langroid 0.20.1__tar.gz → 0.21.0__tar.gz

{langroid-0.20.1 → langroid-0.21.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.20.1
+Version: 0.21.0
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani
@@ -248,6 +248,9 @@ teacher_task.run()
 <details>
 <summary> <b>Click to expand</b></summary>
 
+- **Nov 2024:**
+  - **[0.20.0](https://github.com/langroid/langroid/releases/tag/0.20.0)** Support for 
+    ArangoDB Knowledge Graphs.
 - **Oct 2024:**
   - **[0.18.0]** [LLMConfig.async_stream_quiet](https://langroid.github.io/langroid/notes/async-streaming/) flag to 
     turn off LLM output in async + stream mode.

{langroid-0.20.1 → langroid-0.21.0}/README.md

@@ -135,6 +135,9 @@ teacher_task.run()
 <details>
 <summary> <b>Click to expand</b></summary>
 
+- **Nov 2024:**
+  - **[0.20.0](https://github.com/langroid/langroid/releases/tag/0.20.0)** Support for 
+    ArangoDB Knowledge Graphs.
 - **Oct 2024:**
   - **[0.18.0]** [LLMConfig.async_stream_quiet](https://langroid.github.io/langroid/notes/async-streaming/) flag to 
     turn off LLM output in async + stream mode.

{langroid-0.20.1 → langroid-0.21.0}/langroid/agent/special/arangodb/arangodb_agent.py

@@ -118,6 +118,7 @@ class ArangoChatAgent(ChatAgent):
     def init_state(self) -> None:
         super().init_state()
         self.current_retrieval_aql_query: str = ""
+        self.current_schema_params: ArangoSchemaTool = ArangoSchemaTool()
         self.num_tries = 0  # how many attempts to answer user question
 
     def user_response(
@@ -125,6 +126,8 @@ class ArangoChatAgent(ChatAgent):
         msg: Optional[str | ChatDocument] = None,
     ) -> Optional[ChatDocument]:
         response = super().user_response(msg)
+        if response is None:
+            return None
         response_str = response.content if response is not None else ""
         if response_str != "":
             self.num_tries = 0  # reset number of tries if user responds
@@ -165,7 +168,20 @@ class ArangoChatAgent(ChatAgent):
                 """
             )
 
-        return super().llm_response(message)
+        response = super().llm_response(message)
+        if (
+            response is not None
+            and self.config.chat_mode
+            and self.config.addressing_prefix in response.content
+            and self.has_tool_message_attempt(response)
+        ):
+            # response contains both a user-addressing and a tool, which
+            # is not allowed, so remove the user-addressing prefix
+            response.content = response.content.replace(
+                self.config.addressing_prefix, ""
+            )
+
+        return response
 
     def _validate_config(self) -> None:
         assert isinstance(self.config, ArangoChatAgentConfig)
@@ -363,6 +379,11 @@ class ArangoChatAgent(ChatAgent):
             """
         self.num_tries += 1
         query = msg.aql_query
+        if query == self.current_retrieval_aql_query:
+            return """
+            You have already tried this query, so you will get the same results again!
+            If you need to retry, please MODIFY the query to get different results.
+            """
         self.current_retrieval_aql_query = query
         logger.info(f"Executing AQL query: {query}")
         response = self.read_query(query)
@@ -399,6 +420,17 @@ class ArangoChatAgent(ChatAgent):
         else show all properties and example-docs.
         """
 
+        if (
+            msg is not None
+            and msg.collections == self.current_schema_params.collections
+            and msg.properties == self.current_schema_params.properties
+        ):
+            return """
+            You have already tried this schema TOOL, so you will get the same results 
+            again! Please MODIFY the tool params `collections` or `properties` to get
+            different results.
+            """
+
         if msg is not None:
             collections = msg.collections
             properties = msg.properties

{langroid-0.20.1 → langroid-0.21.0}/langroid/agent/special/arangodb/system_messages.py

@@ -144,14 +144,17 @@ THEN make ANOTHER query, and so on, until you have the answer.
 RETRY-SUGGESTIONS:
 If you receive a null or other unexpected result,
 (a) make sure you use the available TOOLs correctly,
-(b) USE `{arango_schema_tool_name}` tool/function-call to get all collections, 
-    their attributes and graph definitions available in your ArangoDB database.
+(b) learn more about the schema using EITHER:
+ - `{arango_schema_tool_name}` tool/function-call to find properties of specific
+    collections or other parts of the schema, OR
+ - `{aql_retrieval_tool_name}` tool/function-call to use AQL queries to 
+    find specific parts of the schema.
 (c) Collection names are CASE-SENSITIVE -- make sure you adhere to the exact 
     collection name you found in the schema.
 (d) see if you have made an assumption in your AQL query, and try another way, 
     or use `{aql_retrieval_tool_name}` to explore the database contents before 
     submitting your final query. 
-(f) Try APPROXIMATE or PARTIAL MATCHES to strings in the user's query, 
+(e) Try APPROXIMATE or PARTIAL MATCHES to strings in the user's query, 
     e.g. user may ask about "Godfather" instead of "The Godfather",
     or try using CASE-INSENSITIVE MATCHES.
     

{langroid-0.20.1 → langroid-0.21.0}/langroid/language_models/__init__.py

@@ -17,6 +17,7 @@ from .base import (
 from .openai_gpt import (
     OpenAIChatModel,
     AnthropicModel,
+    GeminiModel,
     OpenAICompletionModel,
     OpenAIGPTConfig,
     OpenAIGPT,
@@ -41,6 +42,7 @@ __all__ = [
     "LLMResponse",
     "OpenAIChatModel",
     "AnthropicModel",
+    "GeminiModel",
     "OpenAICompletionModel",
     "OpenAIGPTConfig",
     "OpenAIGPT",

{langroid-0.20.1 → langroid-0.21.0}/langroid/language_models/openai_gpt.py

@@ -66,6 +66,7 @@ if "OLLAMA_HOST" in os.environ:
 else:
     OLLAMA_BASE_URL = "http://localhost:11434/v1"
 
+GEMINI_BASE_URL = "https://generativelanguage.googleapis.com/v1beta/"
 OLLAMA_API_KEY = "ollama"
 DUMMY_API_KEY = "xxx"
 
@@ -92,6 +93,14 @@ class OpenAIChatModel(str, Enum):
     O1_MINI = "o1-mini"
 
 
+class GeminiModel(str, Enum):
+    """Enum for Gemini models"""
+
+    GEMINI_1_5_FLASH = "gemini-1.5-flash"
+    GEMINI_1_5_FLASH_8B = "gemini-1.5-flash-8b"
+    GEMINI_1_5_PRO = "gemini-1.5-pro"
+
+
 class OpenAICompletionModel(str, Enum):
     """Enum for OpenAI Completion models"""
 
@@ -503,6 +512,7 @@ class OpenAIGPT(LanguageModel):
 
         self.is_groq = self.config.chat_model.startswith("groq/")
         self.is_cerebras = self.config.chat_model.startswith("cerebras/")
+        self.is_gemini = self.config.chat_model.startswith("gemini/")
 
         if self.is_groq:
             self.config.chat_model = self.config.chat_model.replace("groq/", "")
@@ -524,6 +534,11 @@ class OpenAIGPT(LanguageModel):
                 api_key=self.api_key,
             )
         else:
+            if self.is_gemini:
+                self.config.chat_model = self.config.chat_model.replace("gemini/", "")
+                self.api_key = os.getenv("GEMINI_API_KEY", DUMMY_API_KEY)
+                self.api_base = GEMINI_BASE_URL
+
             self.client = OpenAI(
                 api_key=self.api_key,
                 base_url=self.api_base,
@@ -623,7 +638,13 @@ class OpenAIGPT(LanguageModel):
         Currently main troublemaker is o1* series.
         """
         match self.config.chat_model:
-            case OpenAIChatModel.O1_MINI | OpenAIChatModel.O1_PREVIEW:
+            case (
+                OpenAIChatModel.O1_MINI
+                | OpenAIChatModel.O1_PREVIEW
+                | GeminiModel.GEMINI_1_5_FLASH
+                | GeminiModel.GEMINI_1_5_FLASH_8B
+                | GeminiModel.GEMINI_1_5_PRO
+            ):
                 return {"max_tokens": "max_completion_tokens"}
             case _:
                 return {}
@@ -1229,13 +1250,10 @@ class OpenAIGPT(LanguageModel):
         cost = 0.0
         prompt_tokens = 0
         completion_tokens = 0
-        if not cached and not self.get_stream():
+        if not cached and not self.get_stream() and response["usage"] is not None:
             prompt_tokens = response["usage"]["prompt_tokens"]
             completion_tokens = response["usage"]["completion_tokens"]
-            cost = self._cost_chat_model(
-                response["usage"]["prompt_tokens"],
-                response["usage"]["completion_tokens"],
-            )
+            cost = self._cost_chat_model(prompt_tokens, completion_tokens)
 
         return LLMTokenUsage(
             prompt_tokens=prompt_tokens, completion_tokens=completion_tokens, cost=cost

{langroid-0.20.1 → langroid-0.21.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.20.1"
+version = "0.21.0"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"