updating UI examples

examples/UI/rag_ui_with_query_topic_with_streamlit.py

@@ -0,0 +1,94 @@
+
+""" This example shows how to build a simple UI RAG application for longer documents in which a retrieval query step
+    is required to build a context from selected text chunks in the document.
+
+    This example is build with a Streamlit UI. To run, it requires a separate `pip install streamlit`, and
+    to execute the script, you should run from the command line with:
+
+     `streamlit run using_with_streamlit_ui.py`
+
+    For more information about Streamlit, check out their docs:  https://docs.streamlit.io/develop/tutorials
+
+    To build out the application, you would replace the very simple 'text search' mechanism used below with
+    techniques outlined in examples in Embeddings and Retrieval.
+
+"""
+
+
+import os
+import streamlit as st
+
+from llmware.prompts import Prompt
+from llmware.setup import Setup
+
+# st.set_page_config(layout="wide")
+
+
+def simple_analyzer_with_topic_query ():
+
+    st.title("Simple RAG Analyzer with Focusing Query")
+
+    prompter = Prompt()
+
+    sample_files_path = Setup().load_sample_files(over_write=False)
+    doc_path = os.path.join(sample_files_path, "Agreements")
+
+    files = os.listdir(doc_path)
+    file_name = st.selectbox("Choose an Agreement", files)
+
+    #   ** topic_query ** = this is a proxy for a more complex focusing retrieval strategy to target only a
+    #   specific part of the document, rather then the whole document
+    #   in this case, this will run a 'text match' search against the topic query to reduce the
+    #   text chunks reviewed in trying to answer the question
+
+    topic_query = st.text_area("Filtering Topic (hint: 'vacation')")
+
+    #   ** prompt_text ** - this is the question that will be passed to the LLM
+    prompt_text = st.text_area("Question (hint: 'how many vacation days will the executive receive'")
+
+    model_name = st.selectbox("Choose a model for answering questions", ["bling-phi-3-gguf",
+                                                                         "bling-tiny-llama-1b",
+                                                                         "bling-stablelm-3b-tool",
+                                                                         "llama-3-instruct-bartowski-gguf",
+                                                                         "dragon-llama-answer-tool"])
+
+    if st.button("Run Analysis"):
+
+        if file_name and prompt_text and model_name:
+
+            prompter.load_model(model_name, temperature=0.0, sample=False)
+
+            #   parse the PDF in memory and attach to the prompt
+            if not topic_query:
+                sources = prompter.add_source_document(doc_path,file_name)
+            else:
+                # this is where we use the topic_query to filter the parsed document
+                sources = prompter.add_source_document(doc_path,file_name, query=topic_query)
+
+            #   run the inference with the source
+            response = prompter.prompt_with_source(prompt_text)
+
+            #   fact checks
+            fc = prompter.evidence_check_numbers(response)
+            cs = prompter.evidence_check_sources(response)
+
+            if len(response) > 0:
+                if "llm_response" in response[0]:
+                    response = response[0]["llm_response"]
+
+                    st.write(f"Answer: {response}")
+
+                    if len(fc) > 0:
+                        if "fact_check" in fc[0]:
+                            fc_out = fc[0]["fact_check"]
+                            st.write(f"Numbers Check: {fc_out}")
+
+                    if len(cs) > 0:
+                        if "source_review" in cs[0]:
+                            sr_out = cs[0]["source_review"]
+                            st.write(f"Source review: {sr_out}")
+
+
+if __name__ == "__main__":
+
+    simple_analyzer_with_topic_query()

examples/UI/simple_rag_ui_with_streamlit.py

@@ -0,0 +1,86 @@
+
+""" This example shows how to build a simple RAG application with UI with Streamlit and LLMWare.
+
+    Note: it requires a separate `pip install streamlit`, and to run the script, you should run from the
+    command line with:
+
+     `streamlit run using_with_streamlit_ui.py`
+
+    For this example, we will be prompting against a set of Invoice documents, provided in the LLMWare
+    sample files.
+
+    If you would like to substitute longer documents then please look at the UI example:
+        -- rag_ui_with_query_topic_with_streamlit.py
+
+    as a framework to get started integrating a retrieval step before the prompt of the source
+
+    For more information about Streamlit, check out their docs:  https://docs.streamlit.io/develop/tutorials
+
+"""
+
+
+import os
+import streamlit as st
+
+from llmware.prompts import Prompt
+from llmware.setup import Setup
+
+# st.set_page_config(layout="wide")
+
+
+def simple_analyzer ():
+
+    st.title("Simple RAG Analyzer")
+
+    prompter = Prompt()
+
+    sample_files_path = Setup().load_sample_files(over_write=False)
+    doc_path = os.path.join(sample_files_path, "Invoices")
+
+    files = os.listdir(doc_path)
+    file_name = st.selectbox("Choose an Invoice", files)
+
+    prompt_text = st.text_area("Question (hint: 'what is the total amount of the invoice?'")
+
+    model_name = st.selectbox("Choose a model for answering questions", ["bling-phi-3-gguf",
+                                                                         "bling-tiny-llama-1b",
+                                                                         "bling-stablelm-3b-tool",
+                                                                         "llama-3-instruct-bartowski-gguf",
+                                                                         "dragon-llama-answer-tool"])
+
+    if st.button("Run Analysis"):
+
+        if file_name and prompt_text and model_name:
+
+            prompter.load_model(model_name, temperature=0.0, sample=False)
+
+            #   parse the PDF in memory and attach to the prompt
+            sources = prompter.add_source_document(doc_path,file_name)
+
+            #   run the inference with the source
+            response = prompter.prompt_with_source(prompt_text)
+
+            #   fact checks
+            fc = prompter.evidence_check_numbers(response)
+            cs = prompter.evidence_check_sources(response)
+
+            if len(response) > 0:
+                if "llm_response" in response[0]:
+                    response = response[0]["llm_response"]
+
+                    st.write(f"Answer: {response}")
+
+                    if len(fc) > 0:
+                        if "fact_check" in fc[0]:
+                            fc_out = fc[0]["fact_check"]
+                            st.write(f"Numbers Check: {fc_out}")
+
+                    if len(cs) > 0:
+                        if "source_review" in cs[0]:
+                            sr_out = cs[0]["source_review"]
+                            st.write(f"Source review: {sr_out}")
+
+
+if __name__ == "__main__":
+
+    simple_analyzer()

examples/UI/using_streamlit_chat_ui.py

@@ -0,0 +1,81 @@
+
+""" This example provides a basic framework to build a Chatbot UI interface in conjunction with LLMWare
+    using Streamlit Chat UI.
+
+    To run this example requires an install of Streamlit, e.g., `pip3 install streamlit`
+
+    To execute the script, run from the command line with:  `streamlit run using_with_streamlit_ui.py`
+
+    Also, please note that the first time you run with a new model, the model will be downloaded and cached locally,
+    so expect a delay on the 'first run' which will be much faster on every successive run.
+
+    All components of the chatbot will be running locally, so the speed will be determined greatly by the
+    CPU/GPU capacities of your machine.
+
+    We have set the max_output at 250 tokens - for faster, set lower ...
+
+    For more information on the Streamlit Chat UI,
+    see https://docs.streamlit.io/develop/tutorials/llms/build-conversational-apps
+
+
+"""
+
+
+import streamlit as st
+from llmware.models import ModelCatalog
+
+
+def simple_chat_ui_app (model_name):
+
+    st.title(f"Simple Chat with {model_name}")
+
+    model = ModelCatalog().load_model(model_name, temperature=0.3, sample=True, max_output=250)
+
+    # initialize chat history
+    if "messages" not in st.session_state:
+        st.session_state.messages = []
+
+    # display chat messages from history on app rerun
+    for message in st.session_state.messages:
+        with st.chat_message(message["role"]):
+            st.markdown(message["content"])
+
+    # accept user input
+    prompt = st.chat_input("Say something")
+    if prompt:
+
+        with st.chat_message("user"):
+            st.markdown(prompt)
+
+        with st.chat_message("assistant"):
+
+            model_response = model.inference(prompt)
+
+            # insert additional error checking / post-processing of output here
+            bot_response = model_response["llm_response"]
+
+            st.markdown(bot_response)
+
+        st.session_state.messages.append({"role": "user", "content": prompt})
+        st.session_state.messages.append({"role": "assistant", "content": bot_response})
+
+    return 0
+
+
+if __name__ == "__main__":
+
+    #   a few representative good chat models that can run locally
+    #   note: will take a minute for the first time it is downloaded and cached locally
+
+    chat_models = ["phi-3-gguf",
+                   "llama-2-7b-chat-gguf",
+                   "llama-3-instruct-bartowski-gguf",
+                   "openhermes-mistral-7b-gguf",
+                   "zephyr-7b-gguf"]
+
+    model_name = chat_models[0]
+
+    simple_chat_ui_app(model_name)
+
+
+