hud-python 0.4.30__py3-none-any.whl → 0.4.32__py3-none-any.whl

hud/cli/rl/remote_runner.py

@@ -10,6 +10,7 @@ import os
 import subprocess
 import time
 from pathlib import Path
+import uuid
 
 from rich.console import Console
 
@@ -29,6 +30,43 @@ GPU_PRICING = {
 }
 
 
+def ensure_vllm_deployed(model_name: str, gpu_type: str = "A100", timeout: int = 600) -> None:
+    """Deploy vLLM for a model if needed and wait until it's ready.
+
+    Args:
+        model_name: The name of the model to deploy vLLM for
+        gpu_type: GPU type to use for deployment (e.g., A100, H100)
+        timeout: Max seconds to wait for vLLM to be ready
+    """
+    # Check current model status
+    info = rl_api.get_model(model_name)
+    if info.vllm_url:
+        hud_console.success("vLLM server already running")
+        return
+
+    hud_console.info(f"Deploying vLLM server for {model_name}...")
+    rl_api.deploy_vllm(model_name, gpu_type=gpu_type)
+    hud_console.success("vLLM deployment started")
+
+    hud_console.info("Waiting for vLLM server to be ready...")
+    start_time = time.time()
+    with hud_console.progress() as progress:
+        progress.update(
+            "Checking deployment status (see live status on https://app.hud.so/models)"
+        )
+        while True:
+            if time.time() - start_time > timeout:
+                hud_console.error("Timeout waiting for vLLM deployment")
+                raise ValueError("vLLM deployment timeout")
+            info = rl_api.get_model(model_name)
+            if info.vllm_url or info.status == "ready":
+                hud_console.success(
+                    f"vLLM server ready at http://rl.hud.so/v1/models/{model_name}/vllm"
+                )
+                break
+            time.sleep(5)
+
+
 def run_remote_training(
     tasks_file: str | None,
     model: str | None,
@@ -128,49 +166,55 @@ def run_remote_training(
             from rich.prompt import Prompt
 
             # Ask for model name
-            default_name = model_type.split("/")[-1].lower()
+            base_default = model_type.split("/")[-1].lower()
+            default_name = base_default
+            existing_names = {m.name for m in active_models}
+            suffix = 1
+            while default_name in existing_names:
+                default_name = f"{base_default}-{suffix}"
+                suffix += 1
+
             hud_console.info(f"Enter model name (default: {default_name}):")
             model_name = Prompt.ask("Model name", default=default_name)
             model_name = model_name.replace("/", "-").lower()
 
-            # Create the model
+            # Create the model with retry on name conflict
             hud_console.info(f"Creating model: {model_name}")
             try:
                 rl_api.create_model(model_name, model_type)
                 hud_console.success(f"Created model: {model_name}")
+                ensure_vllm_deployed(model_name, gpu_type="A100")
 
-                # Deploy vLLM automatically
-                hud_console.info(f"Deploying vLLM server for {model_name}...")
-                rl_api.deploy_vllm(model_name, gpu_type="A100")
-                hud_console.success("vLLM deployment started")
-
-                # Wait for deployment
-                hud_console.info("Waiting for vLLM server to be ready...")
-                max_wait = 600  # 10 minutes
-                start_time = time.time()
-
-                with hud_console.progress() as progress:
-                    progress.update(
-                        "Checking deployment status (see live status on https://app.hud.so/models)"
-                    )
-
+            except Exception as e:
+                # If the name already exists, suggest a new name and prompt once
+                message = str(e)
+                if "already exists" in message or "409" in message:
+                    alt_name = f"{model_name}-1"
+                    i = 1
                     while True:
-                        if time.time() - start_time > max_wait:
-                            hud_console.error("Timeout waiting for vLLM deployment")
-                            raise ValueError("vLLM deployment timeout")
-
-                        model_info = rl_api.get_model(model_name)
-                        if model_info.status == "ready":
-                            hud_console.success(
-                                f"vLLM server ready at http://rl.hud.so/v1/models/{model_name}/vllm"
-                            )
+                        candidate = f"{model_name}-{str(uuid.uuid4())[:4]}"
+                        if candidate not in existing_names:
+                            alt_name = candidate
                             break
-
-                        time.sleep(5)
-
-            except Exception as e:
-                hud_console.error(f"Failed to create model: {e}")
-                raise
+                        i += 1
+                    hud_console.warning(
+                        f"Model '{model_name}' exists. Suggesting '{alt_name}' instead."
+                    )
+                    try:
+                        from rich.prompt import Prompt as _Prompt
+
+                        chosen = _Prompt.ask("Use different name", default=alt_name)
+                        chosen = chosen.replace("/", "-").lower()
+                        rl_api.create_model(chosen, model_type)
+                        hud_console.success(f"Created model: {chosen}")
+                        model_name = chosen
+                        ensure_vllm_deployed(model_name, gpu_type="A100")
+                    except Exception as e2:
+                        hud_console.error(f"Failed to create model: {e2}")
+                        raise
+                else:
+                    hud_console.error(f"Failed to create model: {e}")
+                    raise
 
         else:
             # Existing model selected
@@ -194,36 +238,7 @@ def run_remote_training(
                     return
 
             # Ensure vLLM is deployed
-            if not model_info.vllm_url:
-                hud_console.info(f"Deploying vLLM server for {model_name}...")
-                rl_api.deploy_vllm(model_name, gpu_type="A100")
-                hud_console.success("vLLM deployment started")
-
-                # Wait for deployment
-                hud_console.info("Waiting for vLLM server to be ready...")
-                max_wait = 600  # 10 minutes
-                start_time = time.time()
-
-                with hud_console.progress() as progress:
-                    progress.update(
-                        "Checking deployment status (see live status on https://app.hud.so/models)"
-                    )
-
-                    while True:
-                        if time.time() - start_time > max_wait:
-                            hud_console.error("Timeout waiting for vLLM deployment")
-                            raise ValueError("vLLM deployment timeout")
-
-                        model_info = rl_api.get_model(model_name)
-                        if model_info.vllm_url:
-                            hud_console.success(
-                                f"vLLM server ready at http://rl.hud.so/v1/models/{model_name}/vllm"
-                            )
-                            break
-
-                        time.sleep(5)
-            else:
-                hud_console.success("vLLM server already running")
+            ensure_vllm_deployed(model_name, gpu_type="A100")
     except KeyboardInterrupt:
         hud_console.dim_info("Training cancelled", "")
         return
@@ -281,7 +296,17 @@ def run_remote_training(
             presets=presets,
         )
 
-        config.job_name = f"RL {model_name} on {tasks_file}"
+        # Use a short label for tasks (avoid full absolute paths)
+        try:
+            if tasks_file and Path(tasks_file).exists():
+                tasks_label = Path(tasks_file).name
+            else:
+                # Fallback: last segment of a non-existent path or dataset name
+                tasks_label = str(tasks_file).replace("\\", "/").split("/")[-1]
+        except Exception:
+            tasks_label = str(tasks_file)
+
+        config.job_name = f"RL {model_name} on {tasks_label}"
 
         # Save config for editing
         temp_config_path = Path(f".rl_config_temp_{model_name}.json")

hud/cli/utils/docker.py

@@ -3,6 +3,8 @@
 from __future__ import annotations
 
 import json
+import platform
+import shutil
 import subprocess
 
 
@@ -117,3 +119,95 @@ def generate_container_name(identifier: str, prefix: str = "hud") -> str:
     # Replace special characters with hyphens
     safe_name = identifier.replace(":", "-").replace("/", "-").replace("\\", "-")
     return f"{prefix}-{safe_name}"
+
+
+def _emit_docker_hints(error_text: str) -> None:
+    """Parse common Docker connectivity errors and print platform-specific hints."""
+    from hud.utils.hud_console import hud_console
+
+    text = error_text.lower()
+    system = platform.system()
+
+    markers = [
+        "cannot connect to the docker daemon",
+        "is the docker daemon running",
+        "error during connect",
+        "permission denied while trying to connect",
+        "no such file or directory",
+        "pipe/dockerdesktop",
+        "dockerdesktoplinuxengine",
+        "//./pipe/docker",
+        "/var/run/docker.sock",
+    ]
+
+    if any(m in text for m in markers):
+        hud_console.error("Docker does not appear to be running or accessible")
+        if system == "Windows":
+            hud_console.hint("Open Docker Desktop and wait until it shows 'Running'")
+            hud_console.hint("If using WSL, enable integration for your distro in Docker Desktop")
+        elif system == "Linux":
+            hud_console.hint(
+                "Start the daemon: sudo systemctl start docker (or service docker start)"
+            )
+            hud_console.hint("If permission denied: sudo usermod -aG docker $USER && re-login")
+        elif system == "Darwin":
+            hud_console.hint("Open Docker Desktop and wait until it shows 'Running'")
+        else:
+            hud_console.hint("Start Docker and ensure the daemon is reachable")
+        trimmed = error_text.strip()
+        if len(trimmed) > 300:
+            trimmed = trimmed[:300] + "..."
+        hud_console.dim_info("Details", trimmed)
+    else:
+        from hud.utils.hud_console import hud_console as _hc
+
+        _hc.error("Docker returned an error")
+        trimmed = error_text.strip()
+        if len(trimmed) > 300:
+            trimmed = trimmed[:300] + "..."
+        _hc.dim_info("Details", trimmed)
+        _hc.hint("Is Docker running and accessible?")
+
+
+def require_docker_running() -> None:
+    """Ensure Docker CLI exists and daemon is reachable; print hints and exit if not."""
+    import typer
+
+    from hud.utils.hud_console import hud_console
+
+    docker_path: str | None = shutil.which("docker")
+    if not docker_path:
+        hud_console.error("Docker CLI not found")
+        hud_console.info("Install Docker Desktop (Windows/macOS) or Docker Engine (Linux)")
+        hud_console.hint("After installation, start Docker and re-run this command")
+        raise typer.Exit(1)
+
+    try:
+        result = subprocess.run(  # noqa: UP022, S603
+            [docker_path, "info"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+            timeout=8,
+            check=False,
+        )
+        if result.returncode == 0:
+            return
+
+        error_text = (result.stderr or "") + "\n" + (result.stdout or "")
+        _emit_docker_hints(error_text)
+        raise typer.Exit(1)
+    except FileNotFoundError as e:
+        hud_console.error("Docker CLI not found on PATH")
+        hud_console.hint("Install Docker and ensure 'docker' is on your PATH")
+        raise typer.Exit(1) from e
+    except subprocess.TimeoutExpired as e:
+        hud_console.error("Docker did not respond in time")
+        hud_console.hint(
+            "Is Docker running? Open Docker Desktop and wait until it reports 'Running'"
+        )
+        raise typer.Exit(1) from e
+    except Exception as e:
+        hud_console.error(f"Docker check failed: {e}")
+        hud_console.hint("Is the Docker daemon running?")
+        raise typer.Exit(1) from e

hud/rl/train.py

@@ -232,6 +232,9 @@ async def train(config: Config, tasks: list[Task]) -> None:
                     )
                     learner.save(str(checkpoint_path))
 
+                    # Wait for 6 seconds to ensure the checkpoint is saved
+                    await asyncio.sleep(6)
+
                     adapter_name = f"{config.adapter_prefix}-{checkpoint_id}"
                     if vllm.load_adapter(adapter_name, str(checkpoint_path)):
                         actor.update_adapter(adapter_name)

hud/rl/vllm_adapter.py

@@ -35,20 +35,38 @@ class VLLMAdapter:
         url = f"{self.base_url}/load_lora_adapter"
         headers = {"Authorization": f"Bearer {self.api_key}", "Content-Type": "application/json"}
         payload = {"lora_name": adapter_name, "lora_path": adapter_path}
-
-        try:
-            response = requests.post(
-                url, headers=headers, data=json.dumps(payload), timeout=timeout
-            )
-            response.raise_for_status()
-
-            self.current_adapter = adapter_name
-            hud_console.info(f"[VLLMAdapter] Loaded adapter: {adapter_name}")
-            return True
-
-        except requests.exceptions.RequestException as e:
-            hud_console.error(f"[VLLMAdapter] Failed to load adapter {adapter_name}: {e}")
-            return False
+        # Implement exponential backoff for retrying the adapter load request.
+        max_retries = 5
+        backoff_factor = 2
+        delay = 1  # initial delay in seconds
+
+        for attempt in range(1, max_retries + 1):
+            try:
+                response = requests.post(
+                    url, headers=headers, data=json.dumps(payload), timeout=timeout
+                )
+                response.raise_for_status()
+
+                self.current_adapter = adapter_name
+                hud_console.info(f"[VLLMAdapter] Loaded adapter: {adapter_name}")
+                return True
+
+            except requests.exceptions.RequestException as e:
+                if attempt == max_retries:
+                    hud_console.error(
+                        f"[VLLMAdapter] Failed to load adapter {adapter_name} after {attempt} attempts: {e}"  # noqa: E501
+                    )
+                    return False
+                else:
+                    hud_console.warning(
+                        f"[VLLMAdapter] Load adapter {adapter_name} failed (attempt {attempt}/{max_retries}): {e}. Retrying in {delay} seconds...",  # noqa: E501
+                    )
+                    import time
+
+                    time.sleep(delay)
+                    delay *= backoff_factor
+
+        return False
 
     def unload_adapter(self, adapter_name: str) -> bool:
         """

hud/utils/tests/test_version.py

@@ -5,4 +5,4 @@ def test_import():
     """Test that the package can be imported."""
     import hud
 
-    assert hud.__version__ == "0.4.30"
+    assert hud.__version__ == "0.4.32"

hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.30"
+__version__ = "0.4.32"

{hud_python-0.4.30.dist-info → hud_python-0.4.32.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.30
+Version: 0.4.32
 Summary: SDK for the HUD platform.
 Project-URL: Homepage, https://github.com/hud-evals/hud-python
 Project-URL: Bug Tracker, https://github.com/hud-evals/hud-python/issues
@@ -35,15 +35,20 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Python: <3.13,>=3.11
+Requires-Dist: anthropic
+Requires-Dist: datasets>=2.14.0
 Requires-Dist: httpx<1,>=0.23.0
 Requires-Dist: hud-fastmcp-python-sdk>=0.1.2
 Requires-Dist: hud-mcp-python-sdk>=3.13.2
 Requires-Dist: hud-mcp-use-python-sdk>=2.3.16
+Requires-Dist: numpy>=1.24.0
+Requires-Dist: openai
 Requires-Dist: opentelemetry-api>=1.34.1
 Requires-Dist: opentelemetry-exporter-otlp-proto-http>=1.34.1
 Requires-Dist: opentelemetry-instrumentation-mcp==0.47.0
 Requires-Dist: opentelemetry-sdk>=1.34.1
 Requires-Dist: pathspec>=0.12.1
+Requires-Dist: pillow>=11.1.0
 Requires-Dist: prompt-toolkit==3.0.51
 Requires-Dist: pydantic-settings<3,>=2
 Requires-Dist: pydantic<3,>=2
@@ -54,8 +59,6 @@ Requires-Dist: typer>=0.9.0
 Requires-Dist: watchfiles>=0.21.0
 Requires-Dist: wrapt>=1.14.0
 Provides-Extra: agent
-Requires-Dist: anthropic; extra == 'agent'
-Requires-Dist: datasets>=2.14.0; extra == 'agent'
 Requires-Dist: dotenv>=0.9.9; extra == 'agent'
 Requires-Dist: ipykernel; extra == 'agent'
 Requires-Dist: ipython<9; extra == 'agent'
@@ -64,12 +67,7 @@ Requires-Dist: jupyter-core; extra == 'agent'
 Requires-Dist: langchain; extra == 'agent'
 Requires-Dist: langchain-anthropic; extra == 'agent'
 Requires-Dist: langchain-openai; extra == 'agent'
-Requires-Dist: numpy>=1.24.0; extra == 'agent'
-Requires-Dist: openai; extra == 'agent'
-Requires-Dist: pillow>=11.1.0; extra == 'agent'
 Provides-Extra: agents
-Requires-Dist: anthropic; extra == 'agents'
-Requires-Dist: datasets>=2.14.0; extra == 'agents'
 Requires-Dist: dotenv>=0.9.9; extra == 'agents'
 Requires-Dist: ipykernel; extra == 'agents'
 Requires-Dist: ipython<9; extra == 'agents'
@@ -78,13 +76,8 @@ Requires-Dist: jupyter-core; extra == 'agents'
 Requires-Dist: langchain; extra == 'agents'
 Requires-Dist: langchain-anthropic; extra == 'agents'
 Requires-Dist: langchain-openai; extra == 'agents'
-Requires-Dist: numpy>=1.24.0; extra == 'agents'
-Requires-Dist: openai; extra == 'agents'
-Requires-Dist: pillow>=11.1.0; extra == 'agents'
 Provides-Extra: dev
 Requires-Dist: aiodocker>=0.24.0; extra == 'dev'
-Requires-Dist: anthropic; extra == 'dev'
-Requires-Dist: datasets>=2.14.0; extra == 'dev'
 Requires-Dist: dotenv>=0.9.9; extra == 'dev'
 Requires-Dist: inspect-ai>=0.3.80; extra == 'dev'
 Requires-Dist: ipykernel; extra == 'dev'
@@ -94,8 +87,6 @@ Requires-Dist: jupyter-core; extra == 'dev'
 Requires-Dist: langchain; extra == 'dev'
 Requires-Dist: langchain-anthropic; extra == 'dev'
 Requires-Dist: langchain-openai; extra == 'dev'
-Requires-Dist: numpy>=1.24.0; extra == 'dev'
-Requires-Dist: openai; extra == 'dev'
 Requires-Dist: pillow>=11.1.0; extra == 'dev'
 Requires-Dist: playwright; extra == 'dev'
 Requires-Dist: pyautogui>=0.9.54; extra == 'dev'
@@ -108,9 +99,7 @@ Requires-Dist: ruff>=0.11.8; extra == 'dev'
 Requires-Dist: setuptools; extra == 'dev'
 Requires-Dist: textdistance<5,>=4.5.0; extra == 'dev'
 Provides-Extra: rl
-Requires-Dist: anthropic; extra == 'rl'
 Requires-Dist: bitsandbytes>=0.41.0; (sys_platform == 'linux') and extra == 'rl'
-Requires-Dist: datasets>=2.14.0; extra == 'rl'
 Requires-Dist: dotenv>=0.9.9; extra == 'rl'
 Requires-Dist: ipykernel; extra == 'rl'
 Requires-Dist: ipython<9; extra == 'rl'
@@ -120,10 +109,7 @@ Requires-Dist: langchain; extra == 'rl'
 Requires-Dist: langchain-anthropic; extra == 'rl'
 Requires-Dist: langchain-openai; extra == 'rl'
 Requires-Dist: liger-kernel>=0.5.0; (sys_platform == 'linux') and extra == 'rl'
-Requires-Dist: numpy>=1.24.0; extra == 'rl'
-Requires-Dist: openai; extra == 'rl'
 Requires-Dist: peft>=0.17.1; extra == 'rl'
-Requires-Dist: pillow>=11.1.0; extra == 'rl'
 Requires-Dist: vllm==0.10.1.1; extra == 'rl'
 Description-Content-Type: text/markdown
 
@@ -239,21 +225,34 @@ The above example let's the agent play 2048 ([See replay](https://app.hud.so/tra
 
 ## Reinforcement Learning with GRPO
 
-This is a Qwen-2.5-3B agent training a policy on the [`text-2048`](environments/text_2048/) environment (see above) using [Verifiers](rl/):
+This is a Qwen‑2.5‑VL‑3B agent training a policy on the 2048-basic browser environment:
 
 ![RL curve](https://raw.githubusercontent.com/hud-evals/hud-python/main/docs/src/images/rl_2.png)
 
-To start training, check out the [`rl/README.md`](rl/README.md) folder:
+Train with the new interactive `hud rl` flow:
 
 ```bash
-git clone https://github.com/hud-evals/hud-python
-cd hud-python/rl
-python train_2048.py
+# Install CLI with RL extras
+uv tool install "hud-python[rl]"
+
+# Option A: Run directly from a HuggingFace dataset
+hud rl hud-evals/basic-2048
+
+# Option B: Download first, modify, then train
+hud get hud-evals/basic-2048
+hud rl basic-2048.jsonl
+
+# Optional: baseline evaluation
+hud eval basic-2048.jsonl
 ```
 
-Any hud MCP environment and evaluation works with our RL pipeline. Even our remote configurations!
+Supports multi‑turn RL for both:
+- Language‑only models (e.g., `Qwen/Qwen2.5-7B-Instruct`)
+- Vision‑Language models (e.g., `Qwen/Qwen2.5-VL-3B-Instruct`)
+
+By default, `hud rl` provisions a persistant server and trainer in the cloud, streams telemetry to `app.hud.so`, and lets you monitor/manage models at `app.hud.so/models`. Use `--local` to run entirely on your machines (typically 2+ GPUs: one for vLLM, the rest for training).
 
-> The [`rl/README.md`](rl/README.md) walks you through several examples of RL training and takes less than 15 minutes to set up for your custom agent!
+Any HUD MCP environment and evaluation works with our RL pipeline (including remote configurations). See the guided docs: `https://docs.hud.so/train-agents/quickstart`.
 
 ## Benchmarking Agents
 

{hud_python-0.4.30.dist-info → hud_python-0.4.32.dist-info}/RECORD

@@ -2,14 +2,14 @@ hud/__init__.py,sha256=JMDFUE1pP0J1Xl_miBdt7ERvoffZmTzSFe8yxz512A8,552
 hud/__main__.py,sha256=YR8Dq8OhINOsVfQ55PmRXXg4fEK84Rt_-rMtJ5rvhWo,145
 hud/settings.py,sha256=sMS31iW1m-5VpWk-Blhi5-obLcUA0fwxWE1GgJz-vqU,2708
 hud/types.py,sha256=Cn9suZ_ZitLnxmnknfbCYVvmLsXRWI56kJ3LXtdfI6M,10157
-hud/version.py,sha256=4eP4L4p9KEmmsjMwwEEDGX8nC__z95NS2DhdflB5qqw,105
+hud/version.py,sha256=amJJuAW2iW2bL8JO7j8MSy3wzSXv2pKn1GXk85UJEVM,105
 hud/agents/__init__.py,sha256=UoIkljWdbq4bM0LD-mSaw6w826EqdEjOk7r6glNYwYQ,286
 hud/agents/base.py,sha256=_u1zR3gXzZ1RlTCUYdMcvgHqdJBC4-AB1lZt0yBx8lg,35406
 hud/agents/claude.py,sha256=wHiw8iAnjnRmZyKRKcOhagCDQMhz9Z6rlSBWqH1X--M,15781
 hud/agents/grounded_openai.py,sha256=U-FHjB2Nh1_o0gmlxY5F17lWJ3oHsNRIB2a7z-IKB64,11231
 hud/agents/langchain.py,sha256=1EgCy8jfjunsWxlPC5XfvfLS6_XZVrIF1ZjtHcrvhYw,9584
 hud/agents/openai.py,sha256=ovARRWNuHqKkZ2Q_OCYSVCIZckrh8XY2jUB2p2x1m88,14259
-hud/agents/openai_chat_generic.py,sha256=H33CE8W3eYmQOJp8C5VcrvnG2AjBu02TWigHctEZNoI,11604
+hud/agents/openai_chat_generic.py,sha256=7n7timn3fvNRnL2xzWyOTeNTchej2r9cAL1mU6YnFdY,11605
 hud/agents/misc/__init__.py,sha256=BYi4Ytp9b_vycpZFXnr5Oyw6ncKLNNGml8Jrb7bWUb4,136
 hud/agents/misc/response_agent.py,sha256=OJdQJ76jP9xxQxVYJ-qPcdBxvFr8ABcwbP1f1I5zU5A,3227
 hud/agents/tests/__init__.py,sha256=W-O-_4i34d9TTyEHV-O_q1Ai1gLhzwDaaPo02_TWQIY,34
@@ -27,20 +27,21 @@ hud/cli/debug.py,sha256=jtFW8J5F_3rhq1Hf1_SkJ7aLS3wjnyIs_LsC8k5cnzc,14200
 hud/cli/dev.py,sha256=56vQdH9oe_XGnOcRcFbNIsLEoBnpCl1eANlRFUeddHQ,31734
 hud/cli/eval.py,sha256=W_eY4uoIQwHcSCvxNaQeRfWC10uQA1UhBWiNQzQPuXM,22694
 hud/cli/get.py,sha256=sksKrdzBGZa7ZuSoQkc0haj-CvOGVSSikoVXeaUd3N4,6274
-hud/cli/init.py,sha256=XswZB2ZnzdE0pxP1kRmO3bHfWGCrKAyrME34ZyPzs98,19715
+hud/cli/init.py,sha256=McZwpxZMXD-It_PXINCUy-SwUaPiQ7jdpSU5-F-caO8,19671
 hud/cli/list_func.py,sha256=EVi2Vc3Lb3glBNJxFx4MPnZknZ4xmuJz1OFg_dc8a_E,7177
 hud/cli/pull.py,sha256=Vd1l1-IwskyACzhtC8Df1SYINUZEYmFxrLl0s9cNN6c,12151
-hud/cli/push.py,sha256=JXUxu1QGU7BPWb0erSJq42CIq0sLbaDAO42yYDcvA1g,18347
+hud/cli/push.py,sha256=dmjF-hGlMfq73tquDxsTuM9t50zrkE9PFJqW5vRmYSw,18380
 hud/cli/remove.py,sha256=8vGQyXDqgtjz85_vtusoIG8zurH4RHz6z8UMevQRYM4,6861
 hud/cli/flows/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud/cli/flows/tasks.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud/cli/rl/__init__.py,sha256=_7g1OLhAsDP5gtTSaD35byvlVEYi58K_rQOn_7Kvy-M,22894
+hud/cli/flows/tasks.py,sha256=cOIXs03_iAS68nIYj5NpzqlSh1vSbIZtp3dnt8PEyl4,8855
+hud/cli/rl/__init__.py,sha256=BeqXdmzPwVBptz4j796XJRxSC5B_9tQta5aKd0jDMvo,5000
 hud/cli/rl/config.py,sha256=iNhCxotM33OEiP9gqPvn8A_AxrBVe6fcFCQTvc13xzA,2884
 hud/cli/rl/display.py,sha256=hqJVGmO9csYinladhZwjF-GMvppYWngxDHajTyIJ_gM,5214
 hud/cli/rl/gpu.py,sha256=peXS-NdUF5RyuSs0aZoCzGLboneBUpCy8f9f99WMrG0,2009
 hud/cli/rl/gpu_utils.py,sha256=H5ckPwgj5EVP3yJ5eVihR5R7Y6Gp6pt8ZUfWCCwcLG4,11072
+hud/cli/rl/local_runner.py,sha256=GssmDgCxGfFsi31aFj22vwCiwa9ELllEwQjbActxSXY,21514
 hud/cli/rl/presets.py,sha256=DzOO82xL5QyzdVtlX-Do1CODMvDz9ILMPapjU92jcZg,3051
-hud/cli/rl/remote_runner.py,sha256=dOl6UBTJoE3lDGydETG1oXXRF44Bsy4JzeL62VxeMRM,12710
+hud/cli/rl/remote_runner.py,sha256=nSLPkBek1pBQiOYg-yqkc3La5dkQEN9xn9DaauSlYAA,13747
 hud/cli/rl/rl_api.py,sha256=INJobvSa50ccR037u_GPsDa_9WboWyNwqEaoh9hcXj0,4306
 hud/cli/rl/vllm.py,sha256=Gq_M6KsQArGz7FNIdemuM5mk16mu3xe8abpO2GCCuOE,6093
 hud/cli/tests/__init__.py,sha256=ZrGVkmH7DHXGqOvjOSNGZeMYaFIRB2K8c6hwr8FPJ-8,68
@@ -61,7 +62,7 @@ hud/cli/tests/test_registry.py,sha256=-o9MvQTcBElteqrg0XW8Bg59KrHCt88ZyPqeaAlyyT
 hud/cli/tests/test_utils.py,sha256=_oa2lTvgqJxXe0Mtovxb8x-Sug-f6oJJKvG67r5pFtA,13474
 hud/cli/utils/__init__.py,sha256=L6s0oNzY2LugGp9faodCPnjzM-ZUorUH05-HmYOq5hY,35
 hud/cli/utils/cursor.py,sha256=fy850p0rVp5k_1wwOCI7rK1SggbselJrywFInSQ2gio,3009
-hud/cli/utils/docker.py,sha256=VTUcoPqxh3uXOgvL6NSqYiSDhyCPRp3jTfFbnIDwumg,3774
+hud/cli/utils/docker.py,sha256=-nAj7wRRIilbezG0-pCHA2-tleoqUJN9sDXHxvMWilU,7331
 hud/cli/utils/environment.py,sha256=y_c0ohxWrM054ZKid0KOQPzs2M2vh985AsumPG2wTPc,4282
 hud/cli/utils/interactive.py,sha256=tcwp9HkAyr2_GiM3Raba4h0P_OgCksQKram80BucPo4,16546
 hud/cli/utils/logging.py,sha256=DyOWuzZUg6HeKCqfs6ufb703XS3bW4G2pzaXVAvDqvA,9018
@@ -113,10 +114,10 @@ hud/rl/chat_template.jinja,sha256=XTdzI8oFGEcSA-exKxyHaprwRDmX5Am1KEb0VxvUc6U,49
 hud/rl/config.py,sha256=PAKYPCsKl8yg_j3gJSE5SJUgLM7j0lFy0K_Vt4-otDM,5384
 hud/rl/distributed.py,sha256=8avhrb0lHYkhW22Z7MfkqSnlczWj5jMrUMEtkcoCf74,2473
 hud/rl/learner.py,sha256=FKIgIIghsNiDr_g090xokOO_BxNmTSj1O-TSJzIq_Uw,24703
-hud/rl/train.py,sha256=oZQGo0Wvb2LSrhh-7FLOsGCvI4G4AjgAAvF9P0k9l1Q,13436
+hud/rl/train.py,sha256=ZigkUKj-I1nsYmFByZprqaoDZ88LVDH-6auYneEPOsA,13555
 hud/rl/types.py,sha256=lrLKo7iaqodYth2EyeuOQfLiuzXfYM2eJjPmpObrD7c,3965
 hud/rl/utils.py,sha256=IsgVUUibxnUzb32a4mu1sYrgJC1CwoG9E-Dd5y5VDOA,19115
-hud/rl/vllm_adapter.py,sha256=TBNo5lyNzszg6ATk9JoEZAm-xk_tcUJmq9YXwF1NB5w,3961
+hud/rl/vllm_adapter.py,sha256=O2_TdTGIyNr9zRGhCw18XWjOKYzEM3049wvlyL2x0sc,4751
 hud/rl/tests/__init__.py,sha256=PXmD3Gs6xOAwaYKb4HnwZERDjX05N1QF-aU6ya0dBtE,27
 hud/rl/tests/test_learner.py,sha256=qfSHFFROteRb98TjBuAKjFmZjCGfuWXPysVvTAWJ7wQ,6025
 hud/rl/utils/start_vllm_server.sh,sha256=ThPokrLK_Qm_uh916fHXXBfMlw1TC97P57-AEI5MuOc,910
@@ -197,10 +198,10 @@ hud/utils/tests/test_init.py,sha256=2QLQSGgyP9wJhOvPCusm_zjJad0qApOZi1BXpxcdHXQ,
 hud/utils/tests/test_mcp.py,sha256=0pUa16mL-bqbZDXp5NHBnt1gO5o10BOg7zTMHZ1DNPM,4023
 hud/utils/tests/test_progress.py,sha256=QSF7Kpi03Ff_l3mAeqW9qs1nhK50j9vBiSobZq7T4f4,7394
 hud/utils/tests/test_telemetry.py,sha256=5jl7bEx8C8b-FfFUko5pf4UY-mPOR-9HaeL98dGtVHM,2781
-hud/utils/tests/test_version.py,sha256=jpqi_MLI03sxvfOaaX6i6wE6qhR1oLlH1VDVqdtAj6U,160
+hud/utils/tests/test_version.py,sha256=HVazB_GTYA2lJ3Oc10a6KPOvPoBBQOGuYoJl7yIazWw,160
 hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud_python-0.4.30.dist-info/METADATA,sha256=u3lRTRb4gv7NfN8iEJoEPqLAizjXU_RBWNfPNfwXiJM,21069
-hud_python-0.4.30.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-hud_python-0.4.30.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
-hud_python-0.4.30.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
-hud_python-0.4.30.dist-info/RECORD,,
+hud_python-0.4.32.dist-info/METADATA,sha256=Ve-zWpi0dWcv8bjC4wCcNZnQtsrlFTOwAaWs4RmQ_W0,20861
+hud_python-0.4.32.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+hud_python-0.4.32.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
+hud_python-0.4.32.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
+hud_python-0.4.32.dist-info/RECORD,,