Add shields

README.md

@@ -1,4 +1,10 @@
-# Tweet Critic
+# Online Prompt Optimization
+
+<!-- markdown-link-check-disable -->
+[![GitHub](https://img.shields.io/badge/GitHub-Repository-blue?style=for-the-badge&logo=github)](https://github.com/langchain-ai/tweet-critic)
+
+[![Streamlit App](https://static.streamlit.io/badges/streamlit_badge_black_white.svg)](https://tweet-critic.streamlit.app/)
+<!-- markdown-link-check-enable -->
 
 Welcome to the tweet critic prompt optimization demo! This repository showcases a Streamlit application that demonstrates how to iteratively refine and optimize prompts based on user feedback and dialog.
 
@@ -9,6 +15,8 @@ It employs a two common optimization techniques:
 - Few-shot Learning: approved examples (those given a 👍) are added to a LangSmith few-shot prompt dataset and later dynamically injected in the chat bot's system prompt.
 - Prompt Optimization: When a 👍 or 👎 is provided, an "optimizer" prompt is used to propose updates to the chat bot to help it better satisfy this type of user request in the future. These updates are checked in the the [LangSmith Hub](https://smith.langchain.com/hub) so you can see the updates and roll back prompts if desired.
 
+![Demo](./img/demo.gif)
+
 ## Getting Started
 
 To run the application locally, follow these steps:

main.py

@@ -33,8 +33,6 @@ OPTIMIZER_PROMPT_NAME = "wfh/convo-optimizer"
 NUM_FEWSHOTS = 15
 PROMPT_UPDATE_BATCHSIZE = 5
 
-chat_llm = ChatAnthropic(model="claude-3-haiku-20240307", temperature=1)
-optimizer_llm = ChatAnthropic(model="claude-3-opus-20240229", temperature=1)
 
 # Add a sidebar
 st.sidebar.title("Session Information")
@@ -56,6 +54,12 @@ st.sidebar.markdown(f"[See Prompt in Hub]({prompt_url})")
 optimizer_prompt_url = f"https://smith.langchain.com/hub/{OPTIMIZER_PROMPT_NAME}"
 st.sidebar.markdown(f"[See Optimizer Prompt in Hub]({optimizer_prompt_url})")
 
+chat_llm_model = st.sidebar.selectbox("Chat LLM Model", ["haiku", "opus"])
+if chat_llm_model == "haiku": 
+   chat_llm = ChatAnthropic(model=f"claude-3-{chat_llm_model}-20240307", temperature=1)
+else:
+   chat_llm = ChatAnthropic(model=f"claude-3-{chat_llm_model}-20240229", temperature=1) 
+optimizer_llm = ChatAnthropic(model="claude-3-opus-20240229", temperature=1)
 ## Get few-shot examples from 👍 examples
 client = Client()