Refactor the yaml setup of kernel components

aios/config/config.yaml

@@ -11,14 +11,35 @@ api_keys:
     home: ""        # Optional: HuggingFace models path
 
 # LLM Configuration
-llm:
-  default_model: "gpt-4"
-  max_new_tokens: 256
-  backend: "openai"
-  max_gpu_memory: null
-  eval_device: "cuda:0"
+llms:
+  models:
+    - name: "gpt-4o-mini"
+      backend: "openai"
+      max_new_tokens: 256
+      temperature: 1.0
+
+    - name: "gemini-1.5-flash"
+      backend: "google"
+      max_new_tokens: 1024
+      temperature: 1.0
+
   log_mode: "console"
 
+memory:
+  memory_limit: 524288 # 512KB
+  eviction_k: 3
+
+storage:
+  root_dir: "root"
+  use_vector_db: true
+
+scheduler:
+  log_mode: "console"
+
+agent_factory:
+  log_mode: "console"
+  max_workers: 64  
+
 server:
   host: "localhost"
   port: 8000

aios/config/config_manager.py

@@ -80,24 +80,33 @@ def get_api_key(self, provider: str) -> Optional[str]:
 
         return api_key
 
-    def get_llm_config(self) -> dict:
-        """Get LLM configuration"""
-        llm_config = self.config.get('llm', {})
-        if not llm_config:
-            # Provide default configuration
-            llm_config = {
-                "default_model": "gpt-4",
-                "max_new_tokens": 256,
-                "backend": "openai",
-                "max_gpu_memory": None,
-                "eval_device": "cuda:0",
-                "log_mode": "console"
-            }
-        return llm_config
+    def get_llms_config(self) -> dict:
+        """Get configuration of LLMs"""
+        return self.config.get('llms', {})
+
+    def get_storage_config(self) -> dict:
+        """Get storage configuration"""
+        return self.config.get("storage", {})
+
+    def get_memory_config(self) -> dict:
+        """Get memory configuration"""
+        return self.config.get("memory", {})
+
+    def get_tool_config(self) -> dict:
+        """Get tool configuration"""
+        return self.config.get("tools", {})
+
+    def get_scheduler_config(self) -> dict:
+        """Get scheduler configuration"""
+        return self.config.get("scheduler", {})
+
+    def get_agent_factory_config(self) -> dict:
+        """Get agent factory configuration"""
+        return self.config.get("agent_factory", {})
 
-    def get_kernel_config(self) -> dict:
-        """Get kernel configuration"""
-        return self.config.get("kernel", {})
+    # def get_kernel_config(self) -> dict:
+    #     """Get kernel configuration"""
+    #     return self.config.get("kernel", {})
 
 # Global config instance
 config = ConfigManager() 

aios/hooks/modules/scheduler.py

@@ -12,25 +12,29 @@
     # QueueCheckEmpty,
 )
 
+from aios.hooks.types.llm import LLMRequestQueue
+from aios.hooks.types.memory import MemoryRequestQueue
+from aios.hooks.types.storage import StorageRequestQueue
+from aios.hooks.types.tool import ToolRequestQueue
+
 from contextlib import contextmanager
 
 from aios.hooks.utils.validate import validate
 from aios.hooks.stores import queue as QueueStore, processes as ProcessStore
 from aios.scheduler.fifo_scheduler import FIFOScheduler
+# from aios.scheduler.rr_scheduler import RRScheduler
 
 
 @validate(SchedulerParams)
 def useFIFOScheduler(
     params: SchedulerParams,
-) -> Tuple[Callable[[], None], Callable[[], None]]:
+):
     """
     Initialize and return a scheduler with start and stop functions.
 
     Args:
         params (SchedulerParams): Parameters required for the scheduler.
 
-    Returns:
-        Tuple: A tuple containing the start and stop functions for the scheduler.
     """
     if params.get_llm_syscall is None:
         from aios.hooks.stores._global import global_llm_req_queue_get_message
@@ -47,7 +51,19 @@ def useFIFOScheduler(
     if params.get_tool_syscall is None:
         from aios.hooks.stores._global import global_tool_req_queue_get_message
         params.get_tool_syscall = global_tool_req_queue_get_message
-
+
+    # if params.llm_request_queue is None:
+    #     params.llm_request_queue = LLMRequestQueue
+
+    # if params.memory_request_queue is None:
+    #     params.memory_request_queue = MemoryRequestQueue
+
+    # if params.storage_request_queue is None:
+    #     params.storage_request_queue = StorageRequestQueue
+
+    # if params.tool_request_queue is None:
+    #     params.tool_request_queue = ToolRequestQueue
+
     scheduler = FIFOScheduler(**params.model_dump())
 
     # Function to start the scheduler
@@ -70,21 +86,33 @@ def fifo_scheduler(params: SchedulerParams):
     Args:
         params (SchedulerParams): The parameters for the scheduler.
     """
-    if params.get_llm_syscall is None:
-        from aios.hooks.stores._global import global_llm_req_queue_get_message
-        params.get_llm_syscall = global_llm_req_queue_get_message
+    # if params.get_llm_syscall is None:
+    #     from aios.hooks.stores._global import global_llm_req_queue_get_message
+    #     params.get_llm_syscall = global_llm_req_queue_get_message
 
-    if params.get_memory_syscall is None:
-        from aios.hooks.stores._global import global_memory_req_queue_get_message
-        params.get_memory_syscall = global_memory_req_queue_get_message
+    # if params.get_memory_syscall is None:
+    #     from aios.hooks.stores._global import global_memory_req_queue_get_message
+    #     params.get_memory_syscall = global_memory_req_queue_get_message
 
-    if params.get_storage_syscall is None:
-        from aios.hooks.stores._global import global_storage_req_queue_get_message
-        params.get_storage_syscall = global_storage_req_queue_get_message
+    # if params.get_storage_syscall is None:
+    #     from aios.hooks.stores._global import global_storage_req_queue_get_message
+    #     params.get_storage_syscall = global_storage_req_queue_get_message
 
-    if params.get_tool_syscall is None:
-        from aios.hooks.stores._global import global_tool_req_queue_get_message
-        params.get_tool_syscall = global_tool_req_queue_get_message
+    # if params.get_tool_syscall is None:
+    #     from aios.hooks.stores._global import global_tool_req_queue_get_message
+    #     params.get_tool_syscall = global_tool_req_queue_get_message
+
+    if params.llm_request_queue is None:
+        params.llm_request_queue = LLMRequestQueue
+
+    if params.memory_request_queue is None:
+        params.memory_request_queue = MemoryRequestQueue
+
+    if params.storage_request_queue is None:
+        params.storage_request_queue = StorageRequestQueue
+
+    if params.tool_request_queue is None:
+        params.tool_request_queue = ToolRequestQueue
 
     scheduler = FIFOScheduler(**params.model_dump())
 
@@ -116,6 +144,18 @@ def fifo_scheduler_nonblock(params: SchedulerParams):
         from aios.hooks.stores._global import global_tool_req_queue_get_message
         params.get_tool_syscall = global_tool_req_queue_get_message
 
+    # if params.llm_request_queue is None:
+    #     params.llm_request_queue = LLMRequestQueue
+
+    # if params.memory_request_queue is None:
+    #     params.memory_request_queue = MemoryRequestQueue
+
+    # if params.storage_request_queue is None:
+    #     params.storage_request_queue = StorageRequestQueue
+
+    # if params.tool_request_queue is None:
+    #     params.tool_request_queue = ToolRequestQueue
+
     scheduler = FIFOScheduler(**params.model_dump())
 
     return scheduler

aios/hooks/stores/queue.py

@@ -1,14 +1,18 @@
 from queue import Queue
-
+from typing import List
 REQUEST_QUEUE: dict[str, Queue] = {}
+# REQUEST_QUEUE: dict[str, []] = {}
+
 
-def getMessage(q: Queue):
+def getMessage(q: List):
     return q.get(block=True, timeout=0.1)
+    # return q.pop(0)
 
-def addMessage(q: Queue, message: str):
+def addMessage(q: List, message: str):
+    # q.append(message)
     q.put(message)
-
     return None
 
-def isEmpty(q: Queue):
-    return q.empty()
+def isEmpty(q: List):
+    # return len(q) == 0
+    return q.empty()

aios/hooks/syscall.py

@@ -13,8 +13,17 @@
     global_memory_req_queue_add_message,
     global_storage_req_queue_add_message,
     global_tool_req_queue_add_message,
+    # global_llm_req_queue,
+    # global_memory_req_queue,
+    # global_storage_req_queue,
+    # global_tool_req_queue,
 )
 
+from aios.hooks.types.llm import LLMRequestQueue
+from aios.hooks.types.memory import MemoryRequestQueue
+from aios.hooks.types.storage import StorageRequestQueue
+from aios.hooks.types.tool import ToolRequestQueue
+
 from cerebrum.llm.communication import LLMQuery
 from cerebrum.memory.communication import MemoryQuery
 from cerebrum.storage.communication import StorageQuery
@@ -41,6 +50,7 @@ def storage_syscall_exec(agent_name, query):
             syscall.set_response(None)
 
             global_storage_req_queue_add_message(syscall)
+            # StorageRequestQueue.append(syscall)
 
             syscall.start()
             syscall.join()
@@ -49,6 +59,7 @@ def storage_syscall_exec(agent_name, query):
 
             if syscall.get_status() != "done":
                 pass
+
             start_time = syscall.get_start_time()
             end_time = syscall.get_end_time()
             waiting_time = start_time - syscall.get_created_time()
@@ -85,6 +96,7 @@ def mem_syscall_exec(agent_name, query):
             syscall.set_response(None)
 
             global_memory_req_queue_add_message(syscall)
+            # MemoryRequestQueue.append(syscall)
 
             syscall.start()
             syscall.join()
@@ -129,6 +141,7 @@ def tool_syscall_exec(agent_name, tool_calls):
             syscall.set_response(None)
 
             global_tool_req_queue_add_message(syscall)
+            # ToolRequestQueue.append(syscall)
 
             syscall.start()
             syscall.join()
@@ -175,6 +188,7 @@ def llm_syscall_exec(agent_name, query):
             syscall.set_response(None)
 
             global_llm_req_queue_add_message(syscall)
+            # LLMRequestQueue.append(syscall)
 
             syscall.start()
             syscall.join()
@@ -202,7 +216,7 @@ def llm_syscall_exec(agent_name, query):
         }
 
 
-    def send_request(agent_name, query):
+    def execute_request(agent_name, query):
         if isinstance(query, LLMQuery):
             action_type = query.action_type
             # print(action_type)
@@ -281,4 +295,4 @@ class SysCallWrapper:
         memory = memoryview
         tool = tool_syscall_exec
 
-    return send_request, SysCallWrapper
+    return execute_request, SysCallWrapper

aios/hooks/types/llm.py

@@ -3,6 +3,7 @@
 from queue import Queue
 
 LLMRequestQueue: TypeAlias = Queue
+# LLMRequestQueue: TypeAlias = []
 
 LLMRequestQueueGetMessage: TypeAlias = Callable[[], None]
 LLMRequestQueueAddMessage: TypeAlias = Callable[[str], None]

aios/hooks/types/memory.py

@@ -3,6 +3,7 @@
 from queue import Queue
 
 MemoryRequestQueue: TypeAlias = Queue
+# MemoryRequestQueue: TypeAlias = []
 
 MemoryRequestQueueGetMessage: TypeAlias = Callable[[], None]
 MemoryRequestQueueAddMessage: TypeAlias = Callable[[str], None]

aios/hooks/types/scheduler.py

@@ -5,14 +5,24 @@
 from .memory import MemoryRequestQueueGetMessage
 from .storage import StorageRequestQueueGetMessage
 from .tool import ToolRequestQueueGetMessage
+# from .llm import LLMRequestQueue
+# from .memory import MemoryRequestQueue
+# from .storage import StorageRequestQueue
+# from .tool import ToolRequestQueue
+
 
 class SchedulerParams(BaseModel):
     llm: Any
     memory_manager: Any
     storage_manager: Any
     tool_manager: Any
     log_mode: str
+    # llm_request_queue: Any
+    # memory_request_queue: Any
+    # storage_request_queue: Any
+    # tool_request_queue: Any
     get_llm_syscall: LLMRequestQueueGetMessage | None
     get_memory_syscall: MemoryRequestQueueGetMessage | None
     get_storage_syscall: StorageRequestQueueGetMessage | None
-    get_tool_syscall: ToolRequestQueueGetMessage | None
+    get_tool_syscall: ToolRequestQueueGetMessage | None
+

aios/hooks/types/storage.py

@@ -3,6 +3,7 @@
 from queue import Queue
 
 StorageRequestQueue: TypeAlias = Queue
+# StorageRequestQueue: TypeAlias = []
 
 StorageRequestQueueGetMessage: TypeAlias = Callable[[], None]
 StorageRequestQueueAddMessage: TypeAlias = Callable[[str], None]

aios/hooks/types/tool.py

@@ -3,6 +3,7 @@
 from queue import Queue
 
 ToolRequestQueue: TypeAlias = Queue
+# ToolRequestQueue: TypeAlias = []
 
 ToolRequestQueueGetMessage: TypeAlias = Callable[[], None]
 ToolRequestQueueAddMessage: TypeAlias = Callable[[str], None]

aios/llm_core/llm_cost_map.json

@@ -0,0 +1,11 @@
+[
+    {"name": "Qwen-2.5-7B-Instruct", "hosted_model": "ollama/qwen2.5:7b", "cost_per_input_token": 0.000000267, "cost_per_output_token": 0.000000267},
+    {"name": "Qwen-2.5-14B-Instruct", "hosted_model": "ollama/qwen2.5:14b", "cost_per_input_token": 0.000000534, "cost_per_output_token": 0.000000534},
+    {"name": "Qwen-2.5-32B-Instruct", "hosted_model": "ollama/qwen2.5:32b", "cost_per_input_token": 0.00000122, "cost_per_output_token": 0.00000122},
+    {"name": "Llama-3.1-8B-Instruct", "hosted_model": "ollama/llama3.1:8b", "cost_per_input_token": 0.000000305, "cost_per_output_token": 0.000000305},
+    {"name": "Deepseek-r1-7b", "hosted_model": "ollama/deepseek-r1:7b", "cost_per_input_token": 0.000000267, "cost_per_output_token": 0.000000267},
+    {"name": "Deepseek-r1-14b", "hosted_model": "ollama/deepseek-r1:14b", "cost_per_input_token": 0.000000534, "cost_per_output_token": 0.000000534},
+    {"name": "gpt-4o-mini", "hosted_model": "gpt-4o-mini", "cost_per_input_token": 0.00000015, "cost_per_output_token": 0.0000006},
+    {"name": "gpt-4o", "hosted_model": "gpt-4o", "cost_per_input_token": 0.0000025, "cost_per_output_token": 0.00001},
+    {"name": "gemini-1.5-flash", "hosted_model": "gemini/gemini-1.5-flash", "cost_per_input_token": 0.000000075, "cost_per_output_token": 0.0000003}
+  ]