refactor(llama4): remove duplicate implementation, update imports to llama-models, add comprehensive test for tool calling fix (issue #2584)\n\n- Removes all old llama4 code from llama-stack\n- Updates all relevant imports to use llama-models\n- Adds robust pytest to demonstrate arguments_json fix\n- Updates config/scripts as needed for new structure\n- Resolves merge conflicts with updated main branch\n- Fixes mypy and ruff issues

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

@@ -9,6 +9,8 @@ from collections.abc import Generator
 from typing import Optional
 
 import torch
+from llama_models.llama4.generation import Llama4
+from llama_models.llama4.tokenizer import Tokenizer as Llama4Tokenizer
 from lmformatenforcer import JsonSchemaParser, TokenEnforcer, TokenEnforcerTokenizerData
 
 from llama_stack.apis.inference import (
@@ -21,8 +23,6 @@ from llama_stack.apis.inference import (
 from llama_stack.models.llama.datatypes import QuantizationMode
 from llama_stack.models.llama.llama3.generation import Llama3
 from llama_stack.models.llama.llama3.tokenizer import Tokenizer as Llama3Tokenizer
-from llama_stack.models.llama.llama4.generation import Llama4
-from llama_stack.models.llama.llama4.tokenizer import Tokenizer as Llama4Tokenizer
 from llama_stack.models.llama.sku_types import Model, ModelFamily
 from llama_stack.providers.utils.inference.prompt_adapter import (
     ChatCompletionRequestWithRawContent,
@@ -34,7 +34,7 @@ from .common import model_checkpoint_dir
 from .config import MetaReferenceInferenceConfig
 from .inference import resolve_model
 
-Tokenizer = Llama4Tokenizer | Llama3Tokenizer
+type Tokenizer = Llama4Tokenizer | Llama3Tokenizer
 
 
 class LogitsProcessor:

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

@@ -9,6 +9,8 @@ import os
 import sys
 from collections.abc import AsyncGenerator
 
+from llama_models.llama4.chat_format import ChatFormat as Llama4ChatFormat
+from llama_models.llama4.tokenizer import Tokenizer as Llama4Tokenizer
 from pydantic import BaseModel
 from termcolor import cprint
 
@@ -47,8 +49,6 @@ from llama_stack.apis.models import Model, ModelType
 from llama_stack.log import get_logger
 from llama_stack.models.llama.llama3.chat_format import ChatFormat as Llama3ChatFormat
 from llama_stack.models.llama.llama3.tokenizer import Tokenizer as Llama3Tokenizer
-from llama_stack.models.llama.llama4.chat_format import ChatFormat as Llama4ChatFormat
-from llama_stack.models.llama.llama4.tokenizer import Tokenizer as Llama4Tokenizer
 from llama_stack.models.llama.sku_list import resolve_model
 from llama_stack.models.llama.sku_types import ModelFamily
 from llama_stack.providers.datatypes import ModelsProtocolPrivate

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

@@ -9,8 +9,9 @@ from copy import deepcopy
 from functools import partial
 from typing import Any
 
+from llama_models.llama4.chat_format import ChatFormat as Llama4ChatFormat
+
 from llama_stack.models.llama.llama3.chat_format import ChatFormat as Llama3ChatFormat
-from llama_stack.models.llama.llama4.chat_format import ChatFormat as Llama4ChatFormat
 from llama_stack.providers.utils.inference.prompt_adapter import (
     ChatCompletionRequestWithRawContent,
     CompletionRequestWithRawContent,

llama_stack/providers/utils/inference/prompt_adapter.py

@@ -54,11 +54,11 @@ from llama_stack.models.llama.llama3.tokenizer import Tokenizer
 
 # Conditional imports to avoid heavy dependencies during module loading
 try:
-    from llama_stack.models.llama.llama4.chat_format import ChatFormat as Llama4ChatFormat
-    from llama_stack.models.llama.llama4.prompt_templates.system_prompts import (
+    from llama_models.llama4.chat_format import ChatFormat as Llama4ChatFormat
+    from llama_models.llama4.prompt_templates.system_prompts import (
         PythonListCustomToolGenerator as PythonListCustomToolGeneratorLlama4,
     )
-    from llama_stack.models.llama.llama4.tokenizer import Tokenizer as Llama4Tokenizer
+    from llama_models.llama4.tokenizer import Tokenizer as Llama4Tokenizer
 
     LLAMA4_AVAILABLE = True
 except ImportError: