[WIP] feat: Enable native tool calling for Granite

python/beeai_framework/agents/runners/default/runner.py

@@ -48,6 +48,9 @@
 
 
 class DefaultRunner(BaseRunner):
+    # TODO: 333 - Currently a global variable is used to determine if the tool calling should be used or not
+    use_native_tool_calling: bool = False
+
     def default_templates(self) -> BeeAgentTemplates:
         return BeeAgentTemplates(
             system=SystemPromptTemplate,
@@ -98,8 +101,13 @@ async def new_token(value: tuple[ChatModelOutput, Callable], event: EventMeta) -
         def observe(llm_emitter: Emitter) -> None:
             llm_emitter.on("newToken", new_token)
 
+        # TODO: 333 check this exists, shallow copy
+        tools: list[Tool] = self._input.tools[:]
         output: ChatModelOutput = await self._input.llm.create(
-            ChatModelInput(messages=self.memory.messages[:], stream=True)
+            # For native tool calling we pass the tools to the llm call.
+            ChatModelInput(
+                messages=self.memory.messages[:], stream=True, tools=tools if self.use_native_tool_calling else None
+            )
         ).observe(fn=observe)
 
         # Pick up any remaining lines in parser buffer

python/beeai_framework/agents/runners/granite/prompts.py

@@ -27,7 +27,8 @@
 
 GraniteAssistantPromptTemplate = PromptTemplate(
     schema=AssistantPromptTemplateInput,
-    template="{{#thought}}Thought: {{.}}\n{{/thought}}{{#tool_name}}Tool Name: {{.}}\n{{/tool_name}}{{#tool_input}}Tool Input: {{&.}}\n{{/tool_input}}{{#tool_output}}Tool Output: {{&.}}\n{{/tool_output}}{{#final_answer}}Final Answer: {{.}}{{/final_answer}}",  # noqa: E501
+    # TODO: 333 updated to match ts - not passing tools
+    template="{{#thought}}Thought: {{.}}\n{{/thought}}{{#tool_name}}Tool Name: {{.}}\n{{/tool_name}}{{#tool_input}}Tool Input: {{.}}\n{{/tool_input}}{{#final_answer}}Final Answer: {{.}}{{/final_answer}}",  # noqa: E501
 )
 
 GraniteSystemPromptTemplate = PromptTemplate(
@@ -78,14 +79,33 @@
 """,  # noqa: E501
 )
 
+# TODO: 333 Additional prompts in ts
+# export const GraniteBeeSchemaErrorPrompt = BeeSchemaErrorPrompt.fork((config) => {
+#   config.template = `Error: The generated response does not adhere to the communication structure mentioned in the
+# system prompt.
+# You communicate only in instruction lines. Valid instruction lines are 'Thought' followed by 'Tool Name' and then
+# 'Tool Input' or 'Thought' followed by 'Final Answer'.`;
+# });
+
+# export const GraniteBeeUserPrompt = BeeUserPrompt.fork((config) => {
+#   config.template = `{{input}}`;
+# });
+# TODO: 333 Updated text string to match typescript (minor)
 GraniteToolNotFoundErrorTemplate = PromptTemplate(
     schema=ToolNotFoundErrorTemplateInput,
-    template="""The tool does not exist!
+    template="""Tool does not exist!
 {{#tools.length}}
 Use one of the following tools: {{#trim}}{{#tools}}{{name}},{{/tools}}{{/trim}}
 {{/tools.length}}""",
 )
 
+# TODO: 333 Additional prompt in ts
+# export const GraniteBeeToolErrorPrompt = BeeToolErrorPrompt.fork((config) => {
+#   config.template = `The tool has failed; the error log is shown below. If the tool cannot accomplish what you want,
+# use a different tool or explain why you can't use it.
+
+# {{reason}}`;
+# });
 GraniteToolInputErrorTemplate = PromptTemplate(
     schema=ToolInputErrorTemplateInput,
     template="""{{reason}}

python/beeai_framework/agents/runners/granite/runner.py

@@ -29,6 +29,9 @@
 
 
 class GraniteRunner(DefaultRunner):
+    # TODO:333 - Currently a global variable is used to determine if the tool calling should be used or not
+    use_native_tool_calling: bool = True
+
     def create_parser(self) -> LinePrefixParser:
         """Prefixes are renamed for granite"""
         prefixes = [