hud-python 0.4.15__py3-none-any.whl → 0.4.17__py3-none-any.whl

hud/cli/rl/utils.py

@@ -0,0 +1,165 @@
+"""Shared utilities for RL commands."""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Any
+
+import yaml
+
+from hud.utils.design import HUDDesign
+
+design = HUDDesign()
+
+logger = logging.getLogger(__name__)
+
+
+def read_lock_file() -> dict[str, Any]:
+    """Read and parse hud.lock.yaml file."""
+    lock_file = Path("hud.lock.yaml")
+    if not lock_file.exists():
+        return {}
+
+    try:
+        with open(lock_file) as f:
+            return yaml.safe_load(f) or {}
+    except Exception as e:
+        design.warning(f"Could not read hud.lock.yaml: {e}")
+        return {}
+
+
+def write_lock_file(data: dict[str, Any]) -> bool:
+    """Write data to hud.lock.yaml file."""
+    lock_file = Path("hud.lock.yaml")
+
+    try:
+        with open(lock_file, "w") as f:
+            yaml.dump(data, f, default_flow_style=False, sort_keys=False)
+        return True
+    except Exception as e:
+        design.warning(f"Could not write hud.lock.yaml: {e}")
+        return False
+
+
+def get_mcp_config_from_lock() -> dict[str, Any] | None:
+    """Get MCP configuration from lock file."""
+    lock_data = read_lock_file()
+
+    # Check if there's an image reference
+    image = lock_data.get("image")
+    if image:
+        return {
+            "hud": {
+                "url": "https://mcp.hud.so/v3/mcp",
+                "headers": {"Authorization": "Bearer $HUD_API_KEY", "Mcp-Image": image},
+            }
+        }
+
+    return None
+
+
+def get_primary_dataset() -> str | None:
+    """Get primary dataset name from lock file."""
+    lock_data = read_lock_file()
+    return lock_data.get("primary_dataset", {}).get("name")
+
+
+def get_image_from_lock() -> str | None:
+    """Get image name from lock file."""
+    lock_data = read_lock_file()
+    return lock_data.get("image")
+
+
+def detect_image_name() -> str | None:
+    """Try to detect image name from various sources."""
+    # First check lock file
+    image = get_image_from_lock()
+    if image:
+        return image
+
+    # Check pyproject.toml
+    pyproject = Path("pyproject.toml")
+    if pyproject.exists():
+        try:
+            import tomllib
+
+            with open(pyproject, "rb") as f:
+                data = tomllib.load(f)
+
+            # Check for hud.image_name
+            image = data.get("tool", {}).get("hud", {}).get("image_name")
+            if image:
+                return image
+
+            # Use project name
+            name = data.get("project", {}).get("name")
+            if name:
+                return f"{name}:latest"
+        except Exception:
+            logger.warning("Failed to load pyproject.toml")
+
+    # Use directory name as last resort
+    return f"{Path.cwd().name}:latest"
+
+
+def validate_dataset_name(name: str) -> bool:
+    """Validate HuggingFace dataset name format."""
+    if not name:
+        return False
+
+    if "/" not in name:
+        design.error(f"Invalid dataset name: {name}")
+        design.info("Dataset name should be in format: org/dataset")
+        return False
+
+    parts = name.split("/")
+    if len(parts) != 2:
+        design.error(f"Invalid dataset name: {name}")
+        return False
+
+    org, dataset = parts
+    if not org or not dataset:
+        design.error(f"Invalid dataset name: {name}")
+        return False
+
+    # Check for valid characters (alphanumeric, dash, underscore)
+    import re
+
+    if not re.match(r"^[a-zA-Z0-9_-]+$", org) or not re.match(r"^[a-zA-Z0-9_-]+$", dataset):
+        design.error(f"Invalid characters in dataset name: {name}")
+        design.info("Use only letters, numbers, dashes, and underscores")
+        return False
+
+    return True
+
+
+def create_tasks_template() -> list[dict[str, Any]]:
+    """Create a template for tasks.json file."""
+    return [
+        {
+            "id": "example-task-001",
+            "prompt": "Complete the first TODO item in the list",
+            "mcp_config": {
+                "# TODO": "Add your MCP configuration here",
+                "# Example for remote": {
+                    "hud": {
+                        "url": "https://mcp.hud.so/v3/mcp",
+                        "headers": {
+                            "Authorization": "Bearer $HUD_API_KEY",
+                            "Mcp-Image": "your-org/your-env:latest",
+                        },
+                    }
+                },
+                "# Example for local": {
+                    "local": {"command": "docker", "args": ["run", "--rm", "-i", "your-env:latest"]}
+                },
+            },
+            "setup_tool": {"name": "setup", "arguments": {"name": "todo_seed", "num_items": 5}},
+            "evaluate_tool": {
+                "name": "evaluate",
+                "arguments": {"name": "todo_completed", "expected_count": 1},
+            },
+            "metadata": {"difficulty": "easy", "category": "task_completion"},
+        }
+    ]

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__ == "0.4.15"
+    assert hud.__version__ == "0.4.17"

hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.15"
+__version__ = "0.4.17"

{hud_python-0.4.15.dist-info → hud_python-0.4.17.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.15
+Version: 0.4.17
 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-0.4.15.dist-info → hud_python-0.4.17.dist-info}/RECORD

@@ -2,12 +2,12 @@ hud/__init__.py,sha256=BjAhZtsHbGN371Q8t3o4v4jltedkmDE85xW0yOILU9g,397
 hud/__main__.py,sha256=YR8Dq8OhINOsVfQ55PmRXXg4fEK84Rt_-rMtJ5rvhWo,145
 hud/settings.py,sha256=q9aZiHjvbL4oLE-N8AttTW4rmzS8zPMnsca-iMGyEGc,2362
 hud/types.py,sha256=gNnyS1G7aYHIR5sT3k3bOfSTFnPylUO6lNGLWbjbeYk,5149
-hud/version.py,sha256=gdxZtcCpnCv32fA8VZnQ530MgJHh6ZS7pyG585pb6No,105
+hud/version.py,sha256=EkdQIoYM_NJvpzVexCJEPP60FkfL3uxxcA80nz-GVIE,105
 hud/agents/__init__.py,sha256=UoIkljWdbq4bM0LD-mSaw6w826EqdEjOk7r6glNYwYQ,286
 hud/agents/base.py,sha256=rbwYP_a6XTwhY_5CaBlE7SWflnTq1EOuDiNY2XeUWdM,28275
-hud/agents/claude.py,sha256=in-dRByL7wU1gsuwFvscltQ1PGZURb2bIeU3QFhfpQU,14271
+hud/agents/claude.py,sha256=_eD_XKZhVJ6grkHQfbS6JskztueomQcmJeGJMbfNdmE,14534
 hud/agents/langchain.py,sha256=1EgCy8jfjunsWxlPC5XfvfLS6_XZVrIF1ZjtHcrvhYw,9584
-hud/agents/openai.py,sha256=9G7u17oN9yne7ObiuaBV67a3U3byC-NPJchDT4s5kZ0,13994
+hud/agents/openai.py,sha256=tvFYsZ5yaoLkfjMnHe-COxRttMsLRXBLPdSqgeipQRk,14257
 hud/agents/openai_chat_generic.py,sha256=jTJ-KY6HkglPK0iwZH5v3PVnaUjDsWc9IbRo3AbXlyE,5322
 hud/agents/misc/__init__.py,sha256=BYi4Ytp9b_vycpZFXnr5Oyw6ncKLNNGml8Jrb7bWUb4,136
 hud/agents/misc/response_agent.py,sha256=MsnIVElXM4ABrSJfEc_MMYp1Y_Rxmkq4kEGJ9vDX7hw,3098
@@ -30,6 +30,13 @@ hud/cli/list_func.py,sha256=ENxLL4X5uuqAASWZdQuI0k-tEzmlhUn5LATgz3QPQqQ,7065
 hud/cli/pull.py,sha256=JHwCwUwRO0Nzbgm9mkjsz6EpxbxgwQVhgNSY64nNZ-s,11969
 hud/cli/push.py,sha256=4KrEHj0_i3xJNCB3eRjANmHFhSW4MFfpnld3nfVYENs,17904
 hud/cli/remove.py,sha256=USAvB6pbMA3jd19xUtLEBiMsklVTEfE2Maw9nYcpSAE,6619
+hud/cli/rl/README.md,sha256=3pqRZMrnwD-lJwWGCCNZNhGdZG6zyydLBOer0e8BkLw,5983
+hud/cli/rl/__init__.py,sha256=g_Crqn5o0m9xANrTOkQZENWVlwHAV6MWiobte-FfqiY,3412
+hud/cli/rl/init.py,sha256=GXVOXLrX8CVAgpJ1pHuk6Y6oujbh46Rtz8kG18jGzk8,13789
+hud/cli/rl/pod.py,sha256=ZiXI-RG9YsnKx1EWzufcqklBdaD_d6XFtD45a0H8KpM,18837
+hud/cli/rl/ssh.py,sha256=bHAieonseJPON7P1mwB2GPWKLDlLZuvQniONmr5ZfcE,11523
+hud/cli/rl/train.py,sha256=sjY4J0TCp8647kzuIHyEeIsFVGtE0tllT0GzhkPPrWY,19895
+hud/cli/rl/utils.py,sha256=ZW3sjl5KaHZaOCjAbut_QIpQvxgzlxjPGuM6fuYkU9I,4836
 hud/cli/tests/__init__.py,sha256=ZrGVkmH7DHXGqOvjOSNGZeMYaFIRB2K8c6hwr8FPJ-8,68
 hud/cli/tests/test_analyze.py,sha256=SwxvRlnw-VaEwKN2nd1FJAxfhieujPjh7PdQh_LYJ5E,11050
 hud/cli/tests/test_analyze_metadata.py,sha256=s-N-ETJECloiNQhdnZrfvxDI4pWBI8hl8edQNSRaw14,9982
@@ -150,10 +157,10 @@ 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=2-9cqVBlHbL7SFxZhLBPPAg3ZuiZKQvwFqiXE0lUcyY,160
+hud/utils/tests/test_version.py,sha256=-u0egMJ2Q8IPrenCiVP8k2f_n7BSBRB2tYFGdbUwWrM,160
 hud/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-hud_python-0.4.15.dist-info/METADATA,sha256=B1t1zBp1c5qjI9HkygWl_K2RBix1A27iIqR7moxoLsc,20287
-hud_python-0.4.15.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-hud_python-0.4.15.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
-hud_python-0.4.15.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
-hud_python-0.4.15.dist-info/RECORD,,
+hud_python-0.4.17.dist-info/METADATA,sha256=z3k06ixt6r41EDxaHYtbwILpjU1Z5QVxiQ6Jwfu-BWs,20287
+hud_python-0.4.17.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+hud_python-0.4.17.dist-info/entry_points.txt,sha256=jJbodNFg1m0-CDofe5AHvB4zKBq7sSdP97-ohaQ3ae4,63
+hud_python-0.4.17.dist-info/licenses/LICENSE,sha256=yIzBheVUf86FC1bztAcr7RYWWNxyd3B-UJQ3uddg1HA,1078
+hud_python-0.4.17.dist-info/RECORD,,