Merge branch 'main' into add-watsonx-inference-adapter

llama_stack/models/llama/llama3/prompt_templates/system_prompts.py

@@ -279,6 +279,10 @@ class PythonListCustomToolGenerator(PromptTemplateGeneratorBase):  # noqa: N801
                 {% endif -%}
                 {%- endfor %}
             ]
+
+            You can answer general questions or invoke tools when necessary.
+            In addition to tool calls, you should also augment your responses by using the tool outputs.
+
             """
         )
         return PromptTemplate(

llama_stack/models/llama/llama4/chat_format.py

@@ -216,9 +216,12 @@ class ChatFormat:
                 content = ToolUtils.encode_tool_call(t, tool_prompt_format)
                 _process_content(content)
 
+        # Tool calls and Tool Response messages should be eom
         eom = False
         if message.role == "assistant":
-            eom = message.stop_reason == StopReason.end_of_message
+            eom = message.stop_reason == StopReason.end_of_message or message.tool_calls
+        elif message.role == "tool":
+            eom = True
 
         tokens.append(self.tokenizer.special_tokens["<|eom|>" if eom else "<|eot|>"])
         return tokens, images

llama_stack/providers/inline/inference/meta_reference/inference.py

@@ -6,8 +6,11 @@
 
 import asyncio
 import logging
+import os
 from typing import AsyncGenerator, List, Optional, Union
 
+from termcolor import cprint
+
 from llama_stack.apis.common.content_types import (
     TextDelta,
     ToolCallDelta,
@@ -338,6 +341,9 @@ class MetaReferenceInferenceImpl(
             stop_reason = None
 
             for token_result in self.generator.chat_completion(request):
+                if os.environ.get("LLAMA_MODELS_DEBUG", "0") == "1":
+                    cprint(token_result.text, "cyan", end="")
+
                 tokens.append(token_result.token)
 
                 if token_result.token == tokenizer.eot_id:
@@ -386,6 +392,9 @@ class MetaReferenceInferenceImpl(
             ipython = False
 
             for token_result in self.generator.chat_completion(request):
+                if os.environ.get("LLAMA_MODELS_DEBUG", "0") == "1":
+                    cprint(token_result.text, "cyan", end="")
+
                 tokens.append(token_result.token)
 
                 if not ipython and token_result.text.startswith("<|python_tag|>"):