symbolicai 0.20.1__py3-none-any.whl → 0.21.0__py3-none-any.whl

symai/__init__.py

@@ -33,7 +33,7 @@ os.environ['TOKENIZERS_PARALLELISM'] = "false"
 # Create singleton instance
 config_manager = settings.SymAIConfig()
 
-SYMAI_VERSION = "0.20.1"
+SYMAI_VERSION = "0.21.0"
 __version__   = SYMAI_VERSION
 __root_dir__  = config_manager.config_dir
 

symai/backend/engines/embedding/engine_llama_cpp.py

@@ -84,7 +84,7 @@ class LlamaCppEmbeddingEngine(Engine):
             asyncio.set_event_loop(new_loop)
             return new_loop
 
-    async def _arequest(self, text: str) -> dict:
+    async def _arequest(self, text: str, embd_normalize: str) -> dict:
         """Makes an async HTTP request to the llama.cpp server."""
         @retry(**self.retry_params)
         async def _make_request():
@@ -95,7 +95,7 @@ class LlamaCppEmbeddingEngine(Engine):
             async with aiohttp.ClientSession(timeout=timeout) as session:
                 async with session.post(
                     f"{self.server_endpoint}/v1/embeddings",
-                    json={"input": text}
+                    json={"content": text, "embd_normalize": embd_normalize}
                 ) as res:
                     if res.status != 200:
                         raise ValueError(f"Request failed with status code: {res.status}")
@@ -108,21 +108,22 @@ class LlamaCppEmbeddingEngine(Engine):
         kwargs = argument.kwargs
 
         inp = prepared_input if isinstance(prepared_input, list) else [prepared_input]
+        embd_normalize = kwargs.get('embd_normalize', -1) # -1 = no normalization
+
         new_dim = kwargs.get('new_dim')
+        if new_dim:
+            raise NotImplementedError("new_dim is not yet supported")
 
         nest_asyncio.apply()
         loop = self._get_event_loop()
 
         try:
-            res = loop.run_until_complete(self._arequest(inp))
+            res = loop.run_until_complete(self._arequest(inp, embd_normalize))
         except Exception as e:
             raise ValueError(f"Request failed with error: {str(e)}")
 
-        if new_dim:
-            raise NotImplementedError("new_dim is not yet supported")
-
         if res is not None:
-            output = [r["embedding"] for r in res["data"]]
+            output = [r["embedding"] for r in res] # B x 1 x D
         else:
             output = None
         metadata = {'raw_output': res}

symai/backend/engines/neurosymbolic/engine_llama_cpp.py

@@ -26,8 +26,8 @@ class LlamaCppTokenizer:
     @staticmethod
     async def _encode(text: str) -> list[int]:
         async with aiohttp.ClientSession() as session:
-            async with session.post(f"{LlamaCppTokenizer._server_endpoint}/extras/tokenize", json={
-                "input": text,
+            async with session.post(f"{LlamaCppTokenizer._server_endpoint}/tokenize", json={
+                "content": text,
             }) as res:
                 if res.status != 200:
                     CustomUserWarning(f"Request failed with status code: {res.status}", raise_with=ValueError)
@@ -46,13 +46,13 @@ class LlamaCppTokenizer:
     @staticmethod
     async def _decode(tokens: list[int]) -> str:
         async with aiohttp.ClientSession() as session:
-            async with session.post(f"{LlamaCppTokenizer._server_endpoint}/extras/detokenize", json={
+            async with session.post(f"{LlamaCppTokenizer._server_endpoint}/detokenize", json={
                 "tokens": tokens,
             }) as res:
                 if res.status != 200:
                     CustomUserWarning(f"Request failed with status code: {res.status}", raise_with=ValueError)
                 res = await res.json()
-                return res['text']
+                return res['content']
 
     @staticmethod
     def decode(tokens: list[int]) -> str:
@@ -148,7 +148,7 @@ class LlamaCppEngine(Engine):
     def _prepare_request_payload(self, argument: Argument) -> dict:
         """Prepares the request payload from the argument."""
         kwargs = argument.kwargs
-        return {
+        payload = {
             "messages": argument.prop.prepared_input,
             "temperature": kwargs.get('temperature', 0.6),
             "frequency_penalty": kwargs.get('frequency_penalty', 0),
@@ -162,12 +162,28 @@ class LlamaCppEngine(Engine):
             "repeat_penalty": kwargs.get('repeat_penalty', 1),
             "logits_bias": kwargs.get('logits_bias'),
             "logprobs": kwargs.get('logprobs', False),
-            "functions": kwargs.get('functions'),
-            "function_call": kwargs.get('function_call'),
             "grammar": kwargs.get('grammar'),
             "response_format": kwargs.get('response_format'),
         }
 
+        model = SYMSERVER_CONFIG.get('-m') or SYMSERVER_CONFIG.get('--model')
+        if model:
+            payload["model"] = model
+
+        tools = kwargs.get('tools')
+        if tools:
+            payload["tools"] = tools
+
+        tool_choice = kwargs.get('tool_choice')
+        if tool_choice is not None:
+            payload["tool_choice"] = tool_choice
+
+        extra_body = kwargs.get('extra_body')
+        if isinstance(extra_body, dict):
+            payload.update(extra_body)
+
+        return payload
+
     async def _arequest(self, payload: dict) -> dict:
         """Makes an async HTTP request to the llama.cpp server."""
         @retry(**self.retry_params)
@@ -187,6 +203,19 @@ class LlamaCppEngine(Engine):
 
         return await _make_request()
 
+    @classmethod
+    def _extract_thinking(cls, response):
+        """Extract reasoning traces from llama.cpp responses."""
+        if not isinstance(response, dict):
+            return None
+        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')
+        return None
+
     def forward(self, argument):
         payload = self._prepare_request_payload(argument)
 
@@ -200,11 +229,53 @@ class LlamaCppEngine(Engine):
 
         metadata = {'raw_output': res}
 
+        if payload.get('tools'):
+            metadata = self._process_tool_calls(res, metadata)
+
+        thinking = self._extract_thinking(res)
+        if thinking:
+            metadata['thinking'] = thinking
+
         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
+        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 []
+            if not tool_calls:
+                continue
+            for tool_call in tool_calls:
+                if not isinstance(tool_call, dict):
+                    continue
+                function = tool_call.get('function') or {}
+                if hit:
+                    CustomUserWarning("Multiple function calls detected in the response but only the first one will be processed.")
+                    return metadata
+                arguments = function.get('arguments')
+                try:
+                    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 {}
+                }
+                hit = True
+                break
+            if hit:
+                break
+        return metadata
+
     def _prepare_raw_input(self, argument):
         if not argument.prop.processed_input:
             CustomUserWarning('Need to provide a prompt instruction to the engine if raw_input is enabled.', raise_with=ValueError)

symai/backend/engines/search/engine_openai.py

@@ -334,6 +334,7 @@ class GPTXSearchEngine(Engine):
             }
 
         self.model = kwargs.get('model', self.model) # Important for MetadataTracker to work correctly
+
         payload = {
             "model": self.model,
             "input": messages,
@@ -341,6 +342,10 @@ class GPTXSearchEngine(Engine):
             "tool_choice": {"type": "web_search"} if self.model not in OPENAI_REASONING_MODELS else "auto" # force the use of web search tool for non-reasoning models
         }
 
+        if self.model in OPENAI_REASONING_MODELS:
+            reasoning = kwargs.get('reasoning', { "effort": "low", "summary": "auto" })
+            payload['reasoning'] = reasoning
+
         try:
             res = self.client.responses.create(**payload)
             res = SearchResult(res.dict())

symai/functional.py

@@ -171,7 +171,10 @@ def _execute_query_fallback(func, instance, argument, error=None, stack_trace=No
     This matches the fallback logic used in _process_query by handling errors consistently,
     providing error context to the fallback function, and maintaining the same return format.
     """
-    rsp = func(instance, error=error, stack_trace=stack_trace, *argument.args, **argument.signature_kwargs)
+    try:
+        rsp = func(instance, error=error, stack_trace=stack_trace, *argument.args, **argument.signature_kwargs)
+    except Exception:
+        raise error # re-raise the original error
     if rsp is not None:
         # fallback was implemented
         rsp = dict(data=rsp, error=error, stack_trace=stack_trace)
@@ -242,7 +245,7 @@ def _process_query(
         func: Callable,
         constraints: List[Callable] = [],
         default: Optional[object] = None,
-        limit: int = 1,
+        limit: int | None = None,
         trials: int = 1,
         pre_processors: Optional[List[PreProcessor]] = None,
         post_processors: Optional[List[PostProcessor]] = None,

{symbolicai-0.20.1.dist-info → symbolicai-0.21.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: symbolicai
-Version: 0.20.1
+Version: 0.21.0
 Summary: A Neurosymbolic Perspective on Large Language Models
 Author-email: Marius-Constantin Dinu <marius@extensity.ai>, Leoveanu-Condrei Claudiu <leo@extensity.ai>
 License: BSD 3-Clause License

{symbolicai-0.20.1.dist-info → symbolicai-0.21.0.dist-info}/RECORD

@@ -1,5 +1,5 @@
 symai/TERMS_OF_SERVICE.md,sha256=HN42UXVI_wAVDHjMShzy_k7xAsbjXaATNeMKcIte_eg,91409
-symai/__init__.py,sha256=fsXUaTEBN_pJJ24Fh59ZPC39HPMGEH_w7Qr7Llwpyhk,16464
+symai/__init__.py,sha256=QrAr-FX1eF1p2a0kDDY-JUcpZgoJpvdGmpdvKJD7WQA,16464
 symai/chat.py,sha256=vqEe7NqSWdzr9ixkko_094SR1LIbgPLcZxQ8W7782N4,12775
 symai/components.py,sha256=vgIq-cC8rqZG9PAPUB52Y5RGFEKrxFUCWzqzrPzLNvw,52232
 symai/constraints.py,sha256=S1ywLB8nFQy4-beDoJz6IvLTiZHGR8Fu5RNTY4v5zG0,1641
@@ -7,7 +7,7 @@ symai/context.py,sha256=4M69MJOeWSdPTr2Y9teoNTs-nEvpzcAcr7900UgORXA,189
 symai/core.py,sha256=1g45AjJ5wkz1cNTbtoDbd8QlOUc-v-3sWNmDTxaeqY0,69041
 symai/core_ext.py,sha256=binru2AjB8K-arbNLiu1wnNodtFxgqk26b-iLVhPoSU,9322
 symai/exceptions.py,sha256=BxpxI8q3-7Uh_Kg9Xi2PhF6RR6CofxV1h8R07j4v47U,165
-symai/functional.py,sha256=ftuo3RB34Al4kMAaqAt9viOd84Tm1mH_svNUSLjnztU,18868
+symai/functional.py,sha256=0N3f5aQ8TjZ82k7GRl0X6sXVpfplQUNU3RFky3EgKho,18963
 symai/imports.py,sha256=npVGz9portPb3enIUtLYiwtdhSPVz6ctHLPreS6Jtvo,16354
 symai/interfaces.py,sha256=MwnRvd-0QrMc2t_MLv05ZZASeWEIgbWxadiVOGbPbOQ,2898
 symai/memory.py,sha256=Svie0ozSMOElMzcyAGnShc3VOQYpkiCEN4ZLoc2ofHM,3674
@@ -30,7 +30,7 @@ symai/backend/engines/crawler/engine_selenium.py,sha256=daCEXYh7WSlwbJypM-51rrOm
 symai/backend/engines/drawing/engine_bfl.py,sha256=dnw-V9DVrulYlhzwPF2Tu-2wfLBEJVvYt9CGq3SQnqA,4478
 symai/backend/engines/drawing/engine_dall_e.py,sha256=CKcMfy69kXSs2I92PqNfgRoz5LZYH-1nMhEVIRZKLFk,4896
 symai/backend/engines/drawing/engine_gpt_image.py,sha256=0rooYI-NrbpUeCRMzYMHPNzOCpr_-j7ShWzJ7_dZ9MU,7502
-symai/backend/engines/embedding/engine_llama_cpp.py,sha256=gfhvNTjHd2uJCBqBc_CRVBFkHeu3IJGE90tNeTwa8uQ,5023
+symai/backend/engines/embedding/engine_llama_cpp.py,sha256=H9rL-ZErvVU5W7nJTXrCSpudjhgQRONoQUNbjC-sA_Y,5182
 symai/backend/engines/embedding/engine_openai.py,sha256=0Tw45QvRbaH7jpOimx_e3IcvvJ50bv0Au8lxMCj5Hvo,3322
 symai/backend/engines/embedding/engine_plugin_embeddings.py,sha256=TLOaYAw_1AWMwqcHydP41aWIYSGhguOJgI0rMu1Fl00,623
 symai/backend/engines/execute/engine_python.py,sha256=3rPMBcrCrjggVm7JtGQjAywBZ6C_3TXXNwmKBjGFEqE,3752
@@ -50,13 +50,13 @@ symai/backend/engines/neurosymbolic/engine_deepseekX_reasoning.py,sha256=ZaZvAVp
 symai/backend/engines/neurosymbolic/engine_google_geminiX_reasoning.py,sha256=kxGWYsxnQkpsm40HB1MUGggWmrWz8avCY3jvNrohaKw,25622
 symai/backend/engines/neurosymbolic/engine_groq.py,sha256=CsPd2TOHgd-gAIRJvFbTTy_otUvMdDxJPH7DsfZAooM,12176
 symai/backend/engines/neurosymbolic/engine_huggingface.py,sha256=XIu9BnQo-J2flXFCCKwCJJmVozU9WDNkPndmpi-DlzE,7920
-symai/backend/engines/neurosymbolic/engine_llama_cpp.py,sha256=ZbHaRYOb7QSD0OrFURp4PT0KJgQk3gdJMD_Itxn7QcU,11323
+symai/backend/engines/neurosymbolic/engine_llama_cpp.py,sha256=DnF_f3nB5TYpHPAlwVJwi-Ax2L9JR978QGHCXv8ts7k,13833
 symai/backend/engines/neurosymbolic/engine_openai_gptX_chat.py,sha256=Y-auxUFC4W9dfRzzgI3_rbWbPiOx4xfvKS4sM0KxP40,25250
 symai/backend/engines/neurosymbolic/engine_openai_gptX_completion.py,sha256=YgxRoitmDz2de_W7rkhVXYEkDqTJQlgxK4f8tWlt88Q,13840
 symai/backend/engines/neurosymbolic/engine_openai_gptX_reasoning.py,sha256=QVbyZybUPSAQHiA66V6we2W2dAsk52g1kJ7kMdGqb9I,22951
 symai/backend/engines/ocr/engine_apilayer.py,sha256=hZo4lk0ECRIzaGEpmCSNjR5Xrh8mwkKMD2ddpdgioVU,2399
 symai/backend/engines/output/engine_stdout.py,sha256=2hhyhMHFJTfjVRaODYd_5XPnV9pT03URcpYbeMY_USU,951
-symai/backend/engines/search/engine_openai.py,sha256=s0ZlcIRDGrzm24M9uhPyztF1w02An2tt3XcoGaGx6RQ,13951
+symai/backend/engines/search/engine_openai.py,sha256=ps_BNwiC1K1NeBJ6oaoyRIDiYksOqi53J7gGKaobue8,14136
 symai/backend/engines/search/engine_perplexity.py,sha256=yxuhGaA38d1FRbLv6piLll0QDxCCyBVK6eeomjYNryM,4157
 symai/backend/engines/search/engine_serpapi.py,sha256=UqvGHs1J9BOv05C0FJUQjbz29_VuWncIkeDwlRPUilU,3698
 symai/backend/engines/speech_to_text/engine_local_whisper.py,sha256=LRsXliCpHDFPFaE-vPky3-DLkmYwmwe2mxfF0Brz4Wg,8220
@@ -154,9 +154,9 @@ symai/ops/primitives.py,sha256=EaB2Ekx9yGNDaQa3aKS5KpuEr5awAUbO3OcBbufI-l4,11072
 symai/server/__init__.py,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
 symai/server/huggingface_server.py,sha256=UpSBflnQaenDjY1AAn5LUYeg5J4gJLWiMuC5DcoIV3E,8743
 symai/server/llama_cpp_server.py,sha256=qVCldTdcQhK2YCu7sDNSYziu1p2AQieqMFfY028-yOc,2049
-symbolicai-0.20.1.dist-info/licenses/LICENSE,sha256=9vRFudlJ1ghVfra5lcCUIYQCqnZSYcBLjLHbGRsrQCs,1505
-symbolicai-0.20.1.dist-info/METADATA,sha256=Lg6BX-_4lfl9afWX09E9uuvyenffX8Y4Qwkb5zh_ZKc,23122
-symbolicai-0.20.1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-symbolicai-0.20.1.dist-info/entry_points.txt,sha256=JV5sdydIfUZdDF6QBEQHiZHod6XNPjCjpWQrXh7gTAw,261
-symbolicai-0.20.1.dist-info/top_level.txt,sha256=bOoIDfpDIvCQtQgXcwVKJvxAKwsxpxo2IL4z92rNJjw,6
-symbolicai-0.20.1.dist-info/RECORD,,
+symbolicai-0.21.0.dist-info/licenses/LICENSE,sha256=9vRFudlJ1ghVfra5lcCUIYQCqnZSYcBLjLHbGRsrQCs,1505
+symbolicai-0.21.0.dist-info/METADATA,sha256=DvZzrNpr8abyJkWHr2GBW1oyM6WgZafUVoUfJzZBX28,23122
+symbolicai-0.21.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+symbolicai-0.21.0.dist-info/entry_points.txt,sha256=JV5sdydIfUZdDF6QBEQHiZHod6XNPjCjpWQrXh7gTAw,261
+symbolicai-0.21.0.dist-info/top_level.txt,sha256=bOoIDfpDIvCQtQgXcwVKJvxAKwsxpxo2IL4z92rNJjw,6
+symbolicai-0.21.0.dist-info/RECORD,,