Create appy.py

appy.py

@@ -0,0 +1,66 @@
+from gradio_client import Client
+from langchain import tru
+from trulens_eval.schema import Select
+from trulens_eval.tru import Tru
+from trulens_eval.feedback import Feedback
+from trulens_eval.feedback import OpenAI as Feedback_OpenAI
+from langchain import HuggingFacePipeline
+from langchain.llms import HuggingFacePipeline
+from langchain.chains import ConversationChain
+from langchain.chains.conversation.memory import ConversationBufferWindowMemory
+import os
+
+# Access environment variables
+openai_api_key = os.environ.get("OPENAI_API_KEY")
+huggingface_api_token = os.environ.get("HUGGINGFACE_API_TOKEN")
+
+
+# Define a feedback function for query-statement relevance using OpenAI
+feedback_openai = Feedback_OpenAI()
+qa_relevance = Feedback(feedback_openai.relevance, name="Answer Relevance").on_input_output()
+
+# Create a Tru object
+tru = Tru()
+
+# Initialize the HuggingFacePipeline for local LLM
+local_llm = HuggingFacePipeline.from_model_id(
+    model_id="chavinlo/alpaca-native",
+    task="text-generation",
+    model_kwargs={"temperature": 0.6, "top_p": 0.95, "max_length": 256}
+)
+
+# Set the window memory to go back 4 turns
+window_memory = ConversationBufferWindowMemory(k=4)
+
+# Create the ConversationChain with the given window memory
+conversation = ConversationChain(
+    llm=local_llm,
+    verbose=True,
+    memory=window_memory
+)
+
+# Update the conversation prompt template to prime it as a gardening expert
+conversation.prompt.template = '''The following is a friendly conversation between a human and an AI gardening expert. The AI is an expert on gardening and gives recommendations specific to location and conditions. If the AI does not know the answer to a question, it truthfully says it does not know.
+
+Current conversation:
+{history}
+Human: {input}
+AI:'''
+
+# Wrap the conversation with TruChain to instrument it
+tc_conversation = tru.Chain(conversation, app_id='GardeningAIwithMemory_v1', feedbacks=[qa_relevance])
+
+# Initialize Gradio Client
+client = Client("https://tonic-stablemed-chat.hf.space/")
+
+# Make a prediction using the wrapped conversation
+result = client.predict(
+    "Howdy!",  # str in 'user_input' Textbox component
+    "Howdy!",  # str in 'system_prompt' Textbox component
+    api_name="/predict"
+)
+
+# Print the result
+print(result)
+
+tru.run_dashboard()