Always pass add_generation_prompt=True to apply_chat_template

src/smolagents/models.py

@@ -536,18 +536,12 @@ def generate(
         tools = completion_kwargs.pop("tools", None)
         completion_kwargs.pop("tool_choice", None)
 
-        if tools_to_call_from is not None:
-            prompt = self.tokenizer.apply_chat_template(
-                messages,
-                tools=tools,
-                add_generation_prompt=True,
-                tokenize=False,
-            )
-        else:
-            prompt = self.tokenizer.apply_chat_template(
-                messages,
-                tokenize=False,
-            )
+        prompt = self.tokenizer.apply_chat_template(
+            messages,
+            tools=tools,
+            add_generation_prompt=True,
+            tokenize=False,
+        )
 
         sampling_params = SamplingParams(
             n=kwargs.get("n", 1),
@@ -834,6 +828,7 @@ def _prepare_completion_args(
 
         messages = completion_kwargs.pop("messages")
         stop_sequences = completion_kwargs.pop("stop", None)
+        tools = completion_kwargs.pop("tools", None)
 
         max_new_tokens = (
             kwargs.get("max_new_tokens")
@@ -843,10 +838,10 @@ def _prepare_completion_args(
             or 1024
         )
         prompt_tensor = (self.processor if hasattr(self, "processor") else self.tokenizer).apply_chat_template(
-            messages,  # type: ignore
-            tools=[get_tool_json_schema(tool) for tool in tools_to_call_from] if tools_to_call_from else None,
+            messages,
+            tools=tools,
             return_tensors="pt",
-            add_generation_prompt=True if tools_to_call_from else False,
+            add_generation_prompt=True,
             tokenize=True,
             return_dict=True,
         )

tests/test_models.py

@@ -163,14 +163,14 @@ def test_transformers_message_no_tool(self, monkeypatch):
             do_sample=False,
         )
         messages = [{"role": "user", "content": [{"type": "text", "text": "Hello!"}]}]
-        output = model.generate(messages, stop_sequences=["great"]).content
-        assert output == "assistant\nHello"
+        output = model.generate(messages).content
+        assert output == "Hello! I'm here"
 
         output = model.generate_stream(messages, stop_sequences=["great"])
         output_str = ""
         for el in output:
             output_str += el.content
-        assert output_str == "assistant\nHello"
+        assert output_str == "Hello! I'm here"
 
     def test_transformers_message_vl_no_tool(self, shared_datadir, monkeypatch):
         monkeypatch.setattr("huggingface_hub.constants.HF_HUB_DOWNLOAD_TIMEOUT", 30)  # instead of 10
@@ -183,15 +183,17 @@ def test_transformers_message_vl_no_tool(self, shared_datadir, monkeypatch):
             device_map="cpu",
             do_sample=False,
         )
-        messages = [{"role": "user", "content": [{"type": "text", "text": "Hello!"}, {"type": "image", "image": img}]}]
-        output = model.generate(messages, stop_sequences=["great"]).content
-        assert output == "I am"
+        messages = [
+            {"role": "user", "content": [{"type": "text", "text": "What is this?"}, {"type": "image", "image": img}]}
+        ]
+        output = model.generate(messages).content
+        assert output == "This is a very"
 
         output = model.generate_stream(messages, stop_sequences=["great"])
         output_str = ""
         for el in output:
             output_str += el.content
-        assert output_str == "I am"
+        assert output_str == "This is a very"
 
     def test_parse_json_if_needed(self):
         args = "abc"