symbolicai 0.20.2__tar.gz → 0.21.0__tar.gz

{symbolicai-0.20.2 → symbolicai-0.21.0}/.gitignore

@@ -164,3 +164,5 @@ symsh
 .symai
 zagent-review
 zagent-research
+.codex/context
+llama.cpp

{symbolicai-0.20.2 → symbolicai-0.21.0}/.symai/symsh.config.json

@@ -12,6 +12,5 @@
         "file-completion-selected": "bg:#efefef #b3d7ff"
     },
     "map-nt-cmd": true,
-    "show-splash-screen": false,
-    "plugin_prefix": null
+    "show-splash-screen": true
 }

{symbolicai-0.20.2 → symbolicai-0.21.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: symbolicai
-Version: 0.20.2
+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-0.20.2 → symbolicai-0.21.0}/docs/source/ENGINES/local_engine.md

@@ -3,16 +3,13 @@
 
 You can use a locally hosted instance for the Neuro-Symbolic Engine. We build on top of:
 - [llama.cpp](https://github.com/ggerganov/llama.cpp/tree/master) either through:
-    > ❗️**NOTE**❗️ Latest `llama.cpp` commit on `master` branch that we tested `symai` with is `0f5ccd6fd1a1`. We used the build [setup](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md).
+    > ❗️**NOTE**❗️ Latest `llama.cpp` commit on `master` branch on November 5th, 2025 that we tested `symai` with is `a5c07dcd7b49`. We used the build [setup](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md).
   - Direct C++ server from llama.cpp
   - [llama-cpp-python](https://github.com/abetlen/llama-cpp-python?tab=readme-ov-file)
 - [huggingface/transformers](https://huggingface.co/docs/transformers/en/index) through a custom FastAPI server.
 
 ### llama.cpp backend
-For instance, let's suppose you want to set as a Neuro-Symbolic Engine the latest Llama 3 model. First, download the model with the HuggingFace CLI:
-```bash
-huggingface-cli download TheBloke/LLaMA-Pro-8B-Instruct-GGUF llama-pro-8b-instruct.Q4_K_M.gguf --local-dir .
-```
+For instance, let's suppose you want to set up the Neuro-Symbolic Engine with the `gpt-oss-120b` model. Download the GGUF shards you need (e.g. the `Q4_1` variant).
 
 With `symai`, first set the `NEUROSYMBOLIC_ENGINE_MODEL` to `llamacpp`:
 
@@ -33,15 +30,17 @@ symserver --env python --model ./llama-pro-8b-instruct.Q4_K_M.gguf --n_gpu_layer
 
 2. Using C++ server directly:
 ```bash
-symserver --env cpp --cpp-server-path /path/to/llama.cpp/server -m ./llama-pro-8b-instruct.Q4_K_M.gguf --port 8000 --host localhost
+symserver --env cpp --cpp-server-path /path/to/llama.cpp/llama-server -ngl -1 -m gpt-oss-120b/Q4_1/gpt-oss-120b-Q4_1-00001-of-00002.gguf -fa 'on' -b 8092 -ub 1024 --port 8000 --host localhost -c 0 -n 4096 -t 14 --jinja
 ```
 
 To see all available options, run:
 ```bash
 symserver --env python --help  # for Python bindings
-symserver --env cpp --cpp-server-path /path/to/llama.cpp/server --help  # for C++ server
+symserver --env cpp --cpp-server-path /path/to/llama.cpp/llama-server --help  # for C++ server
 ```
 
+The Neuro-Symbolic Engine now supports tool execution and structured JSON responses out of the box. For concrete examples, review the tests in `tests/engines/neurosymbolic/test_nesy_engine.py::test_tool_usage` and `tests/contract/test_contract.py`.
+
 ### HuggingFace backend
 Let's suppose we want to use `dolphin-2.9.3-mistral-7B-32k` from HuggingFace. First, download the model with the HuggingFace CLI:
 ```bash
@@ -109,7 +108,7 @@ symserver --env python --model nomic-embed-text-v1.5.Q8_0.gguf --embedding True
 
 C++ server:
 ```bash
-symserver --env cpp --cpp-server-path /path/to/llama.cpp/server  -ngl 0 -m nomic-embed-text-v1.5.Q8_0.gguf --embedding -c 8192 -b 8192 --rope-scaling yarn --rope-freq-scale .75 --port 8000 --host localhost
+symserver --env cpp --cpp-server-path /path/to/llama.cpp/llama-server -ngl -1 -m nomic-embed-text-v1.5.Q8_0.gguf --embedding -b 8092 -ub 1024 --port 8000 --host localhost -t 14 --mlock --no-mmap
 ```
 
 The server supports batch processing for embeddings. Here's how to use it with `symai`:
@@ -123,5 +122,5 @@ embedding = Symbol(some_text).embed()  # returns a list (1 x dim)
 
 # Batch processing
 some_batch_of_texts = ["Hello, world!"] * 32
-embeddings = Symbol(some_batch_of_texts).embed()  # returns a list (32 x dim)
+embeddings = Symbol(some_batch_of_texts).embed()  # returns a list (32 x 1 x dim)
 ```

{symbolicai-0.20.2 → symbolicai-0.21.0}/symai/__init__.py

@@ -33,7 +33,7 @@ os.environ['TOKENIZERS_PARALLELISM'] = "false"
 # Create singleton instance
 config_manager = settings.SymAIConfig()
 
-SYMAI_VERSION = "0.20.2"
+SYMAI_VERSION = "0.21.0"
 __version__   = SYMAI_VERSION
 __root_dir__  = config_manager.config_dir
 

{symbolicai-0.20.2 → symbolicai-0.21.0}/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}

{symbolicai-0.20.2 → symbolicai-0.21.0}/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)

{symbolicai-0.20.2 → symbolicai-0.21.0}/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-0.20.2 → symbolicai-0.21.0}/symbolicai.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: symbolicai
-Version: 0.20.2
+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