment. + +This module creates an HTTP server that exposes the PythonCodeActEnv +over HTTP endpoints, making it compatible with HTTPEnvClient. + +Usage: + # Development (with auto-reload): + uvicorn envs.coding_env.server.app:app --reload --host 0.0.0.0 --port 8000 + + # Production: + uvicorn envs.coding_env.server.app:app --host 0.0.0.0 --port 8000 --workers 4 + + # Or run directly: + python -m envs.coding_env.server.app +""" + +from core.env_server import create_app + +from ..models import CodeAction, CodeObservation +from .python_codeact_env import PythonCodeActEnv + +# Create the environment instance +env = PythonCodeActEnv() + +# Create the app with web interface and README integration +app = create_app(env, CodeAction, CodeObservation, env_name="coding_env") + + +if __name__ == "__main__": + import uvicorn + + uvicorn.run(app, host="0.0.0.0", port=8000) diff --git a/src/envs/coding_env/server/python_codeact_env.py b/src/envs/coding_env/server/python_codeact_env.py new file mode 100644 index 0000000000000000000000000000000000000000..14daf2c9179a3bee5d4053b48c37536028ce0615 --- /dev/null +++ b/src/envs/coding_env/server/python_codeact_env.py @@ -0,0 +1,115 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +""" +Python Code Action Environment. + +This module provides a server-side environment implementation for executing +Python code actions using PyExecutor. +""" + +import uuid + +from core.env_server import Action, Environment, Observation +from core.tools import PyExecutor + +from ..models import CodeAction, CodeObservation, CodeState +from .transforms import create_safe_coding_transform + + +class PythonCodeActEnv(Environment): + """ + Python Code Action Environment for executing code and tracking state. + + This environment executes Python code submitted as CodeAction during step, + maintains the last exit code in its state, and returns results wrapped + in CodeObservation. + + Args: + transform: Optional transform to apply to observations + additional_imports: List of additional module imports to authorize + (e.g., ["numpy", "pandas", "matplotlib"]) + + Example: + >>> env = PythonCodeActEnv() + >>> obs = env.reset() + >>> action = CodeAction(code="print('Hello, World!')") + >>> obs = env.step(action) + >>> print(obs.stdout) # "Hello, World!\n" + >>> print(obs.exit_code) # 0 + >>> print(env.state.last_exit_code) # 0 + """ + + def __init__( + self, + ): + self.transform = create_safe_coding_transform() + self._executor = PyExecutor() + self._state = CodeState() + + def reset(self) -> Observation: + """ + Reset environment and start fresh execution session. + + Returns: + Initial observation with empty stdout/stderr and exit_code=0 + """ + # Initialize fresh state + self._state = CodeState(episode_id=str(uuid.uuid4()), step_count=0) + # Add last_exit_code to state + self._state.last_exit_code = 0 + + # Reset executor to clear any previously defined variables/functions + self._executor = PyExecutor() + + # Reset transform to clear any accumulated state + self.transform = create_safe_coding_transform() + + # Return initial observation + observation = CodeObservation( + stdout="", + stderr="", + exit_code=0, + ) + + return self._apply_transform(observation) + + def step(self, action: Action) -> Observation: + """ + Execute code action and return observation. + + Args: + action: CodeAction containing the code to execute + + Returns: + CodeObservation with execution results (stdout, stderr, exit_code) + + Raises: + ValueError: If action is not a CodeAction instance + """ + if not isinstance(action, CodeAction): + raise ValueError(f"Expected CodeAction, got {type(action)}") + + # Execute the code using PyExecutor + result = self._executor.run(action.code) + + # Update state + self._state.step_count += 1 + self._state.last_exit_code = result.exit_code + + # Create observation from execution result + observation = CodeObservation( + stdout=result.stdout, + stderr=result.stderr, + exit_code=result.exit_code, + ) + + return self._apply_transform(observation) + + @property + def state(self) -> CodeState: + """Get current environment state including last exit code.""" + return self._state diff --git a/src/envs/coding_env/server/requirements.txt b/src/envs/coding_env/server/requirements.txt new file mode 100644 index 0000000000000000000000000000000000000000..cf6769a09a61bc4675a0a3e254e777ad0077237f --- /dev/null +++ b/src/envs/coding_env/server/requirements.txt @@ -0,0 +1 @@ +smolagents diff --git a/src/envs/coding_env/server/transforms.py b/src/envs/coding_env/server/transforms.py new file mode 100644 index 0000000000000000000000000000000000000000..38ce952b018d7ff35c0a6289c0f9c0c17d13a208 --- /dev/null +++ b/src/envs/coding_env/server/transforms.py @@ -0,0 +1,94 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. +# All rights reserved. +# +# This source code is licensed under the BSD-style license found in the +# LICENSE file in the root directory of this source tree. + +"""Transforms specific to coding environments.""" + +import ast +import re + +from core.env_server.base_transforms import CompositeTransform +from core.env_server.interfaces import Transform +from core.env_server.types import Observation + +from ..models import CodeObservation + + +class CodeSafetyTransform(Transform): + """Evaluates code safety and assigns penalties for dangerous patterns.""" + + def __init__(self, penalty: float = -1.0): + self.penalty = penalty + self.dangerous_patterns = [ + r"import\s+os", + r"import\s+subprocess", + r"eval\(", + r"exec\(", + r"__import__", + r"open\(", + ] + + def __call__(self, observation: Observation) -> Observation: + if not isinstance(observation, CodeObservation): + return observation + + if "last_code" in observation.metadata: + code = observation.metadata["last_code"] + for pattern in self.dangerous_patterns: + if re.search(pattern, code): + observation.reward = self.penalty + observation.metadata["safety_violation"] = pattern + break + else: + if observation.reward is None: + observation.reward = 0.0 + + return observation + + +class CodeQualityTransform(Transform): + """Evaluates and rewards code quality metrics.""" + + def __init__( + self, + concise_bonus: float = 0.1, + max_length_threshold: int = 100, + syntax_penalty: float = -0.2, + ): + self.concise_bonus = concise_bonus + self.max_length_threshold = max_length_threshold + self.syntax_penalty = syntax_penalty + + def __call__(self, observation: Observation) -> Observation: + if not isinstance(observation, CodeObservation): + return observation + + quality_score = 0.0 + + if "last_code" in observation.metadata: + code = observation.metadata["last_code"] + + # Reward concise code + if len(code.strip()) <= self.max_length_threshold: + quality_score += self.concise_bonus + + # Check syntax (redundant but useful for quality assessment) + try: + ast.parse(code) + except SyntaxError: + quality_score += self.syntax_penalty + + # Add to existing reward + if observation.reward is None: + observation.reward = quality_score + else: + observation.reward += quality_score + + return observation + + +def create_safe_coding_transform() -> CompositeTransform: + """Create a transform focused on safe coding practices and quality.""" + return CompositeTransform([CodeSafetyTransform(), CodeQualityTransform()])