update examples

Author: pchalasani

examples/basic/chat-local.py

@@ -22,7 +22,7 @@
 #   or you can explicitly specify it as `lm.OpenAIChatModel.GPT4` or `lm.OpenAIChatModel.GPT4_TURBO`
 
 llm_config = lm.OpenAIGPTConfig(
-    chat_model="litellm/ollama/mistral",
+    chat_model=lm.OpenAIChatModel.GPT4_TURBO,  # or,e.g. "ollama/mistral"
     max_output_tokens=200,
     chat_context_length=2048,  # adjust based on your local LLM params
 )

examples/basic/chat-search.py

@@ -18,7 +18,7 @@
 -m <model_name>: to run with a different LLM model (default: gpt4-turbo)
 
 You can specify a local in a few different ways, e.g. `-m local/localhost:8000/v1`
-or `-m litellm/ollama_chat/mistral` etc. See here how to use Langroid with local LLMs:
+or `-m ollama/mistral` etc. See here how to use Langroid with local LLMs:
 https://langroid.github.io/langroid/tutorials/local-llm-setup/
 
 

examples/basic/chat.py

@@ -16,7 +16,7 @@
 -q <initial user msg>
 
 For details on running with local or non-OpenAI models, see:
-https://langroid.github.io/langroid/tutorials/non-openai-llms/
+https://langroid.github.io/langroid/tutorials/local-llm-setup/
 """
 import typer
 from rich import print
@@ -27,40 +27,12 @@
 from langroid.agent.chat_agent import ChatAgent, ChatAgentConfig
 from langroid.agent.task import Task
 from langroid.utils.configuration import set_global, Settings
-from langroid.utils.logging import setup_colored_logging
 
 
 app = typer.Typer()
 
 # Create classes for non-OpenAI model configs
 
-# OPTION 1: LiteLLM-supported models
-# -----------------------------------
-# For any model supported by litellm
-# (see list here https://docs.litellm.ai/docs/providers)
-# The `chat_model` should be specified as "litellm/" followed by
-# the chat_model name in the litellm docs.
-# In this case Langroid uses the litellm adapter library to
-# translate between OpenAI API and the model's API.
-# For external (remote) models, typical there will be specific env vars
-# (e.g. API Keys, etc) that need to be set.
-# If those are not set, you will get an err msg saying which vars need to be set.
-
-# OPTION 2: Local models served at an OpenAI-compatible API endpoint
-# -----------------------------------------------------------------
-# Use this config for any model that is locally served at an
-# OpenAI-compatible API endpoint, for example, using either the
-# litellm proxy server https://docs.litellm.ai/docs/proxy_server
-# or the oooba/text-generation-webui server
-# https://github.com/oobabooga/text-generation-webui/tree/main/extensions/openai
-#
-# In this case the `chat_model` name should be specified as
-# "local/localhost:8000/v1" or "local/localhost:8000" or other port number
-# depending on how you launch the model locally.
-# Langroid takes care of extracting the local URL to set the `api_base`
-# of the config so that the `openai.*` completion functions can be used
-# without having to rely on adapter libraries like litellm.
-
 
 @app.command()
 def main(

examples/basic/fn-call-local-numerical.py

@@ -1,7 +1,7 @@
 """
 Function-calling example using a local LLM, with ollama.
 
-"Function-calling" refers to the ability to ability of the LLM to generate
+"Function-calling" refers to the ability of the LLM to generate
 a structured response, typically a JSON object, instead of a plain text response,
 which is then interpreted by your code to perform some action.
 This is also referred to in various scenarios as "Tools", "Actions" or "Plugins".
@@ -31,7 +31,7 @@
 Local model with best results is dolphin-mixtral:
 ```
 ollama run dolphin-mixtral
-python3 examples/basic/fn-call-local-numerical.py -m litellm/ollama_chat/dolphin-mixtral:latest
+python3 examples/basic/fn-call-local-numerical.py -m ollama/dolphin-mixtral:latest
 ```
 
 See here for how to set up a Local LLM to work with Langroid:
@@ -43,6 +43,7 @@
 import fire
 
 import langroid as lr
+from langroid.language_models.openai_gpt import OpenAICallParams
 from langroid.utils.configuration import settings
 from langroid.agent.tool_message import ToolMessage
 import langroid.language_models as lm
@@ -57,24 +58,17 @@
 
 
 class PolinskyTool(lr.agent.ToolMessage):
-    """A fictitious number transformation tool. We intentially use
+    """A fictitious number transformation tool. We intentionally use
     a fictitious tool rather than something like "square" or "double"
     to prevent the LLM from trying to answer the question directly.
     """
 
     request: str = "polinsky"
-    purpose: str = "To respond to user request for the Polinsky transform of a <number>."
+    purpose: str = (
+        "To respond to user request for the Polinsky transform of a <number>."
+    )
     number: int
 
-    def handle(self) -> str:
-        """Handle LLM's structured output if it matches Polinsky tool"""
-        result = self.number * 3 + 1
-        msg = f"""
-        SUCCESS! The Polinksy transform of {self.number} is {result}.
-        Present this result to me, and ask for another number.
-        """
-        return msg
-
     @classmethod
     def json_instructions(cls) -> str:
         inst = super().json_instructions()
@@ -84,20 +78,6 @@ def json_instructions(cls) -> str:
         ONLY USE THIS TOOL AFTER THE USER ASKS FOR A POLINSKY TRANSFORM.
         """
 
-    @staticmethod
-    def handle_message_fallback(
-        agent: lr.ChatAgent, msg: str | ChatDocument
-    ) -> str | ChatDocument | None:
-        """Fallback method when LLM forgets to generate a tool"""
-        if isinstance(msg, ChatDocument) and msg.metadata.sender == "LLM":
-            return """
-                You must use the "polinsky" tool/function to 
-                request the Polinsky transform of a number.
-                You either forgot to use it, or you used it with the wrong format.
-                Make sure all fields are filled out and pay attention to the 
-                required types of the fields.
-                """
-
     @classmethod
     def examples(cls) -> List["ToolMessage"]:
         # Used to provide few-shot examples in the system prompt
@@ -111,6 +91,38 @@ def examples(cls) -> List["ToolMessage"]:
         ]
 
 
+class MyChatAgent(lr.ChatAgent):
+    tool_called: bool = False
+
+    def polinsky(self, msg: PolinskyTool) -> str:
+        """Handle LLM's structured output if it matches Polinsky tool"""
+        self.tool_called = True
+        result = msg.number * 3 + 1
+        response = f"""
+        SUCCESS! The Polinksy transform of {msg.number} is {result}.
+        Present this result to me, and ask for another number.
+        """
+        return response
+
+    def handle_message_fallback(
+        self, msg: str | ChatDocument
+    ) -> str | ChatDocument | None:
+        """Fallback method when LLM does not generate a tool,
+        and agent ends up handling the msg"""
+        if isinstance(msg, ChatDocument) and msg.metadata.sender == "LLM":
+            if self.tool_called:
+                self.tool_called = False
+                return "Ask the user what they need help with"
+            else:
+                return """
+                    You must use the "polinsky" tool/function to 
+                    request the Polinsky transform of a number.
+                    You either forgot to use it, or you used it with the wrong format.
+                    Make sure all fields are filled out and pay attention to the 
+                    required types of the fields.
+                    """
+
+
 def app(
     m: str = DEFAULT_LLM,  # model name
     d: bool = False,  # debug
@@ -123,9 +135,13 @@ def app(
         chat_model=m or DEFAULT_LLM,
         chat_context_length=16_000,  # for dolphin-mixtral
         max_output_tokens=100,
+        params=OpenAICallParams(
+            presence_penalty=0.8,
+            frequency_penalty=0.8,
+        ),
         temperature=0,
         stream=True,
-        timeout=45,
+        timeout=100,
     )
 
     # Recommended: First test if basic chat works with this llm setup as below:
@@ -143,7 +159,6 @@ def app(
     # verify you can interact with this in a chat loop on cmd line:
     # task.run("Concisely answer some questions")
 
-
     # Define a ChatAgentConfig and ChatAgent
     config = lr.ChatAgentConfig(
         llm=llm_cfg,
@@ -182,15 +197,15 @@ def app(
         """,
     )
 
-    agent = lr.ChatAgent(config)
+    agent = MyChatAgent(config)
 
     # (4) Enable the Tool for this agent --> this auto-inserts JSON instructions
     # and few-shot examples into the system message
     agent.enable_message(PolinskyTool)
 
     # (5) Create task and run it to start an interactive loop
     task = lr.Task(agent)
-    task.run("ONLY say this: 'What can I help with?' say NOTHING ELSE.")
+    task.run("Can you help me with some questions?")
 
 
 if __name__ == "__main__":

examples/chainlit/chat-doc-qa.py

@@ -9,7 +9,7 @@
 Note, to run this with a local LLM, you can click the settings symbol
 on the left of the chat window and enter the model name, e.g.:
 
-litellm/ollama_chat/mistral:7b-instruct-v0.2-q8_0
+ollama/mistral:7b-instruct-v0.2-q8_0
 
 or