fix(smolagents): add test for issue #2417

tests/test_issue_2417.py

@@ -0,0 +1,52 @@
+import sys
+import types
+
+from smolagents.models import VLLMModel
+
+
+def test_issue_2417(monkeypatch):
+    class FakeTokenizer:
+        def apply_chat_template(self, messages, **kwargs):
+            return "formatted prompt"
+
+    class FakeLLM:
+        def __init__(self, **kwargs):
+            self.initialization_kwargs = kwargs
+
+        def generate(self, prompt, sampling_params):
+            self.prompt = prompt
+            self.sampling_params = sampling_params
+            return [
+                types.SimpleNamespace(
+                    prompt_token_ids=[1, 2],
+                    outputs=[types.SimpleNamespace(text="4", token_ids=[3])],
+                )
+            ]
+
+    class FakeSamplingParams:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+    vllm = types.ModuleType("vllm")
+    vllm.__path__ = []
+    vllm.LLM = FakeLLM
+    vllm.SamplingParams = FakeSamplingParams
+
+    tokenizers = types.ModuleType("vllm.tokenizers")
+    tokenizers.get_tokenizer = lambda model_id: FakeTokenizer()
+
+    sampling_params = types.ModuleType("vllm.sampling_params")
+    sampling_params.StructuredOutputsParams = lambda **kwargs: kwargs
+
+    monkeypatch.setattr("smolagents.models._is_package_available", lambda package: package == "vllm")
+    monkeypatch.setitem(sys.modules, "vllm", vllm)
+    monkeypatch.setitem(sys.modules, "vllm.tokenizers", tokenizers)
+    monkeypatch.setitem(sys.modules, "vllm.sampling_params", sampling_params)
+    monkeypatch.delitem(sys.modules, "vllm.transformers_utils", raising=False)
+    monkeypatch.delitem(sys.modules, "vllm.transformers_utils.tokenizer", raising=False)
+
+    model = VLLMModel("test-model", max_tokens=4096)
+    output = model.generate([{"role": "user", "content": "What is 2+2?"}])
+
+    assert output.content == "4"
+    assert model.model.sampling_params.kwargs["max_tokens"] == 4096