hud-python 0.4.29__py3-none-any.whl → 0.4.31__py3-none-any.whl

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/native/comparator.py

@@ -513,11 +513,11 @@ def make_alias_tool(name: str, preset_mode: ComparisonMode, description: str) ->
 
 
 # Create MCP server
-comparator_server = MCPServer(name="comparator")
+comparator = MCPServer(name="comparator")
 
 # Register main tool
-comparator_server.add_tool(SubmitTool())
-comparator_server.add_tool(CompareTool())
+comparator.add_tool(SubmitTool())
+comparator.add_tool(CompareTool())
 
 # Register aliases - these are just thin wrappers
 ALIASES = [
@@ -534,13 +534,13 @@ ALIASES = [
 
 for name, mode, desc in ALIASES:
     AliasTool = make_alias_tool(name, mode, desc)
-    comparator_server.add_tool(AliasTool())
+    comparator.add_tool(AliasTool())
 
 # Export for mounting
-__all__ = ["comparator_server"]
+__all__ = ["comparator"]
 
 
 if __name__ == "__main__":
     # Run as standalone server
     logger.info("Starting Comparator MCP Server...")
-    comparator_server.run()
+    comparator.run()

hud/native/tests/test_comparator.py

@@ -11,7 +11,7 @@ from hud.native.comparator import (
     ComparisonResult,
     DataType,
     auto_select_mode,
-    comparator_server,
+    comparator,
     detect_type,
     extract_boolean,
     extract_json,
@@ -321,10 +321,10 @@ class TestAliasTools:
     @pytest.mark.asyncio
     async def test_aliases_work(self):
         """Test that aliases are properly registered and work."""
-        from hud.native.comparator import comparator_server
+        from hud.native.comparator import comparator
 
         # Check that aliases are registered
-        tool_names = [t.name for t in comparator_server._tool_manager._tools.values()]
+        tool_names = [t.name for t in comparator._tool_manager._tools.values()]
 
         expected_aliases = [
             "compare_exact",
@@ -433,7 +433,7 @@ class TestAliasPreprocessing:
     @pytest.mark.asyncio
     async def test_json_alias_preprocessing(self):
         """Test JSON extraction in compare_json tool."""
-        tools = {t.name: t for t in comparator_server._tool_manager._tools.values()}
+        tools = {t.name: t for t in comparator._tool_manager._tools.values()}
         json_tool = tools["compare_json"]
 
         assert isinstance(json_tool, FunctionTool)
@@ -448,7 +448,7 @@ class TestAliasPreprocessing:
     @pytest.mark.asyncio
     async def test_numeric_alias_preprocessing(self):
         """Test number extraction in numeric tools."""
-        tools = {t.name: t for t in comparator_server._tool_manager._tools.values()}
+        tools = {t.name: t for t in comparator._tool_manager._tools.values()}
 
         # Float tool
         float_tool = tools["compare_float"]
@@ -471,7 +471,7 @@ class TestAliasPreprocessing:
     @pytest.mark.asyncio
     async def test_boolean_alias_preprocessing(self):
         """Test boolean extraction in compare_boolean tool."""
-        tools = {t.name: t for t in comparator_server._tool_manager._tools.values()}
+        tools = {t.name: t for t in comparator._tool_manager._tools.values()}
         bool_tool = tools["compare_boolean"]
 
         assert isinstance(bool_tool, FunctionTool)
@@ -485,7 +485,7 @@ class TestAliasPreprocessing:
     @pytest.mark.asyncio
     async def test_list_alias_preprocessing(self):
         """Test list extraction in compare_list tool."""
-        tools = {t.name: t for t in comparator_server._tool_manager._tools.values()}
+        tools = {t.name: t for t in comparator._tool_manager._tools.values()}
         list_tool = tools["compare_list"]
 
         assert isinstance(list_tool, FunctionTool)
@@ -499,7 +499,7 @@ class TestAliasPreprocessing:
     @pytest.mark.asyncio
     async def test_complex_llm_output(self):
         """Test extraction from complex LLM outputs with reasoning."""
-        tools = {t.name: t for t in comparator_server._tool_manager._tools.values()}
+        tools = {t.name: t for t in comparator._tool_manager._tools.values()}
         json_tool = tools["compare_json"]
 
         llm_output = """

hud/native/tests/test_native_init.py

@@ -8,12 +8,12 @@ class TestNativeInit:
 
     def test_comparator_server_import(self):
         """Test that comparator server can be imported."""
-        from hud.native.comparator import comparator_server
+        from hud.native.comparator import comparator
         from hud.server import MCPServer
 
         # Verify comparator is an MCPServer instance
-        assert isinstance(comparator_server, MCPServer)
-        assert comparator_server.name == "comparator"
+        assert isinstance(comparator, MCPServer)
+        assert comparator.name == "comparator"
 
     def test_all_exports(self):
         """Test that __all__ is properly defined."""
@@ -31,11 +31,11 @@ class TestNativeInit:
 
     def test_comparator_tools_registered(self):
         """Test that comparator server has tools registered."""
-        from hud.native.comparator import comparator_server
+        from hud.native.comparator import comparator
 
         # The server should have tools registered
         # We can check that the tool manager has tools
-        tool_names = [t.name for t in comparator_server._tool_manager._tools.values()]
+        tool_names = [t.name for t in comparator._tool_manager._tools.values()]
 
         # Should have the main compare tool
         assert "compare" in tool_names
@@ -64,16 +64,18 @@ class TestNativeInit:
 
     def test_comparator_tool_functionality(self):
         """Test that we can get the CompareTool from the comparator."""
-        from hud.native.comparator import comparator_server
-        from hud.tools import BaseTool
+        from hud.native.comparator import comparator
 
         # Get the compare tool
         compare_tool = None
-        for tool in comparator_server._tool_manager._tools.values():
+        for tool in comparator._tool_manager._tools.values():
             if tool.name == "compare":
                 compare_tool = tool
                 break
 
         assert compare_tool is not None
-        assert isinstance(compare_tool, BaseTool)
-        assert hasattr(compare_tool, "__call__")
+        # FastMCP wraps tools as FunctionTool instances
+        assert hasattr(compare_tool, "name")
+        assert compare_tool.name == "compare"
+        # FunctionTool has a 'fn' attribute for the callable
+        assert hasattr(compare_tool, "fn") or hasattr(compare_tool, "__call__")

hud/rl/README.md

@@ -1,6 +1,6 @@
 We suggest running hud rl (or with the --local flag) for optimal hyperparameters and native HuggingFace running.
 
-Install:
+However, to run this independently, sping up an instance with at least 2 GPUs and run:
 ```bash
 sudo apt-get update -y && sudo apt-get install -y cuda-toolkit-12-6
 uv pip install -e .[rl]
@@ -8,8 +8,7 @@ uv pip install ninja
 uv pip install flash-attn --no-build-isolation
 ```
 
-
-However, if you want to run the training directly, launch a vllm server with:
+Launch a vllm server with:
 ```bash
 export VLLM_ALLOW_RUNTIME_LORA_UPDATING=True
 export TOKENIZERS_PARALLELISM=false

hud/rl/learner.py

@@ -103,10 +103,12 @@ class GRPOLearner:
 
         # Load processor/tokenizer based on model type
         if is_vl_model:
+            # Some environments require remote code for Qwen2.5-VL processors
             processor = AutoProcessor.from_pretrained(
                 model_cfg.base_model,
                 min_pixels=model_cfg.min_pixels,
                 max_pixels=model_cfg.max_pixels,
+                trust_remote_code=True,
             )
         else:
             processor = AutoTokenizer.from_pretrained(model_cfg.base_model)
@@ -123,6 +125,7 @@ class GRPOLearner:
                 model_cfg.base_model,
                 torch_dtype=torch.bfloat16,
                 attn_implementation=attn_implementation,
+                trust_remote_code=True,
             )
             self.log(f"Using {attn_implementation} for attention")
         except (ImportError, ValueError) as 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__ == "0.4.29"
+    assert hud.__version__ == "0.4.31"

hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.29"
+__version__ = "0.4.31"

{hud_python-0.4.29.dist-info → hud_python-0.4.31.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.29
+Version: 0.4.31
 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-0.4.29.dist-info → hud_python-0.4.31.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=I65QTO8NMhRiDAN0vYH7G47SRwt713WHRXjHgx1VsL0,105
+hud/version.py,sha256=wQqwRzN9OMugX4H2oCdAp5bWBbjfBRrvQGeo_0_uaYs,105
 hud/agents/__init__.py,sha256=UoIkljWdbq4bM0LD-mSaw6w826EqdEjOk7r6glNYwYQ,286
-hud/agents/base.py,sha256=y3Aq3DtsDT5X2Q-_C-jqxFtDcHUlnBDOczmZIhqC0-w,35118
+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=T2HTZaSUmvqBhYIfurtreb-gxnlVBqDO64GySu_uFIA,11514
+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/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=HRIWMx5Dmmu2IorrmcyVu-Bs_3B45pwEsCC67M1z3Oo,22670
+hud/cli/flows/tasks.py,sha256=8r-51oon3anwMAi5cyAOgC3iB48jnqlOUO3iTWmqsyI,6372
+hud/cli/rl/__init__.py,sha256=BeqXdmzPwVBptz4j796XJRxSC5B_9tQta5aKd0jDMvo,5000
 hud/cli/rl/config.py,sha256=iNhCxotM33OEiP9gqPvn8A_AxrBVe6fcFCQTvc13xzA,2884
-hud/cli/rl/display.py,sha256=OTFcoIn5k3ehfi5-PmuClIkY04f4tpSMuNf7yd-bBr4,5212
+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=ILjWpjbdZyH7db-C-x1T0GwsrLu3PBKx3IoTZl-qamg,12733
+hud/cli/rl/remote_runner.py,sha256=Umyjjbtw6ikyNNfVGjn6sY6Qnta8Uc1KC7KzMgFaJVw,13146
 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
@@ -92,10 +93,10 @@ hud/datasets/utils.py,sha256=hdZfjWH5l3FVJaWBSHEEpjujAG7DqEam_vHgslL8MLs,4279
 hud/misc/__init__.py,sha256=m_pprQQ-G-Y0Sd0NEiR8MtAMbElnuFZ2OWT8TXrw7c4,43
 hud/misc/claude_plays_pokemon.py,sha256=IthAkjDVr2Q-GNvX-QLJyMzN7-0pHqqJbagGNv2m7yo,10453
 hud/native/__init__.py,sha256=TqM0KaiQnDb2Nv1zOgpEMiLVq8JPd4j_aaK4rUZ0IiA,232
-hud/native/comparator.py,sha256=A16wFGINLHPyuD23e8sEYxhRwWCUBYzYLb6TpvzJG9c,18297
+hud/native/comparator.py,sha256=GCHs7iZa0fB425es6vvG91UW4yrbY6-BsWdabYJaNA4,18255
 hud/native/tests/__init__.py,sha256=gBTLMm6w5f6D-02Se2WleYsEEYyFt95JDcFzp3C2L_k,40
-hud/native/tests/test_comparator.py,sha256=x1gFLXEDRIiJhH8tg5Rd3ptY-modYaHgSm6-hCJ1EdY,18568
-hud/native/tests/test_native_init.py,sha256=00zSF6jBLIYYbDaslI7fk-ANTEofIC9SPrRtX942WM4,2738
+hud/native/tests/test_comparator.py,sha256=pDch3r3xDi2o5YXF_bkoLfIdHcCjse3foAaqyr7PzkQ,18512
+hud/native/tests/test_native_init.py,sha256=Z-2dinbQYEkrbCcfBrBOLGdpXtWWOtkfPzp7ZKri68Y,2839
 hud/otel/__init__.py,sha256=ii17ayoWiS5vAhA7UAmZ8TkmP52gs2pWyHsD46-uYbE,1003
 hud/otel/collector.py,sha256=jLZymZ8r7xt2VDuWexfbnT7PY1-0aiyLMgjBy8KDY1M,4497
 hud/otel/config.py,sha256=mricuAmtFd1yIfOYKw2aHI-u4piku0GXHWv6hjsWQLM,6806
@@ -105,18 +106,18 @@ hud/otel/instrumentation.py,sha256=fsFG9W89RdewFDxWKN9Ft4GUb7WbIKpfucTc16WxaZU,5
 hud/otel/processors.py,sha256=-gGRbwifplcExDQBLfx_9tqWreDImULJNcENgO9q7VU,4700
 hud/otel/tests/__init__.py,sha256=VNJKBMaxTtbn7trW-1Ph50zCvCok_wTSGcI1HD6GOLA,43
 hud/otel/tests/test_processors.py,sha256=np0R4ssd9j6LJSJykJ5bNjl0POwNYNhgb7BqOZHwcMY,6778
-hud/rl/README.md,sha256=OLOmRGOWuQGE-xjLqQJbazzX8ygUG17ECP6kjbM2C0g,1163
+hud/rl/README.md,sha256=uFRpNFaEY8paq9k1C4miF7AGnbqHTGAsPmpcf9JIEeA,1189
 hud/rl/__init__.py,sha256=yYL7U1WV6L3mr3Hig48-4lhnryTaWj4nCXm4lG5vrYI,25
 hud/rl/actor.py,sha256=0YChXyxCz1wVBQ9lKb7vSl64_HQ24-DmYqCCxuORzJc,6747
 hud/rl/buffer.py,sha256=xz4FlvO9l945VsSS4lzRFMwH3rA9HafgbUfADSauXok,15210
 hud/rl/chat_template.jinja,sha256=XTdzI8oFGEcSA-exKxyHaprwRDmX5Am1KEb0VxvUc6U,4965
 hud/rl/config.py,sha256=PAKYPCsKl8yg_j3gJSE5SJUgLM7j0lFy0K_Vt4-otDM,5384
 hud/rl/distributed.py,sha256=8avhrb0lHYkhW22Z7MfkqSnlczWj5jMrUMEtkcoCf74,2473
-hud/rl/learner.py,sha256=JBlzPFgX16uk6f6xYINLXWdDD1i1tiyzm6GLryrzgYg,24545
-hud/rl/train.py,sha256=oZQGo0Wvb2LSrhh-7FLOsGCvI4G4AjgAAvF9P0k9l1Q,13436
+hud/rl/learner.py,sha256=FKIgIIghsNiDr_g090xokOO_BxNmTSj1O-TSJzIq_Uw,24703
+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=cXnWb-tAwTPK40YNpgJ0EwXe7op_7xDXBGf7sf5-ECU,160
+hud/utils/tests/test_version.py,sha256=qN98qLlKKeM-W-AM1q1s0Lci8phBQ7SUd_L-0yVmujA,160
 hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud_python-0.4.29.dist-info/METADATA,sha256=xT3yJCdo-1-xJ1q54lgIyFALitPOiU2-YSParihkYvM,21069
-hud_python-0.4.29.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-hud_python-0.4.29.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
-hud_python-0.4.29.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
-hud_python-0.4.29.dist-info/RECORD,,
+hud_python-0.4.31.dist-info/METADATA,sha256=UZMnmiRCMmHmXOD0BrkQs7Caqk5t8HcllxWPwR2SSzc,20861
+hud_python-0.4.31.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+hud_python-0.4.31.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
+hud_python-0.4.31.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
+hud_python-0.4.31.dist-info/RECORD,,