hud-python 0.4.29__tar.gz → 0.4.30__tar.gz

{hud_python-0.4.29 → hud_python-0.4.30}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.29
+Version: 0.4.30
 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

{hud_python-0.4.29 → hud_python-0.4.30}/hud/agents/base.py

@@ -153,16 +153,24 @@ class MCPAgent(ABC):
             if task.setup_tool:
                 if isinstance(task.setup_tool, list):
                     for tool in task.setup_tool:
-                        if self.agent_tools and tool.name not in self.agent_tools:
+                        if not self.agent_tools or (
+                            self.agent_tools and tool.name not in self.agent_tools
+                        ):
                             self.lifecycle_tools.append(tool.name)
-                elif self.agent_tools and task.setup_tool.name not in self.agent_tools:
+                elif not self.agent_tools or (
+                    self.agent_tools and task.setup_tool.name not in self.agent_tools
+                ):
                     self.lifecycle_tools.append(task.setup_tool.name)
             if task.evaluate_tool:
                 if isinstance(task.evaluate_tool, list):
                     for tool in task.evaluate_tool:
-                        if self.agent_tools and tool.name not in self.agent_tools:
+                        if not self.agent_tools or (
+                            self.agent_tools and tool.name not in self.agent_tools
+                        ):
                             self.lifecycle_tools.append(tool.name)
-                elif self.agent_tools and task.evaluate_tool.name not in self.agent_tools:
+                elif not self.agent_tools or (
+                    self.agent_tools and task.evaluate_tool.name not in self.agent_tools
+                ):
                     self.lifecycle_tools.append(task.evaluate_tool.name)
             if task.system_prompt:
                 self.system_prompt += "\n\n" + task.system_prompt

{hud_python-0.4.29 → hud_python-0.4.30}/hud/agents/openai_chat_generic.py

@@ -230,7 +230,8 @@ class GenericOpenAIChatAgent(MCPAgent):
         if msg.tool_calls:
             for tc in msg.tool_calls:
                 if tc.function.name is not None:  # type: ignore
-                    tool_calls.extend(self._oai_to_mcp(tc))
+                    # _oai_to_mcp returns a single MCPToolCall; append it
+                    tool_calls.append(self._oai_to_mcp(tc)) # noqa: PERF401
 
         # Only stop on length (token limit), never on "stop"
         done = choice.finish_reason == "length"

{hud_python-0.4.29 → hud_python-0.4.30}/hud/cli/rl/__init__.py

@@ -131,7 +131,11 @@ def rl_command(
     if not tasks_file:
         tasks_file = find_tasks_file(tasks_file)
         if not tasks_file:
-            console.print("[red]❌ No tasks file found in current directory[/red]")
+            hud_console.warning("No tasks file found in current directory")
+            hud_console.hint(
+                "Download a HF dataset using `hud get <dataset_name>` (e.g., `hud get hud-evals/2048-basic`)"  # noqa: E501
+            )
+            hud_console.hint("or create a tasks file manually.")
             raise typer.Exit(1)
 
     # Handle remote execution

{hud_python-0.4.29 → hud_python-0.4.30}/hud/cli/rl/display.py

@@ -42,7 +42,7 @@ def display_preset_table(presets: list[dict[str, Any]], gpu_memory_gb: float) ->
     # Add time columns for A100
     if gpu_memory_gb >= 40:
         preset_table.add_column("Tasks/hour", style="green")
-        preset_table.add_column("Steps/hour", style="green")
+        preset_table.add_column("Updates/hour", style="green")
 
     for i, preset in enumerate(presets):
         row = [

{hud_python-0.4.29 → hud_python-0.4.30}/hud/cli/rl/remote_runner.py

@@ -298,7 +298,6 @@ def run_remote_training(
                 {"name": "✏️  Review configuration", "value": "edit"},
                 {"name": "❌ Cancel", "value": "cancel"},
             ],
-            default=0,
         )
 
         if edit_choice == "cancel":

{hud_python-0.4.29 → hud_python-0.4.30}/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_python-0.4.29 → hud_python-0.4.30}/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_python-0.4.29 → hud_python-0.4.30}/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_python-0.4.29 → hud_python-0.4.30}/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_python-0.4.29 → hud_python-0.4.30}/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_python-0.4.29 → hud_python-0.4.30}/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__ == "0.4.30"

{hud_python-0.4.29 → hud_python-0.4.30}/hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.29"
+__version__ = "0.4.30"

{hud_python-0.4.29 → hud_python-0.4.30}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hud-python"
-version = "0.4.29"
+version = "0.4.30"
 description = "SDK for the HUD platform."
 readme = "README.md"
 requires-python = ">=3.11, <3.13"