fix(scores): coerce numpy and Decimal score values to float in create_score

langfuse/_client/client.py

@@ -1947,9 +1947,15 @@ def create_score(
             timestamp = timestamp.replace(tzinfo=timezone.utc)
 
         try:
-            # The model turned bools and ints into floats (True -> 1.0); keep that.
-            if isinstance(value, (bool, int)):
-                value = float(value)
+            # The model coerced every non-str value to float (True -> 1.0,
+            # numpy.bool_ -> 1.0, Decimal("0.75") -> 0.75); keep that.
+            # Typed as float | str, but evaluators often return numpy or Decimal values.
+            raw_value: Any = value
+            if not isinstance(raw_value, (str, float)):
+                try:
+                    value = float(raw_value)
+                except (TypeError, ValueError):
+                    pass
 
             score = {
                 "id": score_id,

tests/unit/test_score_consumer.py

@@ -7,6 +7,8 @@
 import threading
 import time
 from datetime import datetime, timezone
+from decimal import Decimal
+from fractions import Fraction
 from queue import Queue
 from typing import Any, Callable, Dict, List, Sequence
 
@@ -758,6 +760,8 @@ def test_create_score_payload_matches_request_model_serialization(client_factory
         ),
         dict(name="int-value", value=1, trace_id="a" * 32),
         dict(name="boolean", value=True, trace_id="a" * 32, data_type="BOOLEAN"),
+        dict(name="decimal-value", value=Decimal("0.75"), trace_id="a" * 32),
+        dict(name="fraction-value", value=Fraction(3, 4), trace_id="a" * 32),
         dict(
             name="categorical",
             value="good",
@@ -809,6 +813,21 @@ def test_create_score_payload_matches_request_model_serialization(client_factory
     assert sent == expected
 
 
+def test_create_score_sends_numpy_values_as_numbers(client_factory):
+    np = pytest.importorskip("numpy")
+    recorder = Recorder(_responder(ACCEPTED))
+    client = client_factory(recorder)
+
+    # e.g. an evaluator returning np.mean(hits) > 0.5 or np.float32(score)
+    for value in (np.bool_(True), np.float32(0.5), np.int64(3)):
+        client.create_score(name="numpy", value=value, trace_id="a" * 32)
+    client.flush()
+
+    sent = [score["value"] for body in recorder.bodies for score in body]
+    assert sent == [1.0, 0.5, 3.0]
+    assert all(type(value) is float for value in sent)
+
+
 @pytest.fixture
 def non_utc_local_timezone(monkeypatch):
     # Makes treating naive values as local time observable.