symbolicai 1.0.0__py3-none-any.whl → 1.1.0__py3-none-any.whl

symai/backend/engines/neurosymbolic/engine_llama_cpp.py

@@ -24,14 +24,17 @@ class LlamaCppTokenizer:
 
     @staticmethod
     async def _encode(text: str) -> list[int]:
-        async with aiohttp.ClientSession() as session, session.post(
-            f"{LlamaCppTokenizer._server_endpoint}/tokenize",
-            json={"content": text},
-        ) as res:
+        async with (
+            aiohttp.ClientSession() as session,
+            session.post(
+                f"{LlamaCppTokenizer._server_endpoint}/tokenize",
+                json={"content": text},
+            ) as res,
+        ):
             if res.status != 200:
                 UserMessage(f"Request failed with status code: {res.status}", raise_with=ValueError)
             response_json = await res.json()
-            return response_json['tokens']
+            return response_json["tokens"]
 
     @staticmethod
     def encode(text: str) -> list[int]:
@@ -44,14 +47,17 @@ class LlamaCppTokenizer:
 
     @staticmethod
     async def _decode(tokens: list[int]) -> str:
-        async with aiohttp.ClientSession() as session, session.post(
-            f"{LlamaCppTokenizer._server_endpoint}/detokenize",
-            json={"tokens": tokens},
-        ) as res:
+        async with (
+            aiohttp.ClientSession() as session,
+            session.post(
+                f"{LlamaCppTokenizer._server_endpoint}/detokenize",
+                json={"tokens": tokens},
+            ) as res,
+        ):
             if res.status != 200:
                 UserMessage(f"Request failed with status code: {res.status}", raise_with=ValueError)
             response_json = await res.json()
-            return response_json['content']
+            return response_json["content"]
 
     @staticmethod
     def decode(tokens: list[int]) -> str:
@@ -65,70 +71,83 @@ class LlamaCppTokenizer:
 
 class LlamaCppEngine(Engine):
     _retry_params: ClassVar[dict[str, Any]] = {
-        'tries': 5,
-        'delay': 2,
-        'max_delay': 60,
-        'backoff': 2,
-        'jitter': (1, 5),
-        'graceful': True
+        "tries": 5,
+        "delay": 2,
+        "max_delay": 60,
+        "backoff": 2,
+        "jitter": (1, 5),
+        "graceful": True,
     }
     _timeout_params: ClassVar[dict[str, Any]] = {
-        'read': None,
-        'connect': None,
+        "read": None,
+        "connect": None,
     }
+
     def __init__(
-            self,
-            model: str | None = None,
-            retry_params: dict = _retry_params,
-            timeout_params: dict = _timeout_params,
-        ):
+        self,
+        model: str | None = None,
+        retry_params: dict = _retry_params,
+        timeout_params: dict = _timeout_params,
+    ):
         super().__init__()
         self.config = deepcopy(SYMAI_CONFIG)
         # In case we use EngineRepository.register to inject the api_key and model => dynamically change the engine at runtime
         if model is not None:
-            self.config['NEUROSYMBOLIC_ENGINE_MODEL'] = model
-        if self.id() != 'neurosymbolic':
+            self.config["NEUROSYMBOLIC_ENGINE_MODEL"] = model
+        if self.id() != "neurosymbolic":
             return
-        if not SYMSERVER_CONFIG.get('online'):
-            UserMessage('You are using the llama.cpp engine, but the server endpoint is not started. Please start the server with `symserver [--args]` or run `symserver --help` to see the available options for this engine.', raise_with=ValueError)
-        self.server_endpoint = f"http://{SYMSERVER_CONFIG.get('--host')}:{SYMSERVER_CONFIG.get('--port')}"
-        self.tokenizer = LlamaCppTokenizer # backwards compatibility with how we handle tokenization, i.e. self.tokenizer().encode(...)
+        if not SYMSERVER_CONFIG.get("online"):
+            UserMessage(
+                "You are using the llama.cpp engine, but the server endpoint is not started. Please start the server with `symserver [--args]` or run `symserver --help` to see the available options for this engine.",
+                raise_with=ValueError,
+            )
+        self.server_endpoint = (
+            f"http://{SYMSERVER_CONFIG.get('--host')}:{SYMSERVER_CONFIG.get('--port')}"
+        )
+        self.tokenizer = LlamaCppTokenizer  # backwards compatibility with how we handle tokenization, i.e. self.tokenizer().encode(...)
         self.timeout_params = self._validate_timeout_params(timeout_params)
         self.retry_params = self._validate_retry_params(retry_params)
         self.name = self.__class__.__name__
 
     def id(self) -> str:
-        if self.config.get('NEUROSYMBOLIC_ENGINE_MODEL') and self.config.get('NEUROSYMBOLIC_ENGINE_MODEL').startswith('llama'):
-            return 'neurosymbolic'
-        return super().id() # default to unregistered
+        if self.config.get("NEUROSYMBOLIC_ENGINE_MODEL") and self.config.get(
+            "NEUROSYMBOLIC_ENGINE_MODEL"
+        ).startswith("llama"):
+            return "neurosymbolic"
+        return super().id()  # default to unregistered
 
     def command(self, *args, **kwargs):
         super().command(*args, **kwargs)
-        if 'NEUROSYMBOLIC_ENGINE_MODEL' in kwargs:
-            self.model = kwargs['NEUROSYMBOLIC_ENGINE_MODEL']
-        if 'seed' in kwargs:
-            self.seed = kwargs['seed']
-        if 'except_remedy' in kwargs:
-            self.except_remedy = kwargs['except_remedy']
+        if "NEUROSYMBOLIC_ENGINE_MODEL" in kwargs:
+            self.model = kwargs["NEUROSYMBOLIC_ENGINE_MODEL"]
+        if "seed" in kwargs:
+            self.seed = kwargs["seed"]
+        if "except_remedy" in kwargs:
+            self.except_remedy = kwargs["except_remedy"]
 
     def compute_required_tokens(self, _messages) -> int:
-        #@TODO: quite non-trivial how to handle this with the llama.cpp server
-        UserMessage('Not implemented for llama.cpp!', raise_with=NotImplementedError)
+        # @TODO: quite non-trivial how to handle this with the llama.cpp server
+        UserMessage("Not implemented for llama.cpp!", raise_with=NotImplementedError)
 
     def compute_remaining_tokens(self, _prompts: list) -> int:
-        #@TODO: quite non-trivial how to handle this with the llama.cpp server
-        UserMessage('Not implemented for llama.cpp!', raise_with=NotImplementedError)
+        # @TODO: quite non-trivial how to handle this with the llama.cpp server
+        UserMessage("Not implemented for llama.cpp!", raise_with=NotImplementedError)
 
     def _validate_timeout_params(self, timeout_params):
         if not isinstance(timeout_params, dict):
             UserMessage("timeout_params must be a dictionary", raise_with=ValueError)
-        assert all(key in timeout_params for key in ['read', 'connect']), "Available keys: ['read', 'connect']"
+        assert all(key in timeout_params for key in ["read", "connect"]), (
+            "Available keys: ['read', 'connect']"
+        )
         return timeout_params
 
     def _validate_retry_params(self, retry_params):
         if not isinstance(retry_params, dict):
             UserMessage("retry_params must be a dictionary", raise_with=ValueError)
-        assert all(key in retry_params for key in ['tries', 'delay', 'max_delay', 'backoff', 'jitter', 'graceful']), "Available keys: ['tries', 'delay', 'max_delay', 'backoff', 'jitter', 'graceful']"
+        assert all(
+            key in retry_params
+            for key in ["tries", "delay", "max_delay", "backoff", "jitter", "graceful"]
+        ), "Available keys: ['tries', 'delay', 'max_delay', 'backoff', 'jitter', 'graceful']"
         return retry_params
 
     @staticmethod
@@ -149,35 +168,35 @@ class LlamaCppEngine(Engine):
         kwargs = argument.kwargs
         payload = {
             "messages": argument.prop.prepared_input,
-            "temperature": kwargs.get('temperature', 0.6),
-            "frequency_penalty": kwargs.get('frequency_penalty', 0),
-            "presence_penalty": kwargs.get('presence_penalty', 0),
-            "top_p": kwargs.get('top_p', 0.95),
-            "min_p": kwargs.get('min_p', 0.05),
-            "stop": kwargs.get('stop'),
-            "seed": kwargs.get('seed'),
-            "max_tokens": kwargs.get('max_tokens'),
-            "top_k": kwargs.get('top_k', 40),
-            "repeat_penalty": kwargs.get('repeat_penalty', 1),
-            "logits_bias": kwargs.get('logits_bias'),
-            "logprobs": kwargs.get('logprobs', False),
-            "grammar": kwargs.get('grammar'),
-            "response_format": kwargs.get('response_format'),
+            "temperature": kwargs.get("temperature", 0.6),
+            "frequency_penalty": kwargs.get("frequency_penalty", 0),
+            "presence_penalty": kwargs.get("presence_penalty", 0),
+            "top_p": kwargs.get("top_p", 0.95),
+            "min_p": kwargs.get("min_p", 0.05),
+            "stop": kwargs.get("stop"),
+            "seed": kwargs.get("seed"),
+            "max_tokens": kwargs.get("max_tokens"),
+            "top_k": kwargs.get("top_k", 40),
+            "repeat_penalty": kwargs.get("repeat_penalty", 1),
+            "logits_bias": kwargs.get("logits_bias"),
+            "logprobs": kwargs.get("logprobs", False),
+            "grammar": kwargs.get("grammar"),
+            "response_format": kwargs.get("response_format"),
         }
 
-        model = SYMSERVER_CONFIG.get('-m') or SYMSERVER_CONFIG.get('--model')
+        model = SYMSERVER_CONFIG.get("-m") or SYMSERVER_CONFIG.get("--model")
         if model:
             payload["model"] = model
 
-        tools = kwargs.get('tools')
+        tools = kwargs.get("tools")
         if tools:
             payload["tools"] = tools
 
-        tool_choice = kwargs.get('tool_choice')
+        tool_choice = kwargs.get("tool_choice")
         if tool_choice is not None:
             payload["tool_choice"] = tool_choice
 
-        extra_body = kwargs.get('extra_body')
+        extra_body = kwargs.get("extra_body")
         if isinstance(extra_body, dict):
             payload.update(extra_body)
 
@@ -185,18 +204,20 @@ class LlamaCppEngine(Engine):
 
     async def _arequest(self, payload: dict) -> dict:
         """Makes an async HTTP request to the llama.cpp server."""
+
         @retry(**self.retry_params)
         async def _make_request():
             timeout = aiohttp.ClientTimeout(
-                sock_connect=self.timeout_params['connect'],
-                sock_read=self.timeout_params['read']
+                sock_connect=self.timeout_params["connect"], sock_read=self.timeout_params["read"]
             )
-            async with aiohttp.ClientSession(timeout=timeout) as session, session.post(
-                f"{self.server_endpoint}/v1/chat/completions",
-                json=payload
-            ) as res:
+            async with (
+                aiohttp.ClientSession(timeout=timeout) as session,
+                session.post(f"{self.server_endpoint}/v1/chat/completions", json=payload) as res,
+            ):
                 if res.status != 200:
-                    UserMessage(f"Request failed with status code: {res.status}", raise_with=ValueError)
+                    UserMessage(
+                        f"Request failed with status code: {res.status}", raise_with=ValueError
+                    )
                 return await res.json()
 
         return await _make_request()
@@ -206,12 +227,12 @@ class LlamaCppEngine(Engine):
         """Extract reasoning traces from llama.cpp responses."""
         if not isinstance(response, dict):
             return None
-        choices = response.get('choices', [])
+        choices = response.get("choices", [])
         if not isinstance(choices, list) or not choices:
             return None
         for choice in choices:
-            if isinstance(choice, dict) and isinstance(choice.get('message'), dict):
-                return choice['message'].get('reasoning_content')
+            if isinstance(choice, dict) and isinstance(choice.get("message"), dict):
+                return choice["message"].get("reasoning_content")
         return None
 
     def forward(self, argument):
@@ -223,50 +244,54 @@ class LlamaCppEngine(Engine):
         try:
             res = loop.run_until_complete(self._arequest(payload))
         except Exception as e:
-            UserMessage(f'Error during generation. Caused by: {e}', raise_with=ValueError)
+            UserMessage(f"Error during generation. Caused by: {e}", raise_with=ValueError)
 
-        metadata = {'raw_output': res}
+        metadata = {"raw_output": res}
 
-        if payload.get('tools'):
+        if payload.get("tools"):
             metadata = self._process_tool_calls(res, metadata)
 
         thinking = self._extract_thinking(res)
         if thinking:
-            metadata['thinking'] = thinking
+            metadata["thinking"] = thinking
 
-        output = [r['message']['content'] for r in res['choices']]
+        output = [r["message"]["content"] for r in res["choices"]]
         output = output if isinstance(argument.prop.prepared_input, list) else output[0]
 
         return output, metadata
 
     @staticmethod
     def _process_tool_calls(res, metadata):
-        choices = res.get('choices') if isinstance(res, dict) else None
+        choices = res.get("choices") if isinstance(res, dict) else None
         if not choices:
             return metadata
         hit = False
         for choice in choices:
             if not isinstance(choice, dict):
                 continue
-            message = choice.get('message') or {}
-            tool_calls = message.get('tool_calls') or []
+            message = choice.get("message") or {}
+            tool_calls = message.get("tool_calls") or []
             if not tool_calls:
                 continue
             for tool_call in tool_calls:
                 if not isinstance(tool_call, dict):
                     continue
-                function = tool_call.get('function') or {}
+                function = tool_call.get("function") or {}
                 if hit:
-                    UserMessage("Multiple function calls detected in the response but only the first one will be processed.")
+                    UserMessage(
+                        "Multiple function calls detected in the response but only the first one will be processed."
+                    )
                     return metadata
-                arguments = function.get('arguments')
+                arguments = function.get("arguments")
                 try:
-                    args_dict = json.loads(arguments) if isinstance(arguments, str) else arguments or {}
+                    args_dict = (
+                        json.loads(arguments) if isinstance(arguments, str) else arguments or {}
+                    )
                 except json.JSONDecodeError:
                     args_dict = {}
-                metadata['function_call'] = {
-                    'name': function.get('name'),
-                    'arguments': args_dict or {}
+                metadata["function_call"] = {
+                    "name": function.get("name"),
+                    "arguments": args_dict or {},
                 }
                 hit = True
                 break
@@ -276,11 +301,14 @@ class LlamaCppEngine(Engine):
 
     def _prepare_raw_input(self, argument):
         if not argument.prop.processed_input:
-            UserMessage('Need to provide a prompt instruction to the engine if raw_input is enabled.', raise_with=ValueError)
+            UserMessage(
+                "Need to provide a prompt instruction to the engine if raw_input is enabled.",
+                raise_with=ValueError,
+            )
         value = argument.prop.processed_input
         if not isinstance(value, list):
             if not isinstance(value, dict):
-                value = {'role': 'user', 'content': str(value)}
+                value = {"role": "user", "content": str(value)}
             value = [value]
         return value
 
@@ -291,15 +319,15 @@ class LlamaCppEngine(Engine):
 
         _non_verbose_output = """<META_INSTRUCTION/>\n You will NOT output verbose preambles or post explanation, such as "Sure, let me...", "Hope that was helpful...", "Yes, I can help you with that...", etc. You will consider well formatted output, e.g. for sentences you will use punctuation, spaces, etc. or for code indentation, etc.\n"""
 
-        #@TODO: Non-trivial how to handle user/system/assistant roles;
+        # @TODO: Non-trivial how to handle user/system/assistant roles;
         #       For instance Mixtral-8x7B can't use the system role with llama.cpp while other models can, or Mixtral-8x22B expects the conversation roles must
         #       alternate user/assistant/user/assistant/..., so how to handle this?
         #       For now just use the user, as one can rephrase the system from the user perspective.
-        user:   str = ""
+        user: str = ""
 
         if argument.prop.suppress_verbose_output:
             user += _non_verbose_output
-        user = f'{user}\n' if user and len(user) > 0 else ''
+        user = f"{user}\n" if user and len(user) > 0 else ""
 
         ref = argument.prop.instance
         static_ctxt, dyn_ctxt = ref.global_context
@@ -326,5 +354,5 @@ class LlamaCppEngine(Engine):
         user += str(argument.prop.processed_input)
 
         argument.prop.prepared_input = [
-            { "role": "user", "content": user },
+            {"role": "user", "content": user},
         ]