Merge pull request #1559 from stanfordnlp/typed_predictors_fixes_v2_5

Improve o1 support in dspy.LM (QoL errors)

dspy/clients/lm.py

@@ -30,8 +30,10 @@ def __init__(self, model, model_type='chat', temperature=0.0, max_tokens=1000, c
         self.kwargs = dict(temperature=temperature, max_tokens=max_tokens, **kwargs)
         self.history = []
 
-        # Exceptions for O-1 models, which requires a temperature of 1.0.
-        if "o1-" in model: self.kwargs['temperature'] = 1.0
+        if "o1-" in model:
+            assert max_tokens >= 5000 and temperature == 1.0, \
+                "OpenAI's o1-* models require passing temperature=1.0 and max_tokens >= 5000 to `dspy.LM(...)`"
+
 
     def __call__(self, prompt=None, messages=None, **kwargs):
         # Build the request.