Skip to content

armgit / dspy

Public Imported from GitHub
0

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

Python

fix: restrict developer role to responses (#8727)

TomuHiratac1c27b4e592d
diff --git a/dspy/clients/lm.py b/dspy/clients/lm.py
index b04ac1ac9..3c133a7c0 100644
--- a/dspy/clients/lm.py
+++ b/dspy/clients/lm.py
@@ -40,6 +40,7 @@ class LM(BaseLM):
finetuning_model: str | None = None,
launch_kwargs: dict[str, Any] | None = None,
train_kwargs: dict[str, Any] | None = None,
+ use_developer_role: bool = False,
**kwargs,
):
"""
@@ -77,6 +78,7 @@ class LM(BaseLM):
self.finetuning_model = finetuning_model
self.launch_kwargs = launch_kwargs or {}
self.train_kwargs = train_kwargs or {}
+ self.use_developer_role = use_developer_role
self._warned_zero_temp_rollout = False
# Handle model-specific configuration for different model families
@@ -131,6 +133,11 @@ class LM(BaseLM):
cache = kwargs.pop("cache", self.cache)
messages = messages or [{"role": "user", "content": prompt}]
+ if self.use_developer_role and self.model_type == "responses":
+ messages = [
+ {**m, "role": "developer"} if m.get("role") == "system" else m
+ for m in messages
+ ]
kwargs = {**self.kwargs, **kwargs}
self._warn_zero_temp_rollout(kwargs.get("temperature"), kwargs.get("rollout_id"))
if kwargs.get("rollout_id") is None:
@@ -162,6 +169,11 @@ class LM(BaseLM):
cache = kwargs.pop("cache", self.cache)
messages = messages or [{"role": "user", "content": prompt}]
+ if self.use_developer_role and self.model_type == "responses":
+ messages = [
+ {**m, "role": "developer"} if m.get("role") == "system" else m
+ for m in messages
+ ]
kwargs = {**self.kwargs, **kwargs}
self._warn_zero_temp_rollout(kwargs.get("temperature"), kwargs.get("rollout_id"))
if kwargs.get("rollout_id") is None:
diff --git a/tests/clients/test_lm.py b/tests/clients/test_lm.py
index 3bafc72ea..54068e512 100644
--- a/tests/clients/test_lm.py
+++ b/tests/clients/test_lm.py
@@ -506,6 +506,23 @@ def test_responses_api(litellm_test_server):
assert dspy_responses.call_args.kwargs["model"] == "openai/dspy-test-model"
+def test_lm_replaces_system_with_developer_role():
+ with mock.patch(
+ "dspy.clients.lm.litellm_responses_completion", return_value={"choices": []}
+ ) as mock_completion:
+ lm = dspy.LM(
+ "openai/gpt-4o-mini",
+ cache=False,
+ model_type="responses",
+ use_developer_role=True,
+ )
+ lm.forward(messages=[{"role": "system", "content": "hi"}])
+ assert (
+ mock_completion.call_args.kwargs["request"]["messages"][0]["role"]
+ == "developer"
+ )
+
+
def test_responses_api_tool_calls(litellm_test_server):
api_base, _ = litellm_test_server
expected_tool_call = {