langroid 0.39.2__tar.gz → 0.39.4__tar.gz

{langroid-0.39.2 → langroid-0.39.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: langroid
-Version: 0.39.2
+Version: 0.39.4
 Summary: Harness LLMs with Multi-Agent Programming
 Author-email: Prasad Chalasani <pchalasani@gmail.com>
 License: MIT
@@ -30,7 +30,7 @@ Requires-Dist: lxml<5.0.0,>=4.9.3
 Requires-Dist: nest-asyncio<2.0.0,>=1.6.0
 Requires-Dist: nltk<4.0.0,>=3.8.2
 Requires-Dist: onnxruntime<2.0.0,>=1.16.1
-Requires-Dist: openai<2.0.0,>=1.45.0
+Requires-Dist: openai<2.0.0,>=1.61.1
 Requires-Dist: pandas<3.0.0,>=2.0.3
 Requires-Dist: prettytable<4.0.0,>=3.8.0
 Requires-Dist: pydantic<3.0.0,>=1

{langroid-0.39.2 → langroid-0.39.4}/langroid/agent/special/doc_chat_agent.py

@@ -58,7 +58,7 @@ from langroid.utils.object_registry import ObjectRegistry
 from langroid.utils.output import show_if_debug, status
 from langroid.utils.output.citations import (
     extract_markdown_references,
-    format_footnote_text,
+    format_cited_references,
 )
 from langroid.utils.pydantic_utils import dataframe_to_documents, extract_fields
 from langroid.vector_store.base import VectorStore, VectorStoreConfig
@@ -859,18 +859,10 @@ class DocChatAgent(ChatAgent):
         final_answer = answer_doc.content.strip()
         show_if_debug(final_answer, "SUMMARIZE_RESPONSE= ")
 
+        # extract references like [^2], [^3], etc. from the final answer
         citations = extract_markdown_references(final_answer)
-
-        citations_str = ""
-        if len(citations) > 0:
-            # append [i] source, content for each citation
-            citations_str = "\n".join(
-                [
-                    f"[^{c}] {passages[c-1].metadata.source}"
-                    f"\n{format_footnote_text(passages[c-1].content)}"
-                    for c in citations
-                ]
-            )
+        # format the cited references as a string suitable for markdown footnote
+        citations_str = format_cited_references(citations, passages)
 
         return ChatDocument(
             content=final_answer,  # does not contain citations

{langroid-0.39.2 → langroid-0.39.4}/langroid/language_models/model_info.py

@@ -54,6 +54,7 @@ class DeepSeekModel(ModelName):
 
     DEEPSEEK = "deepseek/deepseek-chat"
     DEEPSEEK_R1 = "deepseek/deepseek-reasoner"
+    OPENROUTER_DEEPSEEK_R1 = "openrouter/deepseek/deepseek-r1"
 
 
 class GeminiModel(ModelName):
@@ -66,6 +67,23 @@ class GeminiModel(ModelName):
     GEMINI_2_FLASH_THINKING = "gemini/gemini-2.0-flash-thinking-exp"
 
 
+class OpenAI_API_ParamInfo(BaseModel):
+    """
+    Parameters exclusive to some models, when using OpenAI API
+    """
+
+    params: Dict[str, List[str]] = dict(
+        reasoning_effort=[
+            OpenAIChatModel.O3_MINI.value,
+        ],
+    )
+    extra_parameters: Dict[str, List[str]] = dict(
+        include_reasoning=[
+            DeepSeekModel.OPENROUTER_DEEPSEEK_R1.value,
+        ]
+    )
+
+
 class ModelInfo(BaseModel):
     """
     Consolidated information about LLM, related to capacity, cost and API

{langroid-0.39.2 → langroid-0.39.4}/langroid/language_models/openai_gpt.py

@@ -49,6 +49,7 @@ from langroid.language_models.config import HFPromptFormatterConfig
 from langroid.language_models.model_info import (
     DeepSeekModel,
     GeminiModel,
+    OpenAI_API_ParamInfo,
     get_model_info,
 )
 from langroid.language_models.model_info import (
@@ -203,6 +204,7 @@ class OpenAICallParams(BaseModel):
     stop: str | List[str] | None = None  # (list of) stop sequence(s)
     seed: int | None = 42
     user: str | None = None  # user id for tracking
+    extra_body: Dict[str, Any] | None = None  # additional params for API request body
 
     def to_dict_exclude_none(self) -> Dict[str, Any]:
         return {k: v for k, v in self.dict().items() if v is not None}
@@ -642,7 +644,15 @@ class OpenAIGPT(LanguageModel):
         """
         List of params that are not supported by the current model
         """
-        return get_model_info(self.config.chat_model).unsupported_params
+        model_info = get_model_info(self.config.chat_model)
+        unsupported = set(model_info.unsupported_params)
+        for param, model_list in OpenAI_API_ParamInfo().params.items():
+            if (
+                self.config.chat_model not in model_list
+                and self.chat_model_orig not in model_list
+            ):
+                unsupported.add(param)
+        return list(unsupported)
 
     def rename_params(self) -> Dict[str, str]:
         """
@@ -739,7 +749,10 @@ class OpenAIGPT(LanguageModel):
             delta = choices[0].get("delta", {})
             # capture both content and reasoning_content
             event_text = delta.get("content", "")
-            event_reasoning = delta.get("reasoning_content", "")
+            event_reasoning = delta.get(
+                "reasoning_content",
+                delta.get("reasoning", ""),
+            )
             if "function_call" in delta and delta["function_call"] is not None:
                 if "name" in delta["function_call"]:
                     event_fn_name = delta["function_call"]["name"]
@@ -861,7 +874,10 @@ class OpenAIGPT(LanguageModel):
         if chat:
             delta = choices[0].get("delta", {})
             event_text = delta.get("content", "")
-            event_reasoning = delta.get("reasoning_content", "")
+            event_reasoning = delta.get(
+                "reasoning_content",
+                delta.get("reasoning", ""),
+            )
             if "function_call" in delta and delta["function_call"] is not None:
                 if "name" in delta["function_call"]:
                     event_fn_name = delta["function_call"]["name"]
@@ -1757,6 +1773,18 @@ class OpenAIGPT(LanguageModel):
         for old_param, new_param in param_rename_map.items():
             if old_param in args:
                 args[new_param] = args.pop(old_param)
+
+        # finally, get rid of extra_body params exclusive to certain models
+        extra_params = args.get("extra_body", {})
+        if extra_params:
+            for param, model_list in OpenAI_API_ParamInfo().extra_parameters.items():
+                if (
+                    self.config.chat_model not in model_list
+                    and self.chat_model_orig not in model_list
+                ):
+                    extra_params.pop(param, None)
+            if extra_params:
+                args["extra_body"] = extra_params
         return args
 
     def _process_chat_completion_response(

{langroid-0.39.2 → langroid-0.39.4}/langroid/parsing/urls.py

@@ -221,10 +221,15 @@ def find_urls(
         response = requests.get(url, timeout=5)
         response.raise_for_status()
         soup = BeautifulSoup(response.text, "html.parser")
-        links = [urljoin(url, a["href"]) for a in soup.find_all("a", href=True)]
+        links = [
+            urljoin(url, a["href"])  # type: ignore
+            for a in soup.find_all("a", href=True)
+        ]
 
         # Defrag links: discard links that are to portions of same page
-        defragged_links = list(set(urldefrag(link).url for link in links))
+        defragged_links = list(
+            set(urldefrag(link).url for link in links)  # type: ignore
+        )
 
         # Filter links based on domain matching requirement
         domain_matching_links = [

{langroid-0.39.2 → langroid-0.39.4}/langroid/utils/output/citations.py

@@ -1,4 +1,9 @@
-def extract_markdown_references(md_string: str) -> list[int]:
+from typing import List
+
+from langroid.mytypes import Document
+
+
+def extract_markdown_references(md_string: str) -> List[int]:
     """
     Extracts markdown references (e.g., [^1], [^2]) from a string and returns
     them as a sorted list of integers.
@@ -59,3 +64,28 @@ def format_footnote_text(content: str, width: int = 0) -> str:
 
     # Join them with newline so we preserve the paragraph/blank line structure
     return "\n".join(output_lines)
+
+
+def format_cited_references(citations: List[int], passages: list[Document]) -> str:
+    """
+    Given a list of (integer) citations, and a list of passages, return a string
+    that can be added as a footer to the main text, to show sources cited.
+
+    Args:
+        citations (list[int]): list of citations, presumably from main text
+        passages (list[Document]): list of passages (Document objects)
+
+    Returns:
+        str: formatted string of citations for footnote in markdown
+    """
+    citations_str = ""
+    if len(citations) > 0:
+        # append [i] source, content for each citation
+        citations_str = "\n".join(
+            [
+                f"[^{c}] {passages[c-1].metadata.source}"
+                f"\n{format_footnote_text(passages[c-1].content)}"
+                for c in citations
+            ]
+        )
+    return citations_str

{langroid-0.39.2 → langroid-0.39.4}/langroid/utils/system.py

@@ -14,11 +14,7 @@ from typing import Any, Literal
 
 logger = logging.getLogger(__name__)
 
-DELETION_ALLOWED_PATHS = [
-    ".qdrant",
-    ".chroma",
-    ".lancedb",
-]
+DELETION_ALLOWED_PATHS = [".qdrant", ".chroma", ".lancedb", ".weaviate"]
 
 
 def pydantic_major_version() -> int:

{langroid-0.39.2 → langroid-0.39.4}/langroid/vector_store/weaviatedb.py

@@ -32,6 +32,8 @@ class WeaviateDBConfig(VectorStoreConfig):
     collection_name: str | None = "temp"
     embedding: EmbeddingModelsConfig = OpenAIEmbeddingsConfig()
     distance: str = VectorDistances.COSINE
+    cloud: bool = False
+    storage_path: str = ".weaviate_embedded/data"
 
 
 class WeaviateDB(VectorStore):
@@ -39,19 +41,25 @@ class WeaviateDB(VectorStore):
         super().__init__(config)
         self.config: WeaviateDBConfig = config
         load_dotenv()
-        key = os.getenv("WEAVIATE_API_KEY")
-        url = os.getenv("WEAVIATE_API_URL")
-        if url is None or key is None:
-            raise ValueError(
-                """WEAVIATE_API_KEY, WEAVIATE_API_URL env variable must be set to use
-                WeaviateDB in cloud mode. Please set these values
-                in your .env file.
-                """
+        if not self.config.cloud:
+            self.client = weaviate.connect_to_embedded(
+                version="latest", persistence_data_path=self.config.storage_path
             )
-        self.client = weaviate.connect_to_weaviate_cloud(
-            cluster_url=url,
-            auth_credentials=Auth.api_key(key),
-        )
+        else:  # Cloud mode
+            key = os.getenv("WEAVIATE_API_KEY")
+            url = os.getenv("WEAVIATE_API_URL")
+            if url is None or key is None:
+                raise ValueError(
+                    """WEAVIATE_API_KEY, WEAVIATE_API_URL env variables must be set to 
+                    use WeaviateDB in cloud mode. Please set these values
+                    in your .env file.
+                    """
+                )
+            self.client = weaviate.connect_to_weaviate_cloud(
+                cluster_url=url,
+                auth_credentials=Auth.api_key(key),
+            )
+
         if config.collection_name is not None:
             WeaviateDB.validate_and_format_collection_name(config.collection_name)
 
@@ -267,3 +275,8 @@ class WeaviateDB(VectorStore):
             )
 
         return formatted_name
+
+    def __del__(self) -> None:
+        # Gracefully close the connection with local client
+        if not self.config.cloud:
+            self.client.close()

{langroid-0.39.2 → langroid-0.39.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langroid"
-version = "0.39.2"
+version = "0.39.4"
 authors = [
     {name = "Prasad Chalasani", email = "pchalasani@gmail.com"},
 ]
@@ -33,7 +33,7 @@ dependencies = [
     "nest-asyncio<2.0.0,>=1.6.0",
     "nltk<4.0.0,>=3.8.2",
     "onnxruntime<2.0.0,>=1.16.1",
-    "openai<2.0.0,>=1.45.0",
+    "openai<2.0.0,>=1.61.1",
     "pandas<3.0.0,>=2.0.3",
     "prettytable<4.0.0,>=3.8.0",
     "pydantic<3.0.0,>=1",