validmind 2.13.9__tar.gz → 2.13.11__tar.gz

{validmind-2.13.9 → validmind-2.13.11}/.gitignore

@@ -189,12 +189,8 @@ lending_club_loan_data_*.csv
 
 # Pickle files
 *.pkl
-# Sample application scorecard model for validation notebook — do not remove!
-!notebooks/use_cases/validation/xgb_model_champion.pkl
 # Sample logistic regression model for validation series — do not remove!
 !notebooks/tutorials/validation/lr_model_champion.pkl
-# Sample XGBoost model for validation quickstart — do not remove!
-!notebooks/quickstart/xgboost_model_champion.pkl
 
 notebooks/llm/datasets/*.jsonl
 

{validmind-2.13.9 → validmind-2.13.11}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: validmind
-Version: 2.13.9
+Version: 2.13.11
 Summary: ValidMind Library
 Author-email: Andres Rodriguez <andres@validmind.ai>, Juan Martinez <juan@validmind.ai>, Anil Sorathiya <anil@validmind.ai>, Luis Pallares <luis@validmind.ai>, John Walz <john@validmind.ai>
 License:                          DUAL LICENSE NOTICE
@@ -735,6 +735,7 @@ Requires-Dist: pyarrow<23,>=22; (python_version >= '3.14') and extra == 'all'
 Requires-Dist: pycocoevalcap; extra == 'all'
 Requires-Dist: ragas<=0.2.7,>=0.2.3; extra == 'all'
 Requires-Dist: rouge>=1; extra == 'all'
+Requires-Dist: scikit-learn<1.8; extra == 'all'
 Requires-Dist: scipy; extra == 'all'
 Requires-Dist: scorecardpy==0.1.9.6; extra == 'all'
 Requires-Dist: sentencepiece<0.3.0,>=0.2.0; extra == 'all'
@@ -744,7 +745,7 @@ Requires-Dist: statsmodels<0.15.0,>=0.14.2; extra == 'all'
 Requires-Dist: textblob<0.19.0,>=0.18.0.post0; extra == 'all'
 Requires-Dist: torch>=2.0.0; extra == 'all'
 Requires-Dist: transformers<5.0.0,>=4.32.0; extra == 'all'
-Requires-Dist: xgboost<3,>=1.5.2; extra == 'all'
+Requires-Dist: xgboost<3.1,>=1.5.2; extra == 'all'
 Provides-Extra: credit-risk
 Requires-Dist: scorecardpy==0.1.9.6; extra == 'credit-risk'
 Requires-Dist: setuptools<82; extra == 'credit-risk'
@@ -802,7 +803,8 @@ Requires-Dist: arch>=7.0.0; extra == 'stats'
 Requires-Dist: scipy; extra == 'stats'
 Requires-Dist: statsmodels<0.15.0,>=0.14.2; extra == 'stats'
 Provides-Extra: xgboost
-Requires-Dist: xgboost<3,>=1.5.2; extra == 'xgboost'
+Requires-Dist: scikit-learn<1.8; extra == 'xgboost'
+Requires-Dist: xgboost<3.1,>=1.5.2; extra == 'xgboost'
 Description-Content-Type: text/markdown
 
 # ValidMind Library

{validmind-2.13.9 → validmind-2.13.11}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "validmind"
-version = "2.13.9"
+version = "2.13.11"
 description = "ValidMind Library"
 readme = "README.pypi.md"
 requires-python = ">=3.9,<3.15"
@@ -40,7 +40,8 @@ dependencies = [
 [project.optional-dependencies]
 all = [
   "torch (>=2.0.0)",
-  "xgboost (>=1.5.2,<3)",
+  "xgboost (>=1.5.2,<3.1)",
+  "scikit-learn (<1.8)",
   "transformers (>=4.32.0,<5.0.0)",
   "pycocoevalcap",
   "ragas (>=0.2.3,<=0.2.7)",
@@ -108,7 +109,10 @@ nlp = [
 ]
 pytorch = ["torch (>=2.0.0)"]
 stats = ["scipy", "statsmodels (>=0.14.2,<0.15.0)", "arch (>=7.0.0)"]
-xgboost = ["xgboost (>=1.5.2,<3)"]
+xgboost = [
+  "xgboost (>=1.5.2,<3.1)",
+  "scikit-learn (<1.8)",
+]
 explainability = [
   "shap (>=0.46.0)",
   "numba (>=0.65.1) ; python_version >= '3.14'",

validmind-2.13.11/validmind/__version__.py

@@ -0,0 +1 @@
+__version__ = "2.13.11"

{validmind-2.13.9 → validmind-2.13.11}/validmind/ai/utils.py

@@ -8,7 +8,7 @@ import os
 from openai import AzureOpenAI, OpenAI
 
 from ..logging import get_logger
-from ..utils import md_to_html
+from ..utils import is_html, md_to_html
 
 logger = get_logger(__name__)
 
@@ -38,6 +38,7 @@ class DescriptionFuture:
 
     def __init__(self, future):
         self._future = future
+        self.markdown_source = None
 
     def get_description(self):
         if isinstance(self._future, tuple):
@@ -46,7 +47,9 @@ class DescriptionFuture:
             # This will block until the future is completed
             description = self._future.result()
 
-        return md_to_html(description[0], mathml=True), description[1]
+        source, was_generated = description
+        self.markdown_source = source if not is_html(source) else None
+        return md_to_html(source, mathml=True), was_generated
 
 
 def _get_google_api_key():

{validmind-2.13.9 → validmind-2.13.11}/validmind/api_client.py

@@ -588,6 +588,7 @@ async def alog_metadata(
     text: Optional[str] = None,
     _json: Optional[Dict[str, Any]] = None,
     section_id: Optional[str] = None,
+    text_format: Optional[str] = None,
 ) -> Dict[str, Any]:
     """Logs free-form metadata to ValidMind API.
 
@@ -597,6 +598,9 @@ async def alog_metadata(
         _json (dict, optional): Free-form key-value pairs to assign to the metadata. Defaults to None.
         section_id (str, optional): Section ID to append the text block to when the
             content ID does not already exist.
+        text_format (str, optional): Format of ``text``. Markdown is sent as-is when
+            the backend advertises support so conversion happens after the request
+            passes through the WAF. Older backends receive locally converted HTML.
 
     Raises:
         Exception: If the API call fails.
@@ -604,11 +608,17 @@ async def alog_metadata(
     Returns:
         dict: The response from the API.
     """
+    if text_format == "markdown" and not client_config.supports_log_metadata_markdown():
+        text = md_to_html(text, mathml=True)
+        text_format = None
+
     metadata_dict = {"content_id": content_id}
     if text is not None:
         metadata_dict["text"] = text
     if _json is not None:
         metadata_dict["json"] = _json
+    if text_format is not None:
+        metadata_dict["text_format"] = text_format
 
     request_params = {}
     if section_id:
@@ -808,11 +818,18 @@ def generate_qualitative_text(text_generation_data: Dict[str, Any]) -> Dict[str,
     return r.json()
 
 
-def _normalize_logged_text(text: str, field_name: str) -> str:
-    """Validate text content and convert Markdown to HTML when needed."""
+def _validate_logged_text(text: str, field_name: str) -> str:
+    """Validate text accepted by log_text."""
     if not isinstance(text, str) or not text:
         raise ValueError(f"`{field_name}` must be a non-empty string")
 
+    return text
+
+
+def _normalize_logged_text(text: str, field_name: str) -> str:
+    """Validate text content and convert Markdown to HTML for local rendering."""
+    text = _validate_logged_text(text, field_name)
+
     if not is_html(text):
         return md_to_html(text, mathml=True)
 
@@ -828,7 +845,7 @@ def _validate_manual_log_text_args(
     if context is not None:
         raise ValueError("`context` is only supported when `text` is omitted")
 
-    return _normalize_logged_text(text, "text")
+    return _validate_logged_text(text, "text")
 
 
 def _build_log_text_generation_request(
@@ -860,13 +877,13 @@ def _build_log_text_generation_request(
     return request_data
 
 
-def _generate_log_text(
+def _generate_log_text_source(
     content_id: str,
     prompt: Optional[str],
     context: Optional[Dict[str, Any]],
     section_id: Optional[str] = None,
 ) -> str:
-    """Generate text for log_text and normalize it to HTML."""
+    """Generate and validate source text without converting it to HTML."""
     request_data = _build_log_text_generation_request(
         content_id,
         prompt,
@@ -874,6 +891,22 @@ def _generate_log_text(
         section_id=section_id,
     )
     generated_text = generate_qualitative_text(request_data)["content"]
+    return _validate_logged_text(generated_text, "generated text")
+
+
+def _generate_log_text(
+    content_id: str,
+    prompt: Optional[str],
+    context: Optional[Dict[str, Any]],
+    section_id: Optional[str] = None,
+) -> str:
+    """Generate text and normalize it to HTML for local result rendering."""
+    generated_text = _generate_log_text_source(
+        content_id,
+        prompt,
+        context,
+        section_id=section_id,
+    )
     return _normalize_logged_text(generated_text, "generated text")
 
 
@@ -902,14 +935,21 @@ async def alog_text(
     if text is not None:
         text = _validate_manual_log_text_args(text, prompt, context)
     else:
-        text = _generate_log_text(
+        text = _generate_log_text_source(
             content_id,
             prompt,
             context,
             section_id=section_id,
         )
 
-    return await alog_metadata(content_id, text, _json, section_id=section_id)
+    text_format = None if is_html(text) else "markdown"
+    return await alog_metadata(
+        content_id,
+        text,
+        _json,
+        section_id=section_id,
+        text_format=text_format,
+    )
 
 
 def log_text(
@@ -924,9 +964,9 @@ def log_text(
 
     Args:
         content_id (str): Unique content identifier for the text.
-        text (str, optional): The text to log. Will be converted to HTML with
-            MathML support when Markdown is provided. If omitted, text is
-            generated using the qualitative text generation backend.
+        text (str, optional): The text to log. Markdown is sent to the backend for
+            HTML and math conversion. If omitted, text is generated using the
+            qualitative text generation backend.
         prompt (str, optional): Custom prompt used for AI-assisted text
             generation. Only supported when `text` is omitted.
         context (dict, optional): Context object for AI-assisted text

{validmind-2.13.9 → validmind-2.13.11}/validmind/client.py

@@ -38,7 +38,7 @@ from .template import preview_template as _preview_template
 from .test_suites import get_by_id as get_test_suite_by_id
 from .tests.run import _get_run_metadata
 from .utils import display as vm_display
-from .utils import get_dataset_info, get_model_info
+from .utils import get_dataset_info, get_model_info, is_html
 from .vm_models import TestSuite, TestSuiteRunner
 from .vm_models.dataset import DataFrameDataset, PolarsDataset, TorchDataset, VMDataset
 from .vm_models.html_progress import HTMLLabel, HTMLProgressBar
@@ -479,9 +479,12 @@ def run_text_generation(
         A TextGenerationResult containing the generated text.
     """
     start_time = time.perf_counter()
-    description = api_client._generate_log_text(
+    description_source = api_client._generate_log_text_source(
         content_id, prompt, context, section_id=section_id
     )
+    description = api_client._normalize_logged_text(
+        description_source, "generated text"
+    )
     metadata = _get_run_metadata(duration_seconds=time.perf_counter() - start_time)
     result = TextGenerationResult(
         result_id=content_id,
@@ -494,6 +497,9 @@ def run_text_generation(
         section_id=section_id,
         context=context,
         _was_description_generated=True,
+        _description_source=(
+            description_source if not is_html(description_source) else None
+        ),
     )
 
     if show:

{validmind-2.13.9 → validmind-2.13.11}/validmind/client_config.py

@@ -39,6 +39,10 @@ class ClientConfig:
         """Returns True if the client can generate LLM-based test descriptions."""
         return self.feature_flags.get("llm_test_descriptions", True)
 
+    def supports_log_metadata_markdown(self):
+        """Return whether the backend converts Markdown logged as metadata."""
+        return self.feature_flags.get("log_metadata_markdown", False)
+
 
 client_config = ClientConfig(
     model=None,

{validmind-2.13.9 → validmind-2.13.11}/validmind/experimental/agents.py

@@ -54,7 +54,8 @@ def run_task(
     else:
         raise ValueError(f"Unsupported task: {task}")
 
-    if not is_html(generated_text):
+    description_source = generated_text if not is_html(generated_text) else None
+    if description_source is not None:
         generated_text = md_to_html(generated_text, mathml=True)
 
     # Create a test result with the generated text
@@ -68,6 +69,7 @@ def run_task(
         section_id=input.get("section_id"),
         context=input.get("context"),
         _was_description_generated=True,
+        _description_source=description_source,
     )
     if show:
         result.show()

{validmind-2.13.9 → validmind-2.13.11}/validmind/tests/output.py

@@ -184,7 +184,10 @@ class StringOutputHandler(OutputHandler):
 
     def process(self, item: Any, result: TestResult) -> None:
         if not is_html(item):
+            result._description_source = item
             item = md_to_html(item, mathml=True)
+        else:
+            result._description_source = None
 
         result.description = item
 

{validmind-2.13.9 → validmind-2.13.11}/validmind/tests/run.py

@@ -16,7 +16,7 @@ from validmind.ai.test_descriptions import get_result_description
 from validmind.errors import MissingRequiredTestInputError
 from validmind.input_registry import input_registry
 from validmind.logging import get_logger
-from validmind.utils import test_id_to_name
+from validmind.utils import is_html, test_id_to_name
 from validmind.vm_models.input import VMInput
 from validmind.vm_models.result import TestResult
 
@@ -451,7 +451,7 @@ def run_test(  # noqa: C901
         result = post_process_fn(result)
 
     if not result.description:
-        result.description = get_result_description(
+        description = get_result_description(
             test_id=test_id,
             test_description=result.doc,
             tables=result.tables,
@@ -463,6 +463,9 @@ def run_test(  # noqa: C901
             additional_context=additional_context,
             params=result.params,
         )
+        result.description = description
+        if isinstance(description, str) and result.doc and not is_html(result.doc):
+            result._description_source = result.doc
 
     if show:
         result.show()

{validmind-2.13.9 → validmind-2.13.11}/validmind/vm_models/model.py

@@ -187,10 +187,12 @@ def is_pytorch_model(model):
     Checks if the model is a PyTorch model. Need to extend this
     method to check for all ways a PyTorch model can be created
     """
-    # if we can't import torch, then it's not a PyTorch model
+    # if we can't import torch, then it's not a PyTorch model. Broken torch
+    # installs can fail with errors other than ImportError (e.g. OSError
+    # WinError 1114 when a native DLL fails to load on Windows)
     try:
         import torch.nn as nn
-    except ImportError:
+    except Exception:
         return False
 
     # return False

{validmind-2.13.9 → validmind-2.13.11}/validmind/vm_models/result/result.py

@@ -20,7 +20,13 @@ from ... import api_client
 from ...ai.utils import DescriptionFuture
 from ...errors import InvalidParameterError
 from ...logging import get_logger, log_api_operation
-from ...utils import HumanReadableEncoder, display, run_async, test_id_to_name
+from ...utils import (
+    HumanReadableEncoder,
+    display,
+    md_to_html,
+    run_async,
+    test_id_to_name,
+)
 from ..figure import Figure, create_figure
 from ..html_renderer import StatefulHTMLRenderer
 from ..input import VMInput
@@ -36,6 +42,17 @@ from .utils import (
 logger = get_logger(__name__)
 
 
+def _description_transport(rendered_description, markdown_source):
+    """Use Markdown only while it still corresponds to the rendered description."""
+    if (
+        markdown_source is not None
+        and md_to_html(markdown_source, mathml=True) == rendered_description
+    ):
+        return markdown_source, "markdown"
+
+    return rendered_description, None
+
+
 class RawData:
     """Holds raw data for a test result."""
 
@@ -197,6 +214,7 @@ class TestResult(Result):
     inputs: Optional[Dict[str, Union[List[VMInput], VMInput]]] = None
     metadata: Optional[Dict[str, Any]] = None
     _was_description_generated: bool = False
+    _description_source: Optional[str] = None
     _unsafe: bool = False
     _client_config_cache: Optional[Any] = None
     _is_scorer_result: bool = False
@@ -237,6 +255,7 @@ class TestResult(Result):
                     self.description,
                     self._was_description_generated,
                 ) = description.get_description()
+                self._description_source = description.markdown_source
 
         return super().__getattribute__(name)
 
@@ -597,17 +616,23 @@ class TestResult(Result):
                 if self._was_description_generated
                 else DEFAULT_REVISION_NAME
             )
-
-            tasks.append(
-                update_metadata(
-                    content_id=(
-                        f"{content_id}::{revision_name}"
-                        if content_id
-                        else f"test_description:{self.result_id}::{revision_name}"
-                    ),
-                    text=self.description,
-                )
+            content_id_with_revision = (
+                f"{content_id}::{revision_name}"
+                if content_id
+                else f"test_description:{self.result_id}::{revision_name}"
+            )
+            description_text, text_format = _description_transport(
+                self.description,
+                self._description_source,
             )
+            metadata_kwargs = {
+                "content_id": content_id_with_revision,
+                "text": description_text,
+            }
+            if text_format is not None:
+                metadata_kwargs["text_format"] = text_format
+
+            tasks.append(update_metadata(**metadata_kwargs))
 
         return await asyncio.gather(*tasks)
 
@@ -717,6 +742,7 @@ class TextGenerationResult(Result):
     section_id: Optional[str] = None
     context: Optional[Dict[str, Any]] = None
     _was_description_generated: bool = False
+    _description_source: Optional[str] = None
     _client_config_cache: Optional[Any] = None
 
     def __post_init__(self):
@@ -853,9 +879,13 @@ class TextGenerationResult(Result):
             else f"{resolved_content_id}::{revision_name}"
         )
 
+        description_text, _ = _description_transport(
+            self.description,
+            self._description_source,
+        )
         return await api_client.alog_text(
             content_id=resolved_content_id,
-            text=self.description,
+            text=description_text,
             section_id=resolved_section_id,
         )
 

{validmind-2.13.9 → validmind-2.13.11}/validmind/vm_models/result/utils.py

@@ -3,7 +3,7 @@
 # SPDX-License-Identifier: AGPL-3.0 AND ValidMind Commercial
 
 import os
-from typing import TYPE_CHECKING, Dict, List, Union
+from typing import TYPE_CHECKING, Dict, List, Optional, Union
 
 from jinja2 import Template
 
@@ -35,7 +35,12 @@ def get_result_template():
     return _result_template
 
 
-async def update_metadata(content_id: str, text: str, _json: Union[Dict, List] = None):
+async def update_metadata(
+    content_id: str,
+    text: str,
+    _json: Union[Dict, List] = None,
+    text_format: Optional[str] = None,
+):
     """Create or update a metadata object."""
     parts = content_id.split("::")
     content_id = parts[0]
@@ -46,7 +51,12 @@ async def update_metadata(content_id: str, text: str, _json: Union[Dict, List] =
 
     logger.debug(f"Updating metadata for `{content_id}`")
 
-    await api_client.alog_metadata(content_id, text, _json)
+    await api_client.alog_metadata(
+        content_id,
+        text,
+        _json,
+        text_format=text_format,
+    )
 
 
 def tables_to_html(tables: List["ResultTable"]) -> str:

validmind-2.13.9/validmind/__version__.py

@@ -1 +0,0 @@
-__version__ = "2.13.9"