Update version to 0.3.1 and update DSPyMator documentation and LM param

- Bumped version from 0.3.0 to 0.3.1 in `pyproject.toml`.
- Added detailed configuration instructions for language models in `dspymator.md`, including usage examples for both string identifiers and pre-configured `dspy.LM` objects.
- Updated parameter references to clarify the handling of `temperature` and `max_tokens` when using `dspy.LM` objects.

Author: jrosenfeld13

docs/user-guide/dspymator.md

@@ -247,6 +247,65 @@ multi_input_classifier.fit(movie_data[["title", "review_text", "rating"]], None)
 predictions = multi_input_classifier.predict(movie_data[["title", "review_text", "rating"]])
 ```
 
+### Configuring Language Models
+
+DSPyMator accepts either a model string or a pre-configured `dspy.LM` object for the `lm` parameter.
+
+**Simple usage with model string:**
+
+```python
+# Uses default OpenAI API (requires OPENAI_API_KEY env var)
+classifier = DSPyMator(
+    program=dspy.Predict("text -> label"),
+    target_names="label",
+    lm="openai/gpt-4o-mini",
+    temperature=0.0,
+    max_tokens=1000,
+)
+```
+
+**Using custom providers:**
+
+For custom API configuration, pass a pre-configured `dspy.LM` object. DSPy uses [LiteLLM](https://docs.litellm.ai/) under the hood, so any LiteLLM-supported provider works. Extra kwargs are passed through to LiteLLM:
+
+```python
+import dspy
+
+# Pass a pre-configured LM object
+classifier = DSPyMator(
+    program=dspy.Predict("text -> label"),
+    target_names="label",
+    lm=dspy.LM(
+        "openrouter/anthropic/claude-3-haiku",
+        temperature=0.1,
+        max_tokens=1000,
+        # Additional kwargs are passed to LiteLLM
+    ),
+)
+```
+
+!!! note "Temperature and max_tokens"
+    When passing a `dspy.LM` object, configure `temperature` and `max_tokens` on the LM directly. The DSPyMator parameters are ignored when using a pre-configured LM.
+
+**Environment variables:**
+
+Most providers are configured via environment variables. Set them before calling `fit()`:
+
+```python
+import os
+
+# OpenAI
+os.environ["OPENAI_API_KEY"] = "sk-..."
+
+# OpenRouter
+os.environ["OPENROUTER_API_KEY"] = "sk-or-..."
+
+# Anthropic
+os.environ["ANTHROPIC_API_KEY"] = "sk-ant-..."
+```
+
+See the [DSPy LM documentation](https://dspy.ai/api/models/LM/) and [LiteLLM provider docs](https://docs.litellm.ai/docs/providers) for supported providers and configuration.
+
 ### Async Execution for Speed
 
 By default, DSPyMator uses async execution for faster batch predictions:
@@ -291,4 +350,27 @@ llm_pipeline = make_pipeline(
         model="openai/text-embedding-3-small",      # or your preferred embedding model
         feature_names=["reasoning"],                # specify which columns to embed
     )
-```
+```
+
+## Parameters Reference
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `program` | `dspy.Module` | *required* | DSPy module (e.g., `dspy.Predict`, `dspy.ChainOfThought`) with a signature defining input/output fields |
+| `target_names` | `str \| list[str]` | *required* | Output field name(s) to use as predictions |
+| `feature_names` | `list[str] \| None` | `None` | Column names mapping input data to signature fields. If `None`, inferred from dataframe columns |
+| `lm` | `str \| dspy.LM` | `"openai/gpt-5-nano"` | Language model - either a string identifier or a pre-configured `dspy.LM` object |
+| `temperature` | `float` | `1.0` | Sampling temperature (ignored if `lm` is a `dspy.LM` object) |
+| `max_tokens` | `int` | `16000` | Maximum tokens in responses (ignored if `lm` is a `dspy.LM` object) |
+| `use_async` | `bool` | `True` | Use async execution for batch predictions |
+| `max_concurrent` | `int` | `50` | Maximum concurrent requests in async mode |
+| `verbose` | `bool` | `True` | Show progress bars during prediction |
+
+### fit() Parameters
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `X` | DataFrame/array | *required* | Training features |
+| `y` | Series/array | *required* | Target values (can be `None` for unsupervised) |
+| `optimizer` | `dspy.Optimizer \| None` | `None` | DSPy optimizer instance (e.g., `dspy.GEPA`, `dspy.BootstrapFewShot`) |
+| `validation_data` | `tuple \| float \| None` | `None` | Validation data as `(X_val, y_val)`, a float for train split fraction, or `None` |

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "centimators"
-version = "0.3.0"
+version = "0.3.1"
 description = "essential data transformers and model estimators for ML and data science competitions"
 readme = "README.md"
 authors = [

src/centimators/model_estimators/dspymator.py

@@ -117,7 +117,10 @@ class DSPyMator(TransformerMixin, BaseEstimator):
         feature_names: Column names mapping input data to signature input fields.
             If None, inferred from dataframe columns or uses signature field names
             for numpy arrays. Must match the number of input fields in the signature.
-        lm: Language model identifier (e.g., "openai/gpt-4", "anthropic/claude-3").
+        lm: Language model - either a string identifier (e.g., "openai/gpt-4") or a
+            pre-configured `dspy.LM` object. Pass a `dspy.LM` directly when you need
+            custom configuration like `api_key` or `api_base` for providers like OpenRouter.
+            When passing an LM object, `temperature` and `max_tokens` are ignored.
             Defaults to "openai/gpt-5-nano".
         temperature: Sampling temperature for the language model. Defaults to 1.0.
         max_tokens: Maximum tokens in model responses. Defaults to 16000.
@@ -167,7 +170,7 @@ class DSPyMator(TransformerMixin, BaseEstimator):
     program: dspy.Module
     target_names: str | list[str]
     feature_names: list[str] | None = None
-    lm: str = "openai/gpt-5-nano"
+    lm: str | dspy.LM = "openai/gpt-5-nano"
     temperature: float = 1.0
     max_tokens: int = 16000
     use_async: bool = True
@@ -243,9 +246,12 @@ def fit(
             estimator.fit(X_train, y_train, optimizer=gepa_optimizer, validation_data=0.2)
             ```
         """
-        self.lm_ = dspy.LM(
-            self.lm, temperature=self.temperature, max_tokens=self.max_tokens
-        )
+        if isinstance(self.lm, dspy.LM):
+            self.lm_ = self.lm
+        else:
+            self.lm_ = dspy.LM(
+                self.lm, temperature=self.temperature, max_tokens=self.max_tokens
+            )
 
         self.input_fields_ = list(self.signature_.input_fields.keys())