Upload folder using huggingface_hub

Dockerfile

@@ -54,6 +54,14 @@ RUN --mount=type=cache,target=/root/.cache/uv \
         uv sync --no-editable; \
     fi
 
+# Explicitly install notebook rendering + model inference deps.
+# These may not be in the base image or uv.lock on first deploy.
+RUN .venv/bin/pip install --quiet \
+    nbconvert>=7.0.0 \
+    nbformat>=5.0.0 \
+    transformers>=4.40.0 \
+    accelerate>=0.29.0
+
 # Final runtime stage
 FROM ${BASE_IMAGE}
 

README.md

@@ -1,255 +1,255 @@
----
-title: Incident Response Env Environment Server
-emoji: 🎣
-colorFrom: red
-colorTo: pink
-sdk: docker
-pinned: false
-app_port: 8000
-base_path: /web
-tags:
-  - openenv
----
-
-# Incident Response Env Environment
-
-A simple test environment that echoes back messages. Perfect for testing the env APIs as well as demonstrating environment usage patterns.
-
-## Quick Start
-
-The simplest way to use the Incident Response Env environment is through the `IncidentResponseEnv` class:
-
-```python
-from incident_response_env import IncidentResponseAction, IncidentResponseEnv
-
-try:
-    # Create environment from Docker image
-    incident_response_envenv = IncidentResponseEnv.from_docker_image("incident_response_env-env:latest")
-
-    # Reset
-    result = incident_response_envenv.reset()
-    print(f"Reset: {result.observation.echoed_message}")
-
-    # Send multiple messages
-    messages = ["Hello, World!", "Testing echo", "Final message"]
-
-    for msg in messages:
-        result = incident_response_envenv.step(IncidentResponseAction(message=msg))
-        print(f"Sent: '{msg}'")
-        print(f"  → Echoed: '{result.observation.echoed_message}'")
-        print(f"  → Length: {result.observation.message_length}")
-        print(f"  → Reward: {result.reward}")
-
-finally:
-    # Always clean up
-    incident_response_envenv.close()
-```
-
-That's it! The `IncidentResponseEnv.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 incident_response_env-env:latest -f server/Dockerfile .
-```
-
-## Deploying to Hugging Face Spaces
-
-You can easily deploy your OpenEnv environment to Hugging Face Spaces using the `openenv push` command:
-
-```bash
-# From the environment directory (where openenv.yaml is located)
-openenv push
-
-# Or specify options
-openenv push --namespace my-org --private
-```
-
-The `openenv push` command will:
-1. Validate that the directory is an OpenEnv environment (checks for `openenv.yaml`)
-2. Prepare a custom build for Hugging Face Docker space (enables web interface)
-3. Upload to Hugging Face (ensuring you're logged in)
-
-### Prerequisites
-
-- Authenticate with Hugging Face: The command will prompt for login if not already authenticated
-
-### Options
-
-- `--directory`, `-d`: Directory containing the OpenEnv environment (defaults to current directory)
-- `--repo-id`, `-r`: Repository ID in format 'username/repo-name' (defaults to 'username/env-name' from openenv.yaml)
-- `--base-image`, `-b`: Base Docker image to use (overrides Dockerfile FROM)
-- `--private`: Deploy the space as private (default: public)
-
-### Examples
-
-```bash
-# Push to your personal namespace (defaults to username/env-name from openenv.yaml)
-openenv push
-
-# Push to a specific repository
-openenv push --repo-id my-org/my-env
-
-# Push with a custom base image
-openenv push --base-image ghcr.io/meta-pytorch/openenv-base:latest
-
-# Push as a private space
-openenv push --private
-
-# Combine options
-openenv push --repo-id my-org/my-env --base-image custom-base:latest --private
-```
-
-After deployment, your space will be available at:
-`https://huggingface.co/spaces/<repo-id>`
-
-The deployed space includes:
-- **Web Interface** at `/web` - Interactive UI for exploring the environment
-- **API Documentation** at `/docs` - Full OpenAPI/Swagger interface
-- **Health Check** at `/health` - Container health monitoring
-- **WebSocket** at `/ws` - Persistent session endpoint for low-latency interactions
-
-## Environment Details
-
-### Action
-**IncidentResponseAction**: Contains a single field
-- `message` (str) - The message to echo back
-
-### Observation
-**IncidentResponseObservation**: Contains the echo response and metadata
-- `echoed_message` (str) - The message echoed back
-- `message_length` (int) - Length of the message
-- `reward` (float) - Reward based on message length (length × 0.1)
-- `done` (bool) - Always False for echo environment
-- `metadata` (dict) - Additional info like step count
-
-### Reward
-The reward is calculated as: `message_length × 0.1`
-- "Hi" → reward: 0.2
-- "Hello, World!" → reward: 1.3
-- Empty message → reward: 0.0
-
-## Advanced Usage
-
-### Connecting to an Existing Server
-
-If you already have a Incident Response Env environment server running, you can connect directly:
-
-```python
-from incident_response_env import IncidentResponseEnv
-
-# Connect to existing server
-incident_response_envenv = IncidentResponseEnv(base_url="<ENV_HTTP_URL_HERE>")
-
-# Use as normal
-result = incident_response_envenv.reset()
-result = incident_response_envenv.step(IncidentResponseAction(message="Hello!"))
-```
-
-Note: When connecting to an existing server, `incident_response_envenv.close()` will NOT stop the server.
-
-### Using the Context Manager
-
-The client supports context manager usage for automatic connection management:
-
-```python
-from incident_response_env import IncidentResponseAction, IncidentResponseEnv
-
-# Connect with context manager (auto-connects and closes)
-with IncidentResponseEnv(base_url="http://localhost:8000") as env:
-    result = env.reset()
-    print(f"Reset: {result.observation.echoed_message}")
-    # Multiple steps with low latency
-    for msg in ["Hello", "World", "!"]:
-        result = env.step(IncidentResponseAction(message=msg))
-        print(f"Echoed: {result.observation.echoed_message}")
-```
-
-The client uses WebSocket connections for:
-- **Lower latency**: No HTTP connection overhead per request
-- **Persistent session**: Server maintains your environment state
-- **Efficient for episodes**: Better for many sequential steps
-
-### Concurrent WebSocket Sessions
-
-The server supports multiple concurrent WebSocket connections. To enable this,
-modify `server/app.py` to use factory mode:
-
-```python
-# In server/app.py - use factory mode for concurrent sessions
-app = create_app(
-    IncidentResponseEnvironment,  # Pass class, not instance
-    IncidentResponseAction,
-    IncidentResponseObservation,
-    max_concurrent_envs=4,  # Allow 4 concurrent sessions
-)
-```
-
-Then multiple clients can connect simultaneously:
-
-```python
-from incident_response_env import IncidentResponseAction, IncidentResponseEnv
-from concurrent.futures import ThreadPoolExecutor
-
-def run_episode(client_id: int):
-    with IncidentResponseEnv(base_url="http://localhost:8000") as env:
-        result = env.reset()
-        for i in range(10):
-            result = env.step(IncidentResponseAction(message=f"Client {client_id}, step {i}"))
-        return client_id, result.observation.message_length
-
-# Run 4 episodes concurrently
-with ThreadPoolExecutor(max_workers=4) as executor:
-    results = list(executor.map(run_episode, range(4)))
-```
-
-## Development & Testing
-
-### Direct Environment Testing
-
-Test the environment logic directly without starting the HTTP server:
-
-```bash
-# From the server directory
-python3 server/incident_response_env_environment.py
-```
-
-This verifies that:
-- Environment resets correctly
-- Step executes actions properly
-- State tracking works
-- Rewards are calculated correctly
-
-### Running Locally
-
-Run the server locally for development:
-
-```bash
-uvicorn server.app:app --reload
-```
-
-## Project Structure
-
-```
-incident_response_env/
-├── .dockerignore         # Docker build exclusions
-├── __init__.py            # Module exports
-├── README.md              # This file
-├── openenv.yaml           # OpenEnv manifest
-├── pyproject.toml         # Project metadata and dependencies
-├── uv.lock                # Locked dependencies (generated)
-├── client.py              # IncidentResponseEnv client
-├── models.py              # Action and Observation models
-└── server/
-    ├── __init__.py        # Server module exports
-    ├── incident_response_env_environment.py  # Core environment logic
-    ├── app.py             # FastAPI application (HTTP + WebSocket endpoints)
-    └── Dockerfile         # Container image definition
-```
+---
+title: Incident Response Env Environment Server
+emoji: 🎣
+colorFrom: red
+colorTo: pink
+sdk: docker
+pinned: false
+app_port: 8000
+base_path: /
+tags:
+  - openenv
+---
+
+# Incident Response Env Environment
+
+A simple test environment that echoes back messages. Perfect for testing the env APIs as well as demonstrating environment usage patterns.
+
+## Quick Start
+
+The simplest way to use the Incident Response Env environment is through the `IncidentResponseEnv` class:
+
+```python
+from incident_response_env import IncidentResponseAction, IncidentResponseEnv
+
+try:
+    # Create environment from Docker image
+    incident_response_envenv = IncidentResponseEnv.from_docker_image("incident_response_env-env:latest")
+
+    # Reset
+    result = incident_response_envenv.reset()
+    print(f"Reset: {result.observation.echoed_message}")
+
+    # Send multiple messages
+    messages = ["Hello, World!", "Testing echo", "Final message"]
+
+    for msg in messages:
+        result = incident_response_envenv.step(IncidentResponseAction(message=msg))
+        print(f"Sent: '{msg}'")
+        print(f"  → Echoed: '{result.observation.echoed_message}'")
+        print(f"  → Length: {result.observation.message_length}")
+        print(f"  → Reward: {result.reward}")
+
+finally:
+    # Always clean up
+    incident_response_envenv.close()
+```
+
+That's it! The `IncidentResponseEnv.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 incident_response_env-env:latest -f server/Dockerfile .
+```
+
+## Deploying to Hugging Face Spaces
+
+You can easily deploy your OpenEnv environment to Hugging Face Spaces using the `openenv push` command:
+
+```bash
+# From the environment directory (where openenv.yaml is located)
+openenv push
+
+# Or specify options
+openenv push --namespace my-org --private
+```
+
+The `openenv push` command will:
+1. Validate that the directory is an OpenEnv environment (checks for `openenv.yaml`)
+2. Prepare a custom build for Hugging Face Docker space (enables web interface)
+3. Upload to Hugging Face (ensuring you're logged in)
+
+### Prerequisites
+
+- Authenticate with Hugging Face: The command will prompt for login if not already authenticated
+
+### Options
+
+- `--directory`, `-d`: Directory containing the OpenEnv environment (defaults to current directory)
+- `--repo-id`, `-r`: Repository ID in format 'username/repo-name' (defaults to 'username/env-name' from openenv.yaml)
+- `--base-image`, `-b`: Base Docker image to use (overrides Dockerfile FROM)
+- `--private`: Deploy the space as private (default: public)
+
+### Examples
+
+```bash
+# Push to your personal namespace (defaults to username/env-name from openenv.yaml)
+openenv push
+
+# Push to a specific repository
+openenv push --repo-id my-org/my-env
+
+# Push with a custom base image
+openenv push --base-image ghcr.io/meta-pytorch/openenv-base:latest
+
+# Push as a private space
+openenv push --private
+
+# Combine options
+openenv push --repo-id my-org/my-env --base-image custom-base:latest --private
+```
+
+After deployment, your space will be available at:
+`https://huggingface.co/spaces/<repo-id>`
+
+The deployed space includes:
+- **Web Interface** at `/web` - Interactive UI for exploring the environment
+- **API Documentation** at `/docs` - Full OpenAPI/Swagger interface
+- **Health Check** at `/health` - Container health monitoring
+- **WebSocket** at `/ws` - Persistent session endpoint for low-latency interactions
+
+## Environment Details
+
+### Action
+**IncidentResponseAction**: Contains a single field
+- `message` (str) - The message to echo back
+
+### Observation
+**IncidentResponseObservation**: Contains the echo response and metadata
+- `echoed_message` (str) - The message echoed back
+- `message_length` (int) - Length of the message
+- `reward` (float) - Reward based on message length (length × 0.1)
+- `done` (bool) - Always False for echo environment
+- `metadata` (dict) - Additional info like step count
+
+### Reward
+The reward is calculated as: `message_length × 0.1`
+- "Hi" → reward: 0.2
+- "Hello, World!" → reward: 1.3
+- Empty message → reward: 0.0
+
+## Advanced Usage
+
+### Connecting to an Existing Server
+
+If you already have a Incident Response Env environment server running, you can connect directly:
+
+```python
+from incident_response_env import IncidentResponseEnv
+
+# Connect to existing server
+incident_response_envenv = IncidentResponseEnv(base_url="<ENV_HTTP_URL_HERE>")
+
+# Use as normal
+result = incident_response_envenv.reset()
+result = incident_response_envenv.step(IncidentResponseAction(message="Hello!"))
+```
+
+Note: When connecting to an existing server, `incident_response_envenv.close()` will NOT stop the server.
+
+### Using the Context Manager
+
+The client supports context manager usage for automatic connection management:
+
+```python
+from incident_response_env import IncidentResponseAction, IncidentResponseEnv
+
+# Connect with context manager (auto-connects and closes)
+with IncidentResponseEnv(base_url="http://localhost:8000") as env:
+    result = env.reset()
+    print(f"Reset: {result.observation.echoed_message}")
+    # Multiple steps with low latency
+    for msg in ["Hello", "World", "!"]:
+        result = env.step(IncidentResponseAction(message=msg))
+        print(f"Echoed: {result.observation.echoed_message}")
+```
+
+The client uses WebSocket connections for:
+- **Lower latency**: No HTTP connection overhead per request
+- **Persistent session**: Server maintains your environment state
+- **Efficient for episodes**: Better for many sequential steps
+
+### Concurrent WebSocket Sessions
+
+The server supports multiple concurrent WebSocket connections. To enable this,
+modify `server/app.py` to use factory mode:
+
+```python
+# In server/app.py - use factory mode for concurrent sessions
+app = create_app(
+    IncidentResponseEnvironment,  # Pass class, not instance
+    IncidentResponseAction,
+    IncidentResponseObservation,
+    max_concurrent_envs=4,  # Allow 4 concurrent sessions
+)
+```
+
+Then multiple clients can connect simultaneously:
+
+```python
+from incident_response_env import IncidentResponseAction, IncidentResponseEnv
+from concurrent.futures import ThreadPoolExecutor
+
+def run_episode(client_id: int):
+    with IncidentResponseEnv(base_url="http://localhost:8000") as env:
+        result = env.reset()
+        for i in range(10):
+            result = env.step(IncidentResponseAction(message=f"Client {client_id}, step {i}"))
+        return client_id, result.observation.message_length
+
+# Run 4 episodes concurrently
+with ThreadPoolExecutor(max_workers=4) as executor:
+    results = list(executor.map(run_episode, range(4)))
+```
+
+## Development & Testing
+
+### Direct Environment Testing
+
+Test the environment logic directly without starting the HTTP server:
+
+```bash
+# From the server directory
+python3 server/incident_response_env_environment.py
+```
+
+This verifies that:
+- Environment resets correctly
+- Step executes actions properly
+- State tracking works
+- Rewards are calculated correctly
+
+### Running Locally
+
+Run the server locally for development:
+
+```bash
+uvicorn server.app:app --reload
+```
+
+## Project Structure
+
+```
+incident_response_env/
+├── .dockerignore         # Docker build exclusions
+├── __init__.py            # Module exports
+├── README.md              # This file
+├── openenv.yaml           # OpenEnv manifest
+├── pyproject.toml         # Project metadata and dependencies
+├── uv.lock                # Locked dependencies (generated)
+├── client.py              # IncidentResponseEnv client
+├── models.py              # Action and Observation models
+└── server/
+    ├── __init__.py        # Server module exports
+    ├── incident_response_env_environment.py  # Core environment logic
+    ├── app.py             # FastAPI application (HTTP + WebSocket endpoints)
+    └── Dockerfile         # Container image definition
+```

server/_notebook.html

@@ -9001,9 +9001,9 @@ Unsloth: Fast downloading is enabled - ignore downloading bars which are red col
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="5a82074c-8f9c-45ea-bf9d-7a40d8dfdf0a" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="4665c438-18da-42d7-b933-ffd6eaadd2a6" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('5a82074c-8f9c-45ea-bf9d-7a40d8dfdf0a');
+var element = document.getElementById('4665c438-18da-42d7-b933-ffd6eaadd2a6');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "a43b656123724da68f8ce0b1d3c8ce6f", "version_major": 2, "version_minor": 0}
@@ -9012,9 +9012,9 @@ var element = document.getElementById('5a82074c-8f9c-45ea-bf9d-7a40d8dfdf0a');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="44c9921c-4aaf-4dff-a61d-e71efefd088c" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="724f32b1-4f56-40cc-9650-581bbc9bbc4f" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('44c9921c-4aaf-4dff-a61d-e71efefd088c');
+var element = document.getElementById('724f32b1-4f56-40cc-9650-581bbc9bbc4f');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "55dfc654c378407799fea2077dd9a102", "version_major": 2, "version_minor": 0}
@@ -9023,9 +9023,9 @@ var element = document.getElementById('44c9921c-4aaf-4dff-a61d-e71efefd088c');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="d17747f9-b240-43ef-bd9b-d10a239534a7" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="cd3368e7-baaf-4ad7-a0f7-7bf3935a38fa" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('d17747f9-b240-43ef-bd9b-d10a239534a7');
+var element = document.getElementById('cd3368e7-baaf-4ad7-a0f7-7bf3935a38fa');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "e0b7e6ef22854ded9e3b5334ae2314d7", "version_major": 2, "version_minor": 0}
@@ -9034,9 +9034,9 @@ var element = document.getElementById('d17747f9-b240-43ef-bd9b-d10a239534a7');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="3dd2bf99-e8db-4daf-85c9-81fb8fa46df2" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="93e859d4-7268-41c4-b1a6-bf227811076e" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('3dd2bf99-e8db-4daf-85c9-81fb8fa46df2');
+var element = document.getElementById('93e859d4-7268-41c4-b1a6-bf227811076e');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "d4406d23036646ba87287c7a34db6967", "version_major": 2, "version_minor": 0}
@@ -9045,9 +9045,9 @@ var element = document.getElementById('3dd2bf99-e8db-4daf-85c9-81fb8fa46df2');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="3a32585f-caec-4584-b57c-0d4457510095" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="aa50e38d-d271-40e8-a1f0-74c8717a1af1" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('3a32585f-caec-4584-b57c-0d4457510095');
+var element = document.getElementById('aa50e38d-d271-40e8-a1f0-74c8717a1af1');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "7f5e0fa0d7f04f73a1c71206b2d14d03", "version_major": 2, "version_minor": 0}
@@ -9056,9 +9056,9 @@ var element = document.getElementById('3a32585f-caec-4584-b57c-0d4457510095');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="f25275b2-2a77-4cad-a72b-52811f6d2bab" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="bcc3300f-6eda-4e37-bf3e-2cb2568264ca" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('f25275b2-2a77-4cad-a72b-52811f6d2bab');
+var element = document.getElementById('bcc3300f-6eda-4e37-bf3e-2cb2568264ca');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "d59108a79279418b89082c3c35c791b5", "version_major": 2, "version_minor": 0}
@@ -9067,9 +9067,9 @@ var element = document.getElementById('f25275b2-2a77-4cad-a72b-52811f6d2bab');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="78585ee7-2cc7-4cea-b4cc-921123492381" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="bf615dad-1d33-41f8-87d8-9eb5f2f6b0d1" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('78585ee7-2cc7-4cea-b4cc-921123492381');
+var element = document.getElementById('bf615dad-1d33-41f8-87d8-9eb5f2f6b0d1');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "30b37f3a86b54844b1b522f63f94254e", "version_major": 2, "version_minor": 0}
@@ -9078,9 +9078,9 @@ var element = document.getElementById('78585ee7-2cc7-4cea-b4cc-921123492381');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="d83fb147-f81d-4f90-85f5-11e0de35a95f" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="070b76e1-d222-4ab8-ab76-c199b158522c" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('d83fb147-f81d-4f90-85f5-11e0de35a95f');
+var element = document.getElementById('070b76e1-d222-4ab8-ab76-c199b158522c');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "01e641d29b9d41da93d7cc9c693abdcd", "version_major": 2, "version_minor": 0}
@@ -22886,9 +22886,9 @@ Unsloth: QLoRA and full finetuning all not selected. Switching to 16bit LoRA.
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="3c098df9-c846-4621-a8f5-f74b8267fba7" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="974a857b-d684-46ed-96e9-8d0bb38ae70b" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('3c098df9-c846-4621-a8f5-f74b8267fba7');
+var element = document.getElementById('974a857b-d684-46ed-96e9-8d0bb38ae70b');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "bceb1ef5489f43c18eb7b846041dc50a", "version_major": 2, "version_minor": 0}
@@ -22897,9 +22897,9 @@ var element = document.getElementById('3c098df9-c846-4621-a8f5-f74b8267fba7');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="4269952f-96aa-4518-8633-78b2c00ac1f6" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="38f2742e-e460-4e84-82d3-5485919d4793" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('4269952f-96aa-4518-8633-78b2c00ac1f6');
+var element = document.getElementById('38f2742e-e460-4e84-82d3-5485919d4793');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "1473bf43ffc248e48fdad86849869776", "version_major": 2, "version_minor": 0}
@@ -22908,9 +22908,9 @@ var element = document.getElementById('4269952f-96aa-4518-8633-78b2c00ac1f6');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="34e6c9f1-2b6e-4484-85c5-7771e5e841fc" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="015d43d4-9066-40d4-b556-f819e055c2cd" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('34e6c9f1-2b6e-4484-85c5-7771e5e841fc');
+var element = document.getElementById('015d43d4-9066-40d4-b556-f819e055c2cd');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "2d1950fd251549968618281e1717672c", "version_major": 2, "version_minor": 0}
@@ -22919,9 +22919,9 @@ var element = document.getElementById('34e6c9f1-2b6e-4484-85c5-7771e5e841fc');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="0d0d5df8-4c18-4d89-ae02-649a01945aa5" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="1d5857a3-53df-4584-8197-685d1b35a66b" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('0d0d5df8-4c18-4d89-ae02-649a01945aa5');
+var element = document.getElementById('1d5857a3-53df-4584-8197-685d1b35a66b');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "d37ad2830cd44a94bc54195b03d794c8", "version_major": 2, "version_minor": 0}
@@ -22930,9 +22930,9 @@ var element = document.getElementById('0d0d5df8-4c18-4d89-ae02-649a01945aa5');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="a3c7f50f-a609-44de-acea-2e9d5f3b60ae" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="0b2ee197-0fc0-4092-a292-d43ceb3baaeb" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('a3c7f50f-a609-44de-acea-2e9d5f3b60ae');
+var element = document.getElementById('0b2ee197-0fc0-4092-a292-d43ceb3baaeb');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "cc1680f141514c1d86237fd46ed6ab50", "version_major": 2, "version_minor": 0}
@@ -22941,9 +22941,9 @@ var element = document.getElementById('a3c7f50f-a609-44de-acea-2e9d5f3b60ae');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="0a47160a-f191-47e0-be62-b7f65e10bb2f" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="f70afaaf-4522-47b7-9ce8-9a83244cb046" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('0a47160a-f191-47e0-be62-b7f65e10bb2f');
+var element = document.getElementById('f70afaaf-4522-47b7-9ce8-9a83244cb046');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "fa1de6e7572442a18a37ecdd236be73c", "version_major": 2, "version_minor": 0}
@@ -22952,9 +22952,9 @@ var element = document.getElementById('0a47160a-f191-47e0-be62-b7f65e10bb2f');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="f2ac9dee-fb3b-4d05-b8bb-92872650f76d" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="8e6d6c37-f9cb-4429-ba45-d32f9f6c6cc5" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('f2ac9dee-fb3b-4d05-b8bb-92872650f76d');
+var element = document.getElementById('8e6d6c37-f9cb-4429-ba45-d32f9f6c6cc5');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "fcc3dfc78fba44fea8fd65a09b32b39f", "version_major": 2, "version_minor": 0}
@@ -22963,9 +22963,9 @@ var element = document.getElementById('f2ac9dee-fb3b-4d05-b8bb-92872650f76d');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="0428d5a7-8dcb-41c5-b16a-a68ba7535232" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="4eb97556-5063-473b-b1fe-b961897e3ba4" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('0428d5a7-8dcb-41c5-b16a-a68ba7535232');
+var element = document.getElementById('4eb97556-5063-473b-b1fe-b961897e3ba4');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "2b44dac6f6ed48269f2ba327095e30a7", "version_major": 2, "version_minor": 0}
@@ -23381,9 +23381,9 @@ Total notebook runtime: 133.89 minutes (8033.2 seconds)
 <div class="jp-OutputArea jp-Cell-outputArea">
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="f73d028d-3d4a-4e5a-abc9-62ac4c1a2ffe" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="bf5ca71a-1ddc-4478-a83c-f44fa34eacfb" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('f73d028d-3d4a-4e5a-abc9-62ac4c1a2ffe');
+var element = document.getElementById('bf5ca71a-1ddc-4478-a83c-f44fa34eacfb');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "227594b06c7744f28319d7337a32cb5e", "version_major": 2, "version_minor": 0}
@@ -23392,9 +23392,9 @@ var element = document.getElementById('f73d028d-3d4a-4e5a-abc9-62ac4c1a2ffe');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="9f00317b-9a07-4c3d-869a-2179157b31a2" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="a787076d-47b9-43c0-acaf-67ead9a7f964" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('9f00317b-9a07-4c3d-869a-2179157b31a2');
+var element = document.getElementById('a787076d-47b9-43c0-acaf-67ead9a7f964');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "85bd070deb1f4c3f9a76db4f0269a119", "version_major": 2, "version_minor": 0}
@@ -23439,9 +23439,9 @@ var element = document.getElementById('9f00317b-9a07-4c3d-869a-2179157b31a2');
 <div class="jp-OutputArea jp-Cell-outputArea">
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="0829c2c9-1a21-431d-9ad4-04d76445ad33" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="a59e8d06-5446-4a8b-a07c-6ec7364ae143" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('0829c2c9-1a21-431d-9ad4-04d76445ad33');
+var element = document.getElementById('a59e8d06-5446-4a8b-a07c-6ec7364ae143');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "8b799481419c44f583f316a3eaed4e1c", "version_major": 2, "version_minor": 0}
@@ -23450,9 +23450,9 @@ var element = document.getElementById('0829c2c9-1a21-431d-9ad4-04d76445ad33');
 </div>
 <div class="jp-OutputArea-child">
 <div class="jp-OutputPrompt jp-OutputArea-prompt"></div>
-<div class="jupyter-widgets jp-OutputArea-output" id="d827d3c6-8d95-4b69-9e67-0485301a8ebe" tabindex="0">
+<div class="jupyter-widgets jp-OutputArea-output" id="04de4636-abae-4ee9-95a7-df1c50955830" tabindex="0">
 <script type="text/javascript">
-var element = document.getElementById('d827d3c6-8d95-4b69-9e67-0485301a8ebe');
+var element = document.getElementById('04de4636-abae-4ee9-95a7-df1c50955830');
 </script>
 <script type="application/vnd.jupyter.widget-view+json">
 {"model_id": "c84bd0c0cd6d4e3780221bc8497b4359", "version_major": 2, "version_minor": 0}

server/custom_ui.py

@@ -277,39 +277,28 @@ def build_custom_ui(
 
     async def run_model(role: str):
         """
-        Load the GRPO checkpoint for the current role and auto-generate an action.
-        Streams status updates (model loading, inference) via gr.update on the status box.
-        After inference, automatically calls step_env with the generated action.
+        Load the GRPO checkpoint for the current role and auto-generate an action,
+        then step the environment with the result.
         """
+        _blank = (gr.update(), gr.update(), gr.update(), "", "", "")
+
         if not _MODEL_AVAILABLE:
-            yield (gr.update(), gr.update(), gr.update(), "", "", "transformers not installed — cannot run model.", gr.update())
-            return
+            return (*_blank[:5], "transformers/torch not installed — cannot run model.")
 
         if role not in ROLE_CHOICES:
-            yield (gr.update(), gr.update(), gr.update(), "", "", "Select a role first.", gr.update())
-            return
-
-        status_msgs: List[str] = []
-
-        def _status_cb(msg: str):
-            status_msgs.append(msg)
+            return (*_blank[:5], "Select a role first.")
 
-        loop = asyncio.get_event_loop()
-        # We need the prompt from the current observation to pass to the model.
-        # Retrieve it via the web_manager's current observation.
+        # Ensure an episode is active
         try:
-            current_obs = None
-            if (
-                web_manager.episode_state
-                and web_manager.episode_state.current_observation
-            ):
-                current_obs = web_manager.episode_state.current_observation
+            current_obs = (
+                web_manager.episode_state.current_observation
+                if web_manager.episode_state and web_manager.episode_state.current_observation
+                else None
+            )
         except Exception:
             current_obs = None
 
         if current_obs is None:
-            # Auto-reset first
-            yield (gr.update(), gr.update(), gr.update(), "", "", "Auto-resetting episode…", gr.update())
             await web_manager.reset_environment()
             try:
                 current_obs = web_manager.episode_state.current_observation
@@ -319,49 +308,42 @@ def build_custom_ui(
         prompt = (current_obs or {}).get("prompt", "") if isinstance(current_obs, dict) else ""
 
         if not prompt:
-            yield (gr.update(), gr.update(), gr.update(), "", "", "No prompt available — click Reset first.", gr.update())
-            return
+            return (*_blank[:5], "No prompt yet — click Reset first.")
 
-        yield (gr.update(), gr.update(), gr.update(), "", "", f"Loading {role} model from HuggingFace…", gr.update())
+        status_msgs: List[str] = []
 
+        def _status_cb(msg: str):
+            status_msgs.append(msg)
+
+        loop = asyncio.get_event_loop()
         try:
             command = await loop.run_in_executor(
                 _MODEL_EXECUTOR,
                 lambda: _model_generate_action(role, prompt, _status_cb),
             )
         except Exception as exc:
-            yield (gr.update(), gr.update(), gr.update(), "", "", f"Model error: {exc}", gr.update())
-            return
+            return (*_blank[:5], f"Model error: {exc}")
 
         all_status = " → ".join(status_msgs)
 
         if not command:
-            yield (gr.update(), gr.update(), gr.update(), "", "", f"Model produced invalid output. {all_status}", gr.update())
-            return
+            return (*_blank[:5], f"Model produced no valid action. {all_status}")
 
-        # Map model command back to an action_id for display consistency
-        universe = ATTACKS if role == "attacker" else DEFENSES
-        prefix   = "ATTACK" if role == "attacker" else "DEFEND"
-        action_name = command.replace(f"{prefix}:", "").strip()
-        ai = universe.index(action_name) if action_name in universe else 0
-
-        # Step the environment with the generated command
         try:
             data = await web_manager.step_environment({"command": command, "role": role})
         except Exception as exc:
-            yield (gr.update(), gr.update(), gr.update(), "", "", f"Step error: {exc}", gr.update())
-            return
+            return (*_blank[:5], f"Step error after model action: {exc}")
 
-        obs       = data.get("observation", {}) or {}
-        next_role = obs.get("next_role", role)
-        health    = obs.get("health", 1.0)
+        obs         = data.get("observation", {}) or {}
+        next_role   = obs.get("next_role", role)
+        health      = obs.get("health", 1.0)
         status_text = obs.get("status", "STABLE")
-        done      = bool(obs.get("done", False))
+        done        = bool(obs.get("done", False))
         role_value  = next_role if next_role in ROLE_CHOICES else role
         new_choices = _action_choices(role_value)
         done_note   = " Episode complete — click Reset." if done else ""
 
-        yield (
+        return (
             gr.update(value=role_value),
             gr.update(choices=new_choices, value=new_choices[0][1] if new_choices else None,
                       label=f"Action ID (0..{len(new_choices) - 1})"),
@@ -369,7 +351,6 @@ def build_custom_ui(
             "",
             json.dumps(_strip_prompt_from_response_for_display(data), indent=2),
             f"Model → {command} | health={health:.0%} | {status_text} | next={next_role}.{done_note}",
-            gr.update(value=ai),
         )
 
     initial_role = "attacker"
@@ -382,38 +363,44 @@ def build_custom_ui(
 <style>
   .oe_sidebar {
     position: fixed;
-    left: 16px;
-    top: 16px;
-    width: 220px;
+    left: 14px;
+    top: 14px;
+    width: 210px;
     z-index: 9999;
-    background: rgba(15, 23, 42, 0.92);
-    border: 1px solid rgba(148, 163, 184, 0.25);
+    background: #1e293b;
+    border: 1px solid #475569;
     border-radius: 12px;
-    padding: 12px;
-    backdrop-filter: blur(8px);
+    padding: 14px 12px;
+    box-shadow: 0 4px 24px rgba(0,0,0,0.5);
+  }
+  .oe_sidebar_title {
+    font-weight: 700;
+    margin: 0 0 10px 0;
+    font-size: 12px;
+    letter-spacing: .08em;
+    text-transform: uppercase;
+    color: #94a3b8;
   }
-  .oe_sidebar_title { font-weight: 700; margin: 0 0 8px 0; font-size: 13px; color: #e2e8f0; }
   .oe_sidebar a {
     display: block;
-    padding: 8px 10px;
+    padding: 9px 12px;
     margin: 6px 0;
-    border-radius: 10px;
-    color: #e2e8f0;
+    border-radius: 8px;
+    color: #f1f5f9;
     text-decoration: none;
     font-size: 13px;
-    border: 1px solid rgba(148, 163, 184, 0.18);
-    background: rgba(30, 41, 59, 0.55);
-  }
-  .oe_sidebar a:hover { background: rgba(51, 65, 85, 0.75); }
-  /* Add a little breathing room so sidebar doesn't cover content on narrow screens. */
-  @media (min-width: 1100px) {
-    .col-left { margin-left: 240px; }
+    font-weight: 500;
+    background: #334155;
+    border: 1px solid #475569;
+    transition: background .15s;
   }
+  .oe_sidebar a:hover { background: #4a6080; color: #ffffff; }
+  @media (min-width: 1100px) { .col-left { margin-left: 240px; } }
 </style>
 <nav class="oe_sidebar">
   <div class="oe_sidebar_title">Jump to</div>
-  <a href="#interactive_playground">Interactive playground</a>
-  <a href="#grpo_training_notebook">GRPO training notebook</a>
+  <a href="#interactive_playground">🎮 Interactive Playground</a>
+  <a href="#grpo_training_notebook">📓 GRPO Training Notebook</a>
 </nav>
 """
         )
@@ -487,7 +474,7 @@ def build_custom_ui(
         model_btn.click(
             fn=run_model,
             inputs=[role_dd],
-            outputs=[role_dd, action_dd, mapping_md, obs_md, raw_json, status_box, action_dd],
+            outputs=[role_dd, action_dd, mapping_md, obs_md, raw_json, status_box],
         )
         state_btn.click(fn=get_state_sync, outputs=[raw_json])