Skip to content

armgit / dspy

Public Imported from GitHub
0

DSPy: The framework for programming—not prompting—language models

Python

Adjust reasoning model token requirement (#8750)

Omar Khattab9725d4de612c
diff --git a/dspy/adapters/two_step_adapter.py b/dspy/adapters/two_step_adapter.py
index 9e3f7bbb1..b78375f14 100644
--- a/dspy/adapters/two_step_adapter.py
+++ b/dspy/adapters/two_step_adapter.py
@@ -30,7 +30,7 @@ class TwoStepAdapter(Adapter):
Example:
```
import dspy
- lm = dspy.LM(model="openai/o3-mini", max_tokens=10000, temperature = 1.0)
+ lm = dspy.LM(model="openai/o3-mini", max_tokens=16000, temperature = 1.0)
adapter = dspy.TwoStepAdapter(dspy.LM("openai/gpt-4o-mini"))
dspy.configure(lm=lm, adapter=adapter)
program = dspy.ChainOfThought("question->answer")
diff --git a/dspy/clients/lm.py b/dspy/clients/lm.py
index 02a532fda..b04ac1ac9 100644
--- a/dspy/clients/lm.py
+++ b/dspy/clients/lm.py
@@ -86,10 +86,10 @@ class LM(BaseLM):
model_pattern = re.match(r"^(?:o[1345]|gpt-5)(?:-(?:mini|nano))?", model_family)
if model_pattern:
- if max_tokens < 20000 or temperature != 1.0:
+ if max_tokens < 16000 or temperature != 1.0:
raise ValueError(
- "OpenAI's reasoning models require passing temperature=1.0 and max_tokens >= 20000 to "
- "`dspy.LM(...)`, e.g., dspy.LM('openai/gpt-5', temperature=1.0, max_tokens=20000)"
+ "OpenAI's reasoning models require passing temperature=1.0 and max_tokens >= 16000 to "
+ "`dspy.LM(...)`, e.g., dspy.LM('openai/gpt-5', temperature=1.0, max_tokens=16000)"
)
self.kwargs = dict(temperature=temperature, max_completion_tokens=max_tokens, **kwargs)
if self.kwargs.get("rollout_id") is None:
diff --git a/tests/clients/test_lm.py b/tests/clients/test_lm.py
index fbe626f60..3bafc72ea 100644
--- a/tests/clients/test_lm.py
+++ b/tests/clients/test_lm.py
@@ -269,12 +269,12 @@ def test_reasoning_model_token_parameter():
lm = dspy.LM(
model=model_name,
temperature=1.0 if is_reasoning_model else 0.7,
- max_tokens=20_000 if is_reasoning_model else 1000,
+ max_tokens=16_000 if is_reasoning_model else 1000,
)
if is_reasoning_model:
assert "max_completion_tokens" in lm.kwargs
assert "max_tokens" not in lm.kwargs
- assert lm.kwargs["max_completion_tokens"] == 20_000
+ assert lm.kwargs["max_completion_tokens"] == 16_000
else:
assert "max_completion_tokens" not in lm.kwargs
assert "max_tokens" in lm.kwargs
@@ -285,21 +285,21 @@ def test_reasoning_model_requirements(model_name):
# Should raise assertion error if temperature or max_tokens requirements not met
with pytest.raises(
ValueError,
- match="reasoning models require passing temperature=1.0 and max_tokens >= 20000",
+ match="reasoning models require passing temperature=1.0 and max_tokens >= 16000",
):
dspy.LM(
model=model_name,
temperature=0.7, # Should be 1.0
- max_tokens=1000, # Should be >= 20_000
+ max_tokens=1000, # Should be >= 16_000
)
# Should pass with correct parameters
lm = dspy.LM(
model=model_name,
temperature=1.0,
- max_tokens=20_000,
+ max_tokens=16_000,
)
- assert lm.kwargs["max_completion_tokens"] == 20_000
+ assert lm.kwargs["max_completion_tokens"] == 16_000
def test_dump_state():