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

symai/backend/engines/neurosymbolic/engine_cerebras.py

@@ -0,0 +1,328 @@
+import json
+import logging
+import re
+from copy import deepcopy
+
+from cerebras.cloud.sdk import Cerebras
+
+from ....components import SelfPrompt
+from ....core_ext import retry
+from ....utils import UserMessage
+from ...base import Engine
+from ...settings import SYMAI_CONFIG
+
+logging.getLogger("cerebras").setLevel(logging.ERROR)
+logging.getLogger("requests").setLevel(logging.ERROR)
+logging.getLogger("urllib").setLevel(logging.ERROR)
+logging.getLogger("httpx").setLevel(logging.ERROR)
+logging.getLogger("httpcore").setLevel(logging.ERROR)
+
+
+_NON_VERBOSE_OUTPUT = (
+    "<META_INSTRUCTION/>\n"
+    "You do not output anything else, like verbose preambles or post explanation, such as "
+    '"Sure, let me...", "Hope that was helpful...", "Yes, I can help you with that...", etc. '
+    "Consider well formatted output, e.g. for sentences use punctuation, spaces etc. or for code use "
+    "indentation, etc. Never add meta instructions information to your output!\n\n"
+)
+
+
+class CerebrasEngine(Engine):
+    def __init__(self, api_key: str | None = None, model: str | None = None):
+        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 api_key is not None and model is not None:
+            self.config["NEUROSYMBOLIC_ENGINE_API_KEY"] = api_key
+            self.config["NEUROSYMBOLIC_ENGINE_MODEL"] = model
+        if self.id() != "neurosymbolic":
+            # Do not initialize if not neurosymbolic; avoids conflict with llama.cpp check in
+            # EngineRepository.register_from_package.
+            return
+
+        self.api_key = self.config["NEUROSYMBOLIC_ENGINE_API_KEY"]
+        self.model = self.config["NEUROSYMBOLIC_ENGINE_MODEL"]
+        self.seed = None
+        self.name = self.__class__.__name__
+
+        try:
+            self.client = Cerebras(api_key=self.api_key)
+        except Exception as exc:
+            UserMessage(
+                f"Failed to initialize Cerebras client. Please check your Cerebras SDK installation. Caused by: {exc}",
+                raise_with=ValueError,
+            )
+
+    def id(self) -> str:
+        model_name = self.config.get("NEUROSYMBOLIC_ENGINE_MODEL")
+        if model_name and model_name.startswith("cerebras"):
+            return "neurosymbolic"
+        return super().id()
+
+    def command(self, *args, **kwargs):
+        super().command(*args, **kwargs)
+        if "NEUROSYMBOLIC_ENGINE_API_KEY" in kwargs:
+            self.api_key = kwargs["NEUROSYMBOLIC_ENGINE_API_KEY"]
+            try:
+                self.client = Cerebras(api_key=self.api_key)
+            except Exception as exc:
+                UserMessage(
+                    f"Failed to reinitialize Cerebras client. Caused by: {exc}",
+                    raise_with=ValueError,
+                )
+        if "NEUROSYMBOLIC_ENGINE_MODEL" in kwargs:
+            self.model = kwargs["NEUROSYMBOLIC_ENGINE_MODEL"]
+        if "seed" in kwargs:
+            self.seed = kwargs["seed"]
+
+    def compute_required_tokens(self, _messages):
+        UserMessage(
+            "Token counting not implemented for this engine.", raise_with=NotImplementedError
+        )
+
+    def compute_remaining_tokens(self, _prompts: list) -> int:
+        UserMessage(
+            "Token counting not implemented for this engine.", raise_with=NotImplementedError
+        )
+
+    def _handle_prefix(self, model_name: str) -> str:
+        """Handle prefix for model name."""
+        return model_name.replace("cerebras:", "")
+
+    def _extract_thinking_content(self, outputs: list[str]) -> tuple[str | None, list[str]]:
+        """Extract thinking content from textual output using <think>...</think> tags if present."""
+        if not outputs:
+            return None, outputs
+
+        content = outputs[0]
+        if not content:
+            return None, outputs
+
+        # This regular expression matches a <think>...</think> block and captures any content between the tags,
+        # including newlines, so that we can separate internal reasoning text from the user-facing answer.
+        think_pattern = r"<think>(.*?)</think>"
+        match = re.search(think_pattern, content, re.DOTALL)
+
+        thinking_content = None
+        if match:
+            thinking_content = match.group(1).strip() or None
+
+        cleaned_content = re.sub(think_pattern, "", content, flags=re.DOTALL).strip()
+        cleaned_outputs = [cleaned_content, *outputs[1:]]
+
+        return thinking_content, cleaned_outputs
+
+    # cumulative wait time is < 30s
+    @retry(tries=8, delay=0.5, backoff=1.5, max_delay=5, jitter=(0, 0.5))
+    def forward(self, argument):
+        kwargs = argument.kwargs
+        messages = argument.prop.prepared_input
+        payload = self._prepare_request_payload(messages, argument)
+        except_remedy = kwargs.get("except_remedy")
+
+        try:
+            res = self.client.chat.completions.create(**payload)
+        except Exception as exc:  # pragma: no cover - defensive path
+            res = self._handle_forward_exception(exc, argument, kwargs, except_remedy)
+
+        return self._build_outputs_and_metadata(res, payload)
+
+    def _handle_forward_exception(self, exc, argument, kwargs, except_remedy):
+        if self.api_key is None or self.api_key == "":
+            msg = (
+                "Cerebras API key is not set. Please set it in the config file or "
+                "pass it as an argument to the command method."
+            )
+            UserMessage(msg)
+            config_key = self.config.get("NEUROSYMBOLIC_ENGINE_API_KEY")
+            if config_key is None or config_key == "":
+                UserMessage(msg, raise_with=ValueError)
+            self.api_key = config_key
+            try:
+                self.client = Cerebras(api_key=self.api_key)
+            except Exception as inner_exc:
+                UserMessage(
+                    f"Failed to initialize Cerebras client after missing API key. Caused by: {inner_exc}",
+                    raise_with=ValueError,
+                )
+
+        callback = self.client.chat.completions.create
+        kwargs["model"] = (
+            self._handle_prefix(kwargs["model"])
+            if "model" in kwargs
+            else self._handle_prefix(self.model)
+        )
+
+        if except_remedy is not None:
+            return except_remedy(self, exc, callback, argument)
+
+        UserMessage(f"Error during generation. Caused by: {exc}", raise_with=ValueError)
+        return None
+
+    def _build_outputs_and_metadata(self, res, payload):
+        metadata: dict = {"raw_output": res}
+        if payload.get("tools"):
+            metadata = self._process_function_calls(res, metadata)
+
+        outputs: list[str] = []
+        thinking_content: str | None = None
+
+        for choice in res.choices:
+            message = choice.message
+            outputs.append(getattr(message, "content", "") or "")
+            if thinking_content is None:
+                reasoning = getattr(message, "reasoning", None)
+                if reasoning:
+                    thinking_content = reasoning
+
+        if thinking_content is None:
+            thinking_content, outputs = self._extract_thinking_content(outputs)
+        else:
+            _, outputs = self._extract_thinking_content(outputs)
+
+        if thinking_content:
+            metadata["thinking"] = thinking_content
+
+        return outputs, metadata
+
+    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,
+            )
+        value = argument.prop.processed_input
+        if not isinstance(value, list):
+            if not isinstance(value, dict):
+                value = {"role": "user", "content": str(value)}
+            value = [value]
+        return value
+
+    def prepare(self, argument):
+        if argument.prop.raw_input:
+            argument.prop.prepared_input = self._prepare_raw_input(argument)
+            return
+        self._validate_response_format(argument)
+
+        system_message = self._build_system_message(argument)
+        user_content = self._build_user_content(argument)
+        user_prompt = {"role": "user", "content": user_content}
+        system_message, user_prompt = self._apply_self_prompt_if_needed(
+            argument, system_message, user_prompt
+        )
+
+        argument.prop.prepared_input = [
+            {"role": "system", "content": system_message},
+            user_prompt,
+        ]
+
+    def _validate_response_format(self, argument) -> None:
+        if argument.prop.response_format:
+            response_format = argument.prop.response_format
+            assert response_format.get("type") is not None, (
+                'Expected format `{ "type": "json_object" }` for JSON mode. '
+                "See Cerebras structured outputs documentation for details."
+            )
+
+    def _build_system_message(self, argument) -> str:
+        system_message: str = ""
+        if argument.prop.suppress_verbose_output:
+            system_message += _NON_VERBOSE_OUTPUT
+        if system_message:
+            system_message = f"{system_message}\n"
+
+        ref = argument.prop.instance
+        static_context, dynamic_context = ref.global_context
+        if len(static_context) > 0:
+            system_message += f"<STATIC CONTEXT/>\n{static_context}\n\n"
+
+        if len(dynamic_context) > 0:
+            system_message += f"<DYNAMIC CONTEXT/>\n{dynamic_context}\n\n"
+
+        if argument.prop.payload:
+            system_message += f"<ADDITIONAL CONTEXT/>\n{argument.prop.payload!s}\n\n"
+
+        examples = argument.prop.examples
+        if examples and len(examples) > 0:
+            system_message += f"<EXAMPLES/>\n{examples!s}\n\n"
+
+        if argument.prop.prompt is not None and len(argument.prop.prompt) > 0:
+            prompt_value = str(argument.prop.prompt)
+            system_message += f"<INSTRUCTION/>\n{prompt_value}\n\n"
+
+        if argument.prop.template_suffix:
+            system_message += (
+                " You will only generate content for the placeholder "
+                f"`{argument.prop.template_suffix!s}` following the instructions and the provided context information.\n\n"
+            )
+
+        return system_message
+
+    def _build_user_content(self, argument) -> str:
+        return str(argument.prop.processed_input)
+
+    def _apply_self_prompt_if_needed(self, argument, system_message, user_prompt):
+        if argument.prop.instance._kwargs.get("self_prompt", False) or argument.prop.self_prompt:
+            self_prompter = SelfPrompt()
+            result = self_prompter({"user": user_prompt["content"], "system": system_message})
+            if result is None:
+                UserMessage("Self-prompting failed!", raise_with=ValueError)
+            return result["system"], {"role": "user", "content": result["user"]}
+        return system_message, user_prompt
+
+    def _process_function_calls(self, res, metadata):
+        hit = False
+        if (
+            hasattr(res, "choices")
+            and res.choices
+            and hasattr(res.choices[0], "message")
+            and res.choices[0].message
+            and hasattr(res.choices[0].message, "tool_calls")
+            and res.choices[0].message.tool_calls
+        ):
+            for tool_call in res.choices[0].message.tool_calls:
+                if hasattr(tool_call, "function") and tool_call.function:
+                    if hit:
+                        UserMessage(
+                            "Multiple function calls detected in the response but only the first one will be processed."
+                        )
+                        break
+                    try:
+                        args_dict = json.loads(tool_call.function.arguments)
+                    except json.JSONDecodeError:
+                        args_dict = {}
+                    metadata["function_call"] = {
+                        "name": tool_call.function.name,
+                        "arguments": args_dict,
+                    }
+                    hit = True
+        return metadata
+
+    def _prepare_request_payload(self, messages, argument):
+        """Prepares the request payload from the argument."""
+        kwargs = argument.kwargs
+
+        n = kwargs.get("n", 1)
+        if n > 1:
+            UserMessage(
+                "If N is supplied, it must be equal to 1. We default to 1 to avoid unexpected batch behavior."
+            )
+            n = 1
+
+        response_format = kwargs.get("response_format")
+
+        return {
+            "messages": messages,
+            "model": self._handle_prefix(kwargs.get("model", self.model)),
+            "max_completion_tokens": kwargs.get("max_completion_tokens"),
+            "stop": kwargs.get("stop"),
+            "temperature": kwargs.get("temperature", 1),
+            "top_p": kwargs.get("top_p", 1),
+            "n": n,
+            "tools": kwargs.get("tools"),
+            "parallel_tool_calls": kwargs.get("parallel_tool_calls"),
+            "response_format": response_format,
+            "reasoning_effort": kwargs.get("reasoning_effort"),
+            "disable_reasoning": kwargs.get("disable_reasoning"),
+            "stream": kwargs.get("stream", False),
+        }

symai/backend/engines/neurosymbolic/engine_deepseekX_reasoning.py

@@ -22,12 +22,12 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
         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 api_key is not None and model is not None:
-            self.config['NEUROSYMBOLIC_ENGINE_API_KEY'] = api_key
-            self.config['NEUROSYMBOLIC_ENGINE_MODEL'] = model
-        if self.id() != 'neurosymbolic':
-            return # do not initialize if not neurosymbolic; avoids conflict with llama.cpp check in EngineRepository.register_from_package
-        self.api_key = self.config['NEUROSYMBOLIC_ENGINE_API_KEY']
-        self.model = self.config['NEUROSYMBOLIC_ENGINE_MODEL']
+            self.config["NEUROSYMBOLIC_ENGINE_API_KEY"] = api_key
+            self.config["NEUROSYMBOLIC_ENGINE_MODEL"] = model
+        if self.id() != "neurosymbolic":
+            return  # do not initialize if not neurosymbolic; avoids conflict with llama.cpp check in EngineRepository.register_from_package
+        self.api_key = self.config["NEUROSYMBOLIC_ENGINE_API_KEY"]
+        self.model = self.config["NEUROSYMBOLIC_ENGINE_MODEL"]
         self.name = self.__class__.__name__
         self.tokenizer = None
         self.max_context_tokens = self.api_max_context_tokens()
@@ -37,71 +37,92 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
         try:
             self.client = OpenAI(api_key=self.api_key, base_url="https://api.deepseek.com")
         except Exception as e:
-            UserMessage(f'Failed to initialize the DeepSeek client. Please check your library version. Caused by: {e}', raise_with=RuntimeError)
+            UserMessage(
+                f"Failed to initialize the DeepSeek client. Please check your library version. Caused by: {e}",
+                raise_with=RuntimeError,
+            )
 
     def id(self) -> str:
-        if self.config.get('NEUROSYMBOLIC_ENGINE_MODEL') and \
-           self.config.get('NEUROSYMBOLIC_ENGINE_MODEL').startswith('deepseek'):
-            return 'neurosymbolic'
-        return super().id() # default to unregistered
+        if self.config.get("NEUROSYMBOLIC_ENGINE_MODEL") and self.config.get(
+            "NEUROSYMBOLIC_ENGINE_MODEL"
+        ).startswith("deepseek"):
+            return "neurosymbolic"
+        return super().id()  # default to unregistered
 
     def command(self, *args, **kwargs):
         super().command(*args, **kwargs)
-        if 'NEUROSYMBOLIC_ENGINE_API_KEY' in kwargs:
-            self.api_key = kwargs['NEUROSYMBOLIC_ENGINE_API_KEY']
-        if 'NEUROSYMBOLIC_ENGINE_MODEL' in kwargs:
-            self.model = kwargs['NEUROSYMBOLIC_ENGINE_MODEL']
-        if 'seed' in kwargs:
-            self.seed = kwargs['seed']
+        if "NEUROSYMBOLIC_ENGINE_API_KEY" in kwargs:
+            self.api_key = kwargs["NEUROSYMBOLIC_ENGINE_API_KEY"]
+        if "NEUROSYMBOLIC_ENGINE_MODEL" in kwargs:
+            self.model = kwargs["NEUROSYMBOLIC_ENGINE_MODEL"]
+        if "seed" in kwargs:
+            self.seed = kwargs["seed"]
 
     def compute_required_tokens(self, _messages):
-        UserMessage('Method "compute_required_tokens" not implemented for DeepSeekXReasoningEngine.', raise_with=NotImplementedError)
+        UserMessage(
+            'Method "compute_required_tokens" not implemented for DeepSeekXReasoningEngine.',
+            raise_with=NotImplementedError,
+        )
 
     def compute_remaining_tokens(self, _prompts: list) -> int:
-        UserMessage('Method "compute_remaining_tokens" not implemented for DeepSeekXReasoningEngine.', raise_with=NotImplementedError)
-
-    def truncate(self, _prompts: list[dict], _truncation_percentage: float | None, _truncation_type: str) -> list[dict]:
-        UserMessage('Method "truncate" not implemented for DeepSeekXReasoningEngine.', raise_with=NotImplementedError)
+        UserMessage(
+            'Method "compute_remaining_tokens" not implemented for DeepSeekXReasoningEngine.',
+            raise_with=NotImplementedError,
+        )
+
+    def truncate(
+        self, _prompts: list[dict], _truncation_percentage: float | None, _truncation_type: str
+    ) -> list[dict]:
+        UserMessage(
+            'Method "truncate" not implemented for DeepSeekXReasoningEngine.',
+            raise_with=NotImplementedError,
+        )
 
     def forward(self, argument):
         kwargs = argument.kwargs
         messages = argument.prop.prepared_input
         payload = self._prepare_request_payload(argument)
-        except_remedy = kwargs.get('except_remedy')
+        except_remedy = kwargs.get("except_remedy")
 
         try:
             res = self.client.chat.completions.create(messages=messages, **payload)
 
         except Exception as e:
-            if self.api_key is None or self.api_key == '':
-                msg = 'DeepSeek API key is not set. Please set it in the config file or pass it as an argument to the command method.'
+            if self.api_key is None or self.api_key == "":
+                msg = "DeepSeek API key is not set. Please set it in the config file or pass it as an argument to the command method."
                 UserMessage(msg)
-                if self.config['NEUROSYMBOLIC_ENGINE_API_KEY'] is None or self.config['NEUROSYMBOLIC_ENGINE_API_KEY'] == '':
+                if (
+                    self.config["NEUROSYMBOLIC_ENGINE_API_KEY"] is None
+                    or self.config["NEUROSYMBOLIC_ENGINE_API_KEY"] == ""
+                ):
                     UserMessage(msg, raise_with=ValueError)
-                self.api_key = self.config['NEUROSYMBOLIC_ENGINE_API_KEY']
+                self.api_key = self.config["NEUROSYMBOLIC_ENGINE_API_KEY"]
 
             callback = self.client.chat.completions.create
-            kwargs['model'] = kwargs.get('model', self.model)
+            kwargs["model"] = kwargs.get("model", self.model)
 
             if except_remedy is not None:
                 res = except_remedy(self, e, callback, argument)
             else:
-                UserMessage(f'Error during generation. Caused by: {e}', raise_with=ValueError)
+                UserMessage(f"Error during generation. Caused by: {e}", raise_with=ValueError)
 
         reasoning_content = res.choices[0].message.reasoning_content
         content = res.choices[0].message.content
-        metadata = {'raw_output': res, 'thinking': reasoning_content}
+        metadata = {"raw_output": res, "thinking": reasoning_content}
 
         return [content], metadata
 
     def _prepare_raw_input(self, argument):
         if not argument.prop.processed_input:
-            UserMessage('A prompt instruction is required for DeepSeekXReasoningEngine when raw_input is enabled.', raise_with=ValueError)
+            UserMessage(
+                "A prompt instruction is required for DeepSeekXReasoningEngine when raw_input is enabled.",
+                raise_with=ValueError,
+            )
         value = argument.prop.processed_input
         # convert to dict if not already
         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
 
@@ -112,14 +133,17 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
 
         if prop.suppress_verbose_output:
             system += _non_verbose_output
-        system = f'{system}\n' if system and len(system) > 0 else ''
+        system = f"{system}\n" if system and len(system) > 0 else ""
 
         if prop.response_format:
             _rsp_fmt = prop.response_format
-            if not (_rsp_fmt.get('type') is not None):
-                UserMessage('Response format type is required! Expected format `{"type": "json_object"}` or other supported types.', raise_with=AssertionError)
+            if not (_rsp_fmt.get("type") is not None):
+                UserMessage(
+                    'Response format type is required! Expected format `{"type": "json_object"}` or other supported types.',
+                    raise_with=AssertionError,
+                )
             system += _non_verbose_output
-            system += f'<RESPONSE_FORMAT/>\n{_rsp_fmt["type"]}\n\n'
+            system += f"<RESPONSE_FORMAT/>\n{_rsp_fmt['type']}\n\n"
 
         ref = prop.instance
         static_ctxt, dyn_ctxt = ref.global_context
@@ -142,7 +166,7 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
             system += f"<INSTRUCTION/>\n{val}\n\n"
 
         if prop.template_suffix:
-            system += f' You will only generate content for the placeholder `{prop.template_suffix!s}` following the instructions and the provided context information.\n\n'
+            system += f" You will only generate content for the placeholder `{prop.template_suffix!s}` following the instructions and the provided context information.\n\n"
 
         return system
 
@@ -151,15 +175,17 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
 
     def _apply_self_prompt(self, argument, system, user_prompt):
         prop = argument.prop
-        if prop.instance._kwargs.get('self_prompt', False) or prop.self_prompt:
+        if prop.instance._kwargs.get("self_prompt", False) or prop.self_prompt:
             self_prompter = SelfPrompt()
 
-            res = self_prompter({'user': user_prompt['content'], 'system': system})
+            res = self_prompter({"user": user_prompt["content"], "system": system})
             if res is None:
-                UserMessage("Self-prompting failed for DeepSeekXReasoningEngine.", raise_with=ValueError)
+                UserMessage(
+                    "Self-prompting failed for DeepSeekXReasoningEngine.", raise_with=ValueError
+                )
 
-            user_prompt = { "role": "user", "content": res['user'] }
-            system = res['system']
+            user_prompt = {"role": "user", "content": res["user"]}
+            system = res["system"]
 
         return system, user_prompt
 
@@ -173,7 +199,7 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
         system, user_prompt = self._apply_self_prompt(argument, system, user_prompt)
 
         argument.prop.prepared_input = [
-            { "role": "system", "content": system },
+            {"role": "system", "content": system},
             user_prompt,
         ]
 
@@ -184,10 +210,10 @@ class DeepSeekXReasoningEngine(Engine, DeepSeekMixin):
         # Not Supported Features: Function Call、Json Output、FIM (Beta)
         # Not Supported Parameters: temperature、top_p、presence_penalty、frequency_penalty、logprobs、top_logprobs
         return {
-            "model": kwargs.get('model', self.model),
-            "seed": kwargs.get('seed', self.seed),
-            "max_tokens": kwargs.get('max_tokens', self.max_response_tokens),
-            "stop": kwargs.get('stop', '<|endoftext|>'),
-            "n": kwargs.get('n', 1),
-            "logit_bias": kwargs.get('logit_bias'),
+            "model": kwargs.get("model", self.model),
+            "seed": kwargs.get("seed", self.seed),
+            "max_tokens": kwargs.get("max_tokens", self.max_response_tokens),
+            "stop": kwargs.get("stop", "<|endoftext|>"),
+            "n": kwargs.get("n", 1),
+            "logit_bias": kwargs.get("logit_bias"),
         }