Skip to content

armgit / dspy

Public Imported from GitHub
0

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

Python

Fix save predict model with LM (#7861)

TomuHirata08478cc84723
diff --git a/dspy/clients/lm.py b/dspy/clients/lm.py
index 6f5468b84..47e69f95f 100644
--- a/dspy/clients/lm.py
+++ b/dspy/clients/lm.py
@@ -246,6 +246,10 @@ class LM(BaseLM):
new_instance.kwargs[key] = value
return new_instance
+
+ def dump_state(self):
+ state_keys = ["model", "model_type", "cache", "cache_in_memory", "num_retries", "finetuning_model", "launch_kwargs", "train_kwargs"]
+ return { key: getattr(self, key) for key in state_keys } | self.kwargs
def request_cache(maxsize: Optional[int] = None):
diff --git a/dspy/predict/predict.py b/dspy/predict/predict.py
index c0e13ec62..e064a966e 100644
--- a/dspy/predict/predict.py
+++ b/dspy/predict/predict.py
@@ -7,6 +7,7 @@ from dspy.primitives.prediction import Prediction
from dspy.primitives.program import Module
from dspy.signatures.signature import ensure_signature
from dspy.utils.callback import with_callbacks
+from dspy.clients.lm import LM
class Predict(Module, Parameter):
@@ -24,7 +25,7 @@ class Predict(Module, Parameter):
self.demos = []
def dump_state(self):
- state_keys = ["lm", "traces", "train"]
+ state_keys = ["traces", "train"]
state = {k: getattr(self, k) for k in state_keys}
state["demos"] = []
@@ -38,6 +39,7 @@ class Predict(Module, Parameter):
state["demos"].append(demo)
state["signature"] = self.signature.dump_state()
+ state["lm"] = self.lm.dump_state() if self.lm else None
return state
def load_state(self, state):
@@ -49,13 +51,14 @@ class Predict(Module, Parameter):
Returns:
self: Returns self to allow method chaining
"""
- excluded_keys = ["signature", "extended_signature"]
+ excluded_keys = ["signature", "extended_signature", "lm"]
for name, value in state.items():
# `excluded_keys` are fields that go through special handling.
if name not in excluded_keys:
setattr(self, name, value)
self.signature = self.signature.load_state(state["signature"])
+ self.lm = LM(**state["lm"]) if state["lm"] else None
if "extended_signature" in state: # legacy, up to and including 2.5, for CoT.
raise NotImplementedError("Loading extended_signature is no longer supported in DSPy 2.6+")
diff --git a/tests/clients/test_lm.py b/tests/clients/test_lm.py
index 46572c8c3..9d218ff90 100644
--- a/tests/clients/test_lm.py
+++ b/tests/clients/test_lm.py
@@ -226,3 +226,27 @@ def test_reasoning_model_requirements():
max_tokens=5000,
)
assert lm.kwargs["max_completion_tokens"] == 5000
+
+def test_dump_state():
+ lm = dspy.LM(
+ model="openai/gpt-4o-mini",
+ model_type="chat",
+ temperature=1,
+ max_tokens=100,
+ num_retries=10,
+ launch_kwargs={ "temperature": 1 },
+ train_kwargs={ "temperature": 5 },
+ )
+
+ assert lm.dump_state() == {
+ "model": "openai/gpt-4o-mini",
+ "model_type": "chat",
+ "temperature": 1,
+ "max_tokens": 100,
+ "num_retries": 10,
+ "cache": True,
+ "cache_in_memory": True,
+ "finetuning_model": None,
+ "launch_kwargs": { "temperature": 1 },
+ "train_kwargs": { "temperature": 5 },
+ }
diff --git a/tests/predict/test_predict.py b/tests/predict/test_predict.py
index 037a49404..c734f6219 100644
--- a/tests/predict/test_predict.py
+++ b/tests/predict/test_predict.py
@@ -35,11 +35,31 @@ def test_reset_method():
def test_lm_after_dump_and_load_state():
predict_instance = Predict("input -> output")
- predict_instance.lm = "lm_state"
+ lm = dspy.LM(
+ model="openai/gpt-4o-mini",
+ model_type="chat",
+ temperature=1,
+ max_tokens=100,
+ num_retries=10,
+ )
+ predict_instance.lm = lm
+ expected_lm_state = {
+ "model": "openai/gpt-4o-mini",
+ "model_type": "chat",
+ "temperature": 1,
+ "max_tokens": 100,
+ "num_retries": 10,
+ "cache": True,
+ "cache_in_memory": True,
+ "finetuning_model": None,
+ "launch_kwargs": {},
+ "train_kwargs": {},
+ }
+ assert lm.dump_state() == expected_lm_state
dumped_state = predict_instance.dump_state()
new_instance = Predict("input -> output")
new_instance.load_state(dumped_state)
- assert new_instance.lm == "lm_state"
+ assert new_instance.lm.dump_state() == expected_lm_state
def test_call_method():
@@ -209,6 +229,28 @@ def test_signature_fields_after_dump_and_load_state(tmp_path):
new_instance.load(file_path)
assert new_instance.signature.dump_state() == original_instance.signature.dump_state()
[email protected]("filename", ["model.json", "model.pkl"])
+def test_lm_field_after_dump_and_load_state(tmp_path, filename):
+ file_path = tmp_path / filename
+ lm = dspy.LM(
+ model="openai/gpt-4o-mini",
+ model_type="chat",
+ temperature=1,
+ max_tokens=100,
+ num_retries=10,
+ )
+ original_predict = dspy.Predict("q->a")
+ original_predict.lm = lm
+
+ original_predict.save(file_path)
+
+ assert file_path.exists()
+
+ loaded_predict = dspy.Predict("q->a")
+ loaded_predict.load(file_path)
+
+ assert original_predict.dump_state() == loaded_predict.dump_state()
+
def test_forward_method():
program = Predict("question -> answer")