feat(router): honor per-deployment model_info for continuation prefill capability

The deployment filter only read supports_assistant_prefill from the cost map, so
a model configured with `model_info: {supports_assistant_prefill: true}` on the
deployment (the usual way to declare a capability for a model that is not in the
cost map, or to override it) was wrongly skipped. Check the deployment's
model_info first (True or False both authoritative), then fall back to the cost
map. Adds tests for the override in both directions.

litellm/router_utils/pre_call_checks/continuation_prefill_check.py

@@ -35,6 +35,19 @@ _STR_KEYED_DICT_ADAPTER: Final = TypeAdapter(dict[str, object])
 def _deployment_supports_prefill(deployment: object) -> bool:
     try:
         deployment_map: Final = _STR_KEYED_DICT_ADAPTER.validate_python(deployment)
+    except ValidationError:
+        return False
+    # A per-deployment model_info override wins, so a model that is not in the cost
+    # map (or is registered generically) can still opt in or out explicitly with
+    # `model_info: {"supports_assistant_prefill": true|false}`.
+    try:
+        model_info: Final = _STR_KEYED_DICT_ADAPTER.validate_python(deployment_map.get("model_info"))
+        declared: Final = model_info.get("supports_assistant_prefill")
+        if isinstance(declared, bool):
+            return declared
+    except ValidationError:
+        pass
+    try:
         litellm_params: Final = _STR_KEYED_DICT_ADAPTER.validate_python(deployment_map.get("litellm_params"))
     except ValidationError:
         return False

tests/test_litellm/router_utils/pre_call_checks/test_continuation_prefill_check.py

@@ -20,6 +20,25 @@ def test_deployment_supports_prefill_reads_capability():
     assert _deployment_supports_prefill(_deployment(NON_PREFILL_MODEL, "b")) is False
 
 
+def test_deployment_model_info_override_wins_over_cost_map():
+    # model_info True opts in a model that is not in the cost map
+    assert (
+        _deployment_supports_prefill(
+            {"litellm_params": {"model": "vendor/custom-model"}, "model_info": {"supports_assistant_prefill": True}}
+        )
+        is True
+    )
+    # model_info False opts out a model the cost map would otherwise allow
+    assert (
+        _deployment_supports_prefill(
+            {"litellm_params": {"model": PREFILL_MODEL}, "model_info": {"supports_assistant_prefill": False}}
+        )
+        is False
+    )
+    # model_info without the key falls through to the cost map
+    assert _deployment_supports_prefill(_deployment(PREFILL_MODEL, "z")) is True
+
+
 def test_deployment_supports_prefill_rejects_malformed_deployments():
     assert _deployment_supports_prefill({}) is False
     assert _deployment_supports_prefill({"litellm_params": {}}) is False