0.1.1 refactor and ui changes

app.py

@@ -33,6 +33,7 @@ index_model = "Writer/camel-5b-hf"
 INDEX_NAME = f"{index_model.replace('/', '-')}-default-no-coref"
 persist_path = f"storage/{INDEX_NAME}"
 MAX_LENGTH = 1024
+MAX_NEW_TOKENS = 250
 
 import baseten
 @st.cache_resource
@@ -68,10 +69,6 @@ f'''
 )
     st.caption('''###### corpus by [@ArxivHealthcareNLP@sigmoid.social](https://sigmoid.social/@ArxivHealthcareNLP)''')
     st.caption('''###### KG Questions by [arylwen](https://github.com/arylwen/mlk8s)''')
-#    st.write(
-#f'''
-##### How can <what most are doing> help with <what few are doing>?
-#''')
     
 from llama_index import StorageContext
 from llama_index import ServiceContext
@@ -99,7 +96,6 @@ enc = tiktoken.get_encoding("gpt2")
 tokenizer = lambda text: enc.encode(text, allowed_special={"<|endoftext|>"})
 globals_helper._tokenizer = tokenizer
 
-
 def set_openai_local():
     openai.api_key = os.environ['LOCAL_OPENAI_API_KEY']
     openai.api_base = os.environ['LOCAL_OPENAI_API_BASE']
@@ -111,12 +107,15 @@ def set_openai():
     openai.api_base = os.environ['DAVINCI_OPENAI_API_BASE']
     os.environ['OPENAI_API_KEY'] = os.environ['DAVINCI_OPENAI_API_KEY']
     os.environ['OPENAI_API_BASE'] = os.environ['DAVINCI_OPENAI_API_BASE']
-
+ 
+from kron.llm_predictor.KronHFHubLLM import KronHuggingFaceHub
 def get_hf_predictor(query_model):
     # no embeddings for now
     set_openai_local()
-    llm=HuggingFaceHub(repo_id=query_model, task="text-generation", 
-                       model_kwargs={"temperature": 0.01, "max_new_tokens": MAX_LENGTH, 'frequency_penalty':1.17}, 
+    #llm=HuggingFaceHub(repo_id=query_model, task="text-generation", 
+    llm=KronHuggingFaceHub(repo_id=query_model, task="text-generation", 
+#                       model_kwargs={"temperature": 0.01, "max_new_tokens": MAX_NEW_TOKENS, 'frequency_penalty':1.17}, 
+                       model_kwargs={"temperature": 0.01, "max_new_tokens": MAX_NEW_TOKENS }, 
                        huggingfacehub_api_token=hf_api_key)
     llm_predictor = LLMPredictor(llm)
     return llm_predictor
@@ -264,7 +263,7 @@ else :
     with query:
         answer_model = st.radio(
             "Choose the model used for inference:",
-            ('Local-Camel', 'HF-TKI', 'hf/tiiuae/falcon-7b-instruct', 'openai/text-davinci-003')
+            ('Writer/camel-5b-hf', 'mosaicml/mpt-7b-instruct', 'hf/tiiuae/falcon-7b-instruct', 'cohere/command', 'baseten/Camel-5b', 'openai/text-davinci-003')
         )
 
 if answer_model == 'openai/text-davinci-003':
@@ -304,7 +303,7 @@ elif answer_model == 'baseten/Camel-5b':
     most_connected = random.sample(graph_nodes[:100], 5)
     low_connected = get_networkx_low_connected_components( "", persist_path)
     least_connected = random.sample(low_connected, 5)
-elif answer_model == 'Local-Camel':
+elif answer_model == 'Writer/camel-5b-hf':
     query_model = 'Writer/camel-5b-hf'
     print(answer_model)
     clear_question(query_model)
@@ -314,8 +313,8 @@ elif answer_model == 'Local-Camel':
     most_connected = random.sample(graph_nodes[:100], 5)
     low_connected = get_networkx_low_connected_components( "", persist_path)
     least_connected = random.sample(low_connected, 5)
-elif answer_model == 'HF-TKI':    
-    query_model = 'allenai/tk-instruct-3b-def-pos-neg-expl'
+elif answer_model == 'mosaicml/mpt-7b-instruct':    
+    query_model = 'mosaicml/mpt-7b-instruct'
     clear_question(query_model)
     query_engine = build_hf_query_engine(query_model, persist_path)
     graph_nodes = get_networkx_graph_nodes( "", persist_path)
@@ -325,21 +324,25 @@ elif answer_model == 'HF-TKI':
 else:
     print('This is a bug.')
 
-# to clear input box
+# to clear the input box
 def submit():
     st.session_state.question = st.session_state.question_input
     st.session_state.question_input = ''
     st.session_state.question_answered = False
 
 with st.sidebar:
-    option_1 = st.selectbox("What most are studying:", most_connected, disabled=True)
+    import gensim
+    m_connected = []
+    for item in most_connected:
+        if not item[0].lower() in gensim.parsing.preprocessing.STOPWORDS:
+            m_connected.extend([item[0].lower()])
+    option_1 = st.selectbox("What most are studying:", m_connected, disabled=True)
     option_2 = st.selectbox("What few are studying:", least_connected, disabled=True)
 
 with query:
-    st.caption(f'''###### Only intended for educational and research purposes. Please do not enter any private or confidential information. Model, question, answer and rating are logged to improve KG Questions.''')
-    #st.caption(f'''Model, question, answer and rating are logged to improve KG Questions.''')
+    st.caption(f'''###### Intended for educational and research purpose. Please do not enter any private or confidential information.   Model, question, answer and rating are logged to improve KG Questions.''')
     question = st.text_input("Enter a question, e.g. What benchmarks can we use for QA?", key='question_input',  on_change=submit )
- 
+
 if(st.session_state.question):    
     try :
         with query:
@@ -363,121 +366,13 @@ if(st.session_state.question):
                 from streamlit_star_rating import st_star_rating
                 stars = st_star_rating("", maxValue=5, defaultValue=3, key="answer_rating")
             st.write(answer_str)
-
         with measurable:
-            from wordcloud import WordCloud, STOPWORDS, ImageColorGenerator
-            import matplotlib.pyplot as plt
-            from PIL import Image
-            wc_all, wc_question, wc_reference  = st.columns([3, 3, 3])
-            wordcloud = WordCloud(max_font_size=50, max_words=1000, background_color="white")
-            with wc_all:
-                image = Image.open('docs/images/all_papers_wordcloud.png')
-                st.image(image)
-                st.caption('''###### Corpus term frequecy.''')
-            with wc_question:
-                wordcloud_q = wordcloud.generate(answer_str)
-                st.image(wordcloud_q.to_array())
-                st.caption('''###### Answer term frequecy.''')
-            with wc_reference:
-                all_reference_texts = ''
-                for nodewithscore in answer.source_nodes:  
-                    node = nodewithscore.node
-                    from llama_index.schema import NodeRelationship
-                    #if NodeRelationship.SOURCE in node.relationships:
-                    all_reference_texts = all_reference_texts + '\n' + node.text
-                wordcloud_r = wordcloud.generate(all_reference_texts)
-                st.image(wordcloud_r.to_array())
-                st.caption('''###### Reference plus graph term frequecy.''')
-                
+            from measurable import display_wordcloud
+            display_wordcloud(answer, answer_str)
         with explainable:
-            #st.write(answer.source_nodes)
-            from pyvis.network import Network
-            graph = Network(height="450px", width="100%")
-            sources_table = []
-            #all_reference_texts = ''
-            for nodewithscore in answer.source_nodes:  
-                node = nodewithscore.node
-                from llama_index.schema import NodeRelationship
-                if NodeRelationship.SOURCE in node.relationships:
-                    node_id = node.relationships[NodeRelationship.SOURCE].node_id
-                    node_id = node_id.split('/')[-1]
-                    title = node_id.split('.')[2].replace('_', ' ')
-                    link = '.'.join(node_id.split('.')[:2])[:10]
-                    link = f'https://arxiv.org/abs/{link}'
-                    href = f'<a  target="_blank" href="{link}">{title}</a>'
-                    sources_table.extend([[href, node.text]])
-                    #all_reference_texts = all_reference_texts + '\n' + node.text
-                else:
-                    #st.write(node.text) TODO second level relationships
-                    rel_map = node.metadata['kg_rel_map']
-                    for concept in rel_map.keys():
-                        #st.write(concept)
-                        graph.add_node(concept,  concept, title=concept)
-                        rels = rel_map[concept]
-                        for rel in rels:
-                            graph.add_node(rel[1],  rel[1], title=rel[1])
-                            graph.add_edge(concept,  rel[1], title=rel[0])
-            # --- display the query terms graph
-            st.session_state.graph_name = 'graph.html'
-            graph.save_graph(st.session_state.graph_name)            
-            import streamlit.components.v1 as components
-            graphHtml = open(st.session_state.graph_name, 'r', encoding='utf-8')
-            source_code = graphHtml.read() 
-            components.html(source_code, height = 500)
-            # --- display the reference texts table
-            import pandas as pd
-            df = pd.DataFrame(sources_table)
-            df.columns = ['paper', 'relevant text']
-            st.markdown(""" <style> 
-                                table[class*="dataframe"] {
-                                font-size: 10px;
-                            }
-                            </style> """, unsafe_allow_html=True)
-            st.write(df.to_html(escape=False), unsafe_allow_html=True)
-            # reference text wordcloud
-            #st.session_state.reference_wcloud = all_reference_texts
-            
-        with ethical:
-            st.write('##### Bias, risks, limitations and terms of use for the models.')
-            ethics_statement = []
-            falcon = ['hf/tiiuae/falcon-7b-instruct', '<a  target="_blank" href="https://huggingface.co/tiiuae/falcon-7b">Bias, Risks, and Limitations</a>']
-            cohere = ['cohere/command', '<a  target="_blank" href="https://cohere.com/terms-of-use">Terms of use</a>']
-            camel = ['baseten/Camel-5b', '<a  target="_blank" href="https://huggingface.co/Writer/camel-5b-hf">Bias, Risks, and Limitations</a>']
-            davinci = ['openai/text-davinci-003', '<a  target="_blank" href="https://openai.com/policies/terms-of-use">Terms of Use</a>']
-            
-            ethics_statement.extend([falcon, cohere, camel, davinci])
-            df = pd.DataFrame(ethics_statement)
-            df.columns = ['model', 'model link']
-            st.markdown(""" <style> 
-                                table[class*="dataframe"] {
-                                font-size: 14px;
-                            }
-                            </style> """, unsafe_allow_html=True)
-            st.write(df.to_html(escape=False), unsafe_allow_html=True)
-            # license
-            st.write('')
-            st.write('##### How papers were included in the index based on license.')
-            st.caption(f'The paper id and title has been included in the index for a full attribution to the authors')
-            ccby = ['<a target="_blank" href="https://creativecommons.org/licenses/by/4.0/">CC BY</a>', 
-                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">full content KG questions pipeline</a>']
-            ccbysa = ['<a target="_blank" href="https://creativecommons.org/licenses/by-sa/4.0/">CC BY-SA</a>', 
-                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">full content KG questions pipeline</a>']
-            ccbyncsa = ['<a target="_blank" href="https://creativecommons.org/licenses/by-nc-sa/4.0/">CC NC-BY-NC-SA</a>', 
-                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">full content KG questions pipeline</a>']
-            ccbyncnd = ['<a target="_blank" href="https://creativecommons.org/licenses/by-nc-nd/4.0/">CC NC-BY-NC-ND</a>', 
-                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">arxiv metadata abstract KG questions pipeline</a>']
-            license_statement = [ccby, ccbysa, ccbyncsa, ccbyncnd]
-            df = pd.DataFrame(license_statement)
-            df.columns = ['license', 'how papers are used']
-            st.markdown(""" <style> 
-                                table[class*="dataframe"] {
-                                font-size: 14px;
-                            }
-                            </style> """, unsafe_allow_html=True)
-            st.write(df.to_html(escape=False), unsafe_allow_html=True)
-
+            from explainable import explain
+            explain(answer)
     except Exception as e:
-        #print(f'{type(e)}, {e}')
         answer_str = f'{type(e)}, {e}'
         st.session_state.answer_rating = -1
         st.write(f'An error occured, please try again. \n{answer_str}')
@@ -486,4 +381,11 @@ if(st.session_state.question):
             req = st.session_state.question
             if(__spaces__):
                 st.session_state.request_log.add_request_log_entry(query_model, req, answer_str, st.session_state.answer_rating)
- 
+else:
+    with measurable:
+        st.write(f'###### Ask a question to see a comparison between the corpus, answer and reference documents.') 
+    with explainable:
+        st.write(f'###### Ask a question to see the knowledge graph and a list of reference documents.')
+with ethical:
+    from ethics import display_ethics
+    display_ethics()

ethics.py

@@ -0,0 +1,48 @@
+import streamlit as st
+import pandas as pd
+
+def display_ethics():
+    # ethics statement
+    display_ethics_statement()
+    # license
+    display_license_statement()
+
+def display_license_statement():
+    st.write('')
+    st.write('##### How papers were included in the index based on license.')
+    st.caption(f'The paper id and title has been included in the index for a full attribution to the authors')
+    ccby = ['<a target="_blank" href="https://creativecommons.org/licenses/by/4.0/">CC BY</a>', 
+                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">full content KG questions pipeline</a>']
+    ccbysa = ['<a target="_blank" href="https://creativecommons.org/licenses/by-sa/4.0/">CC BY-SA</a>', 
+                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">full content KG questions pipeline</a>']
+    ccbyncsa = ['<a target="_blank" href="https://creativecommons.org/licenses/by-nc-sa/4.0/">CC NC-BY-NC-SA</a>', 
+                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">full content KG questions pipeline</a>']
+    ccbyncnd = ['<a target="_blank" href="https://creativecommons.org/licenses/by-nc-nd/4.0/">CC NC-BY-NC-ND</a>', 
+                      '<a target="_blank" href="https://github.com/arylwen/mlk8s/tree/main/apps/papers-kg">arxiv metadata abstract KG questions pipeline</a>']
+    license_statement = [ccby, ccbysa, ccbyncsa, ccbyncnd]
+    df = pd.DataFrame(license_statement)
+    df.columns = ['license', 'how papers are used']
+    st.markdown(""" <style> 
+                                table[class*="dataframe"] {
+                                font-size: 14px;
+                            }
+                            </style> """, unsafe_allow_html=True)
+    st.write(df.to_html(escape=False), unsafe_allow_html=True)
+
+def display_ethics_statement():
+    st.write('##### Bias, risks, limitations and terms of use for the models.')
+    ethics_statement = []
+    falcon = ['hf/tiiuae/falcon-7b-instruct', '<a  target="_blank" href="https://huggingface.co/tiiuae/falcon-7b">Bias, Risks, and Limitations</a>']
+    cohere = ['cohere/command', '<a  target="_blank" href="https://cohere.com/terms-of-use">Terms of use</a>']
+    camel = ['baseten/Camel-5b', '<a  target="_blank" href="https://huggingface.co/Writer/camel-5b-hf">Bias, Risks, and Limitations</a>']
+    davinci = ['openai/text-davinci-003', '<a  target="_blank" href="https://openai.com/policies/terms-of-use">Terms of Use</a>']
+            
+    ethics_statement.extend([falcon, cohere, camel, davinci])
+    df = pd.DataFrame(ethics_statement)
+    df.columns = ['model', 'model link']
+    st.markdown(""" <style> 
+                                table[class*="dataframe"] {
+                                font-size: 14px;
+                            }
+                            </style> """, unsafe_allow_html=True)
+    st.write(df.to_html(escape=False), unsafe_allow_html=True)

explainable.py

@@ -0,0 +1,46 @@
+from pyvis.network import Network
+import streamlit as st
+
+def explain(answer):
+    graph = Network(height="450px", width="100%")
+    sources_table = []
+            #all_reference_texts = ''
+    for nodewithscore in answer.source_nodes:  
+        node = nodewithscore.node
+        from llama_index.schema import NodeRelationship
+        if NodeRelationship.SOURCE in node.relationships:
+            node_id = node.relationships[NodeRelationship.SOURCE].node_id
+            node_id = node_id.split('/')[-1]
+            title = node_id.split('.')[2].replace('_', ' ')
+            link = '.'.join(node_id.split('.')[:2])[:10]
+            link = f'https://arxiv.org/abs/{link}'
+            href = f'<a  target="_blank" href="{link}">{title}</a>'
+            sources_table.extend([[href, node.text]])
+                    #all_reference_texts = all_reference_texts + '\n' + node.text
+        else:
+                    #st.write(node.text) TODO second level relationships
+            rel_map = node.metadata['kg_rel_map']
+            for concept in rel_map.keys():
+                        #st.write(concept)
+                graph.add_node(concept,  concept, title=concept)
+                rels = rel_map[concept]
+                for rel in rels:
+                    graph.add_node(rel[1],  rel[1], title=rel[1])
+                    graph.add_edge(concept,  rel[1], title=rel[0])
+            # --- display the query terms graph
+    st.session_state.graph_name = 'graph.html'
+    graph.save_graph(st.session_state.graph_name)            
+    import streamlit.components.v1 as components
+    graphHtml = open(st.session_state.graph_name, 'r', encoding='utf-8')
+    source_code = graphHtml.read() 
+    components.html(source_code, height = 500)
+            # --- display the reference texts table
+    import pandas as pd
+    df = pd.DataFrame(sources_table)
+    df.columns = ['paper', 'relevant text']
+    st.markdown(""" <style> 
+                                table[class*="dataframe"] {
+                                font-size: 10px;
+                            }
+                            </style> """, unsafe_allow_html=True)
+    st.write(df.to_html(escape=False), unsafe_allow_html=True)

graph.html

@@ -88,8 +88,8 @@
                   
 
                   // parsing and collecting nodes and edges from the python
-                  nodes = new vis.DataSet([{"color": "#97c2fc", "id": "help", "label": "help", "shape": "dot", "title": "help"}, {"color": "#97c2fc", "id": "KEYWORDS", "label": "KEYWORDS", "shape": "dot", "title": "KEYWORDS"}, {"color": "#97c2fc", "id": "following", "label": "following", "shape": "dot", "title": "following"}, {"color": "#97c2fc", "id": "could", "label": "could", "shape": "dot", "title": "could"}, {"color": "#97c2fc", "id": "sentences", "label": "sentences", "shape": "dot", "title": "sentences"}, {"color": "#97c2fc", "id": "What systems could help with the following sentences?\n\nKEYWORDS: systems", "label": "What systems could help with the following sentences?\n\nKEYWORDS: systems", "shape": "dot", "title": "What systems could help with the following sentences?\n\nKEYWORDS: systems"}, {"color": "#97c2fc", "id": "systems", "label": "systems", "shape": "dot", "title": "systems"}, {"color": "#97c2fc", "id": "to shared task leaderboard", "label": "to shared task leaderboard", "shape": "dot", "title": "to shared task leaderboard"}, {"color": "#97c2fc", "id": "What", "label": "What", "shape": "dot", "title": "What"}, {"color": "#97c2fc", "id": "the capital of France", "label": "the capital of France", "shape": "dot", "title": "the capital of France"}, {"color": "#97c2fc", "id": "the capital", "label": "the capital", "shape": "dot", "title": "the capital"}, {"color": "#97c2fc", "id": "Alice", "label": "Alice", "shape": "dot", "title": "Alice"}, {"color": "#97c2fc", "id": "Bob", "label": "Bob", "shape": "dot", "title": "Bob"}, {"color": "#97c2fc", "id": "Carol", "label": "Carol", "shape": "dot", "title": "Carol"}, {"color": "#97c2fc", "id": "David", "label": "David", "shape": "dot", "title": "David"}, {"color": "#97c2fc", "id": "Jane", "label": "Jane", "shape": "dot", "title": "Jane"}, {"color": "#97c2fc", "id": "Peter", "label": "Peter", "shape": "dot", "title": "Peter"}, {"color": "#97c2fc", "id": "Sarah", "label": "Sarah", "shape": "dot", "title": "Sarah"}, {"color": "#97c2fc", "id": "Tom", "label": "Tom", "shape": "dot", "title": "Tom"}, {"color": "#97c2fc", "id": "William", "label": "William", "shape": "dot", "title": "William"}, {"color": "#97c2fc", "id": "Yolanda", "label": "Yolanda", "shape": "dot", "title": "Yolanda"}, {"color": "#97c2fc", "id": "Zane", "label": "Zane", "shape": "dot", "title": "Zane"}, {"color": "#97c2fc", "id": "Charlie", "label": "Charlie", "shape": "dot", "title": "Charlie"}, {"color": "#97c2fc", "id": "Sara", "label": "Sara", "shape": "dot", "title": "Sara"}, {"color": "#97c2fc", "id": "mother of", "label": "mother of", "shape": "dot", "title": "mother of"}, {"color": "#97c2fc", "id": "wife of", "label": "wife of", "shape": "dot", "title": "wife of"}, {"color": "#97c2fc", "id": "daughter of", "label": "daughter of", "shape": "dot", "title": "daughter of"}, {"color": "#97c2fc", "id": "Sam", "label": "Sam", "shape": "dot", "title": "Sam"}, {"color": "#97c2fc", "id": "Susan", "label": "Susan", "shape": "dot", "title": "Susan"}, {"color": "#97c2fc", "id": "Zara", "label": "Zara", "shape": "dot", "title": "Zara"}, {"color": "#97c2fc", "id": "Wendy", "label": "Wendy", "shape": "dot", "title": "Wendy"}, {"color": "#97c2fc", "id": "Davi", "label": "Davi", "shape": "dot", "title": "Davi"}, {"color": "#97c2fc", "id": "Caro", "label": "Caro", "shape": "dot", "title": "Caro"}, {"color": "#97c2fc", "id": "John", "label": "John", "shape": "dot", "title": "John"}, {"color": "#97c2fc", "id": "Mary", "label": "Mary", "shape": "dot", "title": "Mary"}, {"color": "#97c2fc", "id": "granddaughter of", "label": "granddaughter of", "shape": "dot", "title": "granddaughter of"}, {"color": "#97c2fc", "id": "niece of", "label": "niece of", "shape": "dot", "title": "niece of"}, {"color": "#97c2fc", "id": "nephew of", "label": "nephew of", "shape": "dot", "title": "nephew of"}, {"color": "#97c2fc", "id": "George", "label": "George", "shape": "dot", "title": "George"}, {"color": "#97c2fc", "id": "Bill", "label": "Bill", "shape": "dot", "title": "Bill"}, {"color": "#97c2fc", "id": "Emily", "label": "Emily", "shape": "dot", "title": "Emily"}, {"color": "#97c2fc", "id": "Charlie and Jane", "label": "Charlie and Jane", "shape": "dot", "title": "Charlie and Jane"}, {"color": "#97c2fc", "id": "Jane and Charlie", "label": "Jane and Charlie", "shape": "dot", "title": "Jane and Charlie"}, {"color": "#97c2fc", "id": "Jan", "label": "Jan", "shape": "dot", "title": "Jan"}, {"color": "#97c2fc", "id": "is being prescribed Lipitor for treatment", "label": "is being prescribed Lipitor for treatment", "shape": "dot", "title": "is being prescribed Lipitor for treatment"}, {"color": "#97c2fc", "id": "0.8", "label": "0.8", "shape": "dot", "title": "0.8"}, {"color": "#97c2fc", "id": "Greg", "label": "Greg", "shape": "dot", "title": "Greg"}, {"color": "#97c2fc", "id": "Jennifer", "label": "Jennifer", "shape": "dot", "title": "Jennifer"}, {"color": "#97c2fc", "id": "Mark", "label": "Mark", "shape": "dot", "title": "Mark"}, {"color": "#97c2fc", "id": "father of Charlie", "label": "father of Charlie", "shape": "dot", "title": "father of Charlie"}, {"color": "#97c2fc", "id": "Jack", "label": "Jack", "shape": "dot", "title": "Jack"}, {"color": "#97c2fc", "id": "Kevin", "label": "Kevin", "shape": "dot", "title": "Kevin"}, {"color": "#97c2fc", "id": "Lisa", "label": "Lisa", "shape": "dot", "title": "Lisa"}, {"color": "#97c2fc", "id": "Tim", "label": "Tim", "shape": "dot", "title": "Tim"}, {"color": "#97c2fc", "id": "HIV", "label": "HIV", "shape": "dot", "title": "HIV"}, {"color": "#97c2fc", "id": "flu", "label": "flu", "shape": "dot", "title": "flu"}, {"color": "#97c2fc", "id": "HIV and flu", "label": "HIV and flu", "shape": "dot", "title": "HIV and flu"}, {"color": "#97c2fc", "id": "mother of Alice", "label": "mother of Alice", "shape": "dot", "title": "mother of Alice"}]);
-                  edges = new vis.DataSet([{"from": "systems", "title": "submitted", "to": "to shared task leaderboard"}, {"from": "What", "title": "is", "to": "the capital of France"}, {"from": "What", "title": "is", "to": "the capital"}, {"from": "What", "title": "is your name?", "to": "Alice"}, {"from": "What", "title": "is mother of", "to": "Bob"}, {"from": "What", "title": "is father of", "to": "Carol"}, {"from": "What", "title": "is father of", "to": "David"}, {"from": "What", "title": "is father of", "to": "Jane"}, {"from": "What", "title": "is father of", "to": "Peter"}, {"from": "What", "title": "is father of", "to": "Sarah"}, {"from": "What", "title": "is father of", "to": "Tom"}, {"from": "What", "title": "is father of", "to": "William"}, {"from": "What", "title": "is father of", "to": "Yolanda"}, {"from": "What", "title": "is father of", "to": "Zane"}, {"from": "What", "title": "is father of", "to": "Charlie"}, {"from": "What", "title": "is mother of", "to": "Sara"}, {"from": "What", "title": "is", "to": "mother of"}, {"from": "What", "title": "is", "to": "wife of"}, {"from": "What", "title": "is", "to": "daughter of"}, {"from": "What", "title": "is father of", "to": "Sam"}, {"from": "What", "title": "is father of", "to": "Susan"}, {"from": "What", "title": "is father of", "to": "Zara"}, {"from": "What", "title": "is father of", "to": "Wendy"}, {"from": "What", "title": "is father of", "to": "Davi"}, {"from": "What", "title": "is father of", "to": "Caro"}, {"from": "What", "title": "is father of", "to": "John"}, {"from": "What", "title": "is father of", "to": "Mary"}, {"from": "What", "title": "is", "to": "granddaughter of"}, {"from": "What", "title": "is", "to": "niece of"}, {"from": "What", "title": "is", "to": "nephew of"}, {"from": "What", "title": "is brother of", "to": "George"}, {"from": "What", "title": "is wife of", "to": "Bill"}, {"from": "What", "title": "is daughter of", "to": "Emily"}, {"from": "What", "title": "is mother of", "to": "Charlie and Jane"}, {"from": "What", "title": "is father of", "to": "Jane and Charlie"}, {"from": "What", "title": "is mother of", "to": "Jan"}, {"from": "What", "title": "has hyperlipidmia", "to": "is being prescribed Lipitor for treatment"}, {"from": "What", "title": "0.8", "to": "0.8"}, {"from": "What", "title": "is mother of", "to": "Greg"}, {"from": "What", "title": "is father of", "to": "Jennifer"}, {"from": "What", "title": "is father of", "to": "Mark"}, {"from": "What", "title": "is", "to": "father of Charlie"}, {"from": "What", "title": "is father of", "to": "Jack"}, {"from": "What", "title": "is father of", "to": "Kevin"}, {"from": "What", "title": "is father of", "to": "Lisa"}, {"from": "What", "title": "is father of", "to": "Tim"}, {"from": "What", "title": "has", "to": "HIV"}, {"from": "What", "title": "has", "to": "flu"}, {"from": "What", "title": "has", "to": "HIV and flu"}, {"from": "What", "title": "is", "to": "mother of Alice"}]);
+                  nodes = new vis.DataSet([]);
+                  edges = new vis.DataSet([]);
 
                   nodeColors = {};
                   allNodes = nodes.get({ returnType: "Object" });

kron/llm_predictor/KronHFHubLLM.py

@@ -0,0 +1,85 @@
+from __future__ import annotations
+
+import time
+
+from typing import Any, Callable, List, Optional
+
+from langchain.callbacks.manager import CallbackManagerForLLMRun
+from langchain.llms.huggingface_hub import HuggingFaceHub
+
+import logging
+logger = logging.getLogger(__name__)
+
+from tenacity import (
+    before_sleep_log,
+    retry,
+    retry_if_exception_type,
+    stop_after_attempt,
+    wait_exponential,
+)
+
+
+def _create_retry_decorator(llm: KronHuggingFaceHub) -> Callable[[Any], Any]:
+    #import cohere
+
+    min_seconds = 4
+    max_seconds = 10
+    # Wait 2^x * 1 second between each retry starting with
+    # 4 seconds, then up to 10 seconds, then 10 seconds afterwards
+    return retry(
+        reraise=True,
+        stop=stop_after_attempt(llm.max_retries),
+        wait=wait_exponential(multiplier=1, min=min_seconds, max=max_seconds),
+        retry=(retry_if_exception_type(KronHFHubRateExceededException)),
+        before_sleep=before_sleep_log(logger, logging.WARNING),
+    )
+
+
+def completion_with_retry(llm: KronHuggingFaceHub, **kwargs: Any) -> Any:
+    """Use tenacity to retry the completion call."""
+    retry_decorator = _create_retry_decorator(llm)
+
+    @retry_decorator
+    def _completion_with_retry(**kwargs: Any) -> Any:
+        return llm.internal_call(**kwargs)
+
+    return _completion_with_retry(**kwargs)
+
+class KronHFHubRateExceededException(Exception):
+    def __init__(self, message="HF Hub Service Unavailable: Rate exceeded."):
+        self.message = message
+        super().__init__(self.message)
+
+
+class KronHuggingFaceHub(HuggingFaceHub):
+
+    max_retries: int = 10
+    """Maximum number of retries to make when generating."""
+
+    def internal_call(
+        self,
+        prompt: str,
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> str:
+        try:
+            print(f'**************************************\n{prompt}')
+            response =  super()._call(prompt, stop, run_manager, **kwargs)
+            print(f'**************************************\n{response}')
+            return response
+        except ValueError as ve:
+            if "Service Unavailable" in str(ve):
+                raise KronHFHubRateExceededException()
+            else:
+                raise ve
+
+    def _call(
+        self,
+        prompt: str,
+        stop: Optional[List[str]] = None,
+        run_manager: Optional[CallbackManagerForLLMRun] = None,
+        **kwargs: Any,
+    ) -> str:
+        response = completion_with_retry(self, prompt=prompt, stop=stop, run_manager=run_manager, **kwargs)
+        return response

measurable.py

@@ -0,0 +1,26 @@
+from wordcloud import WordCloud, STOPWORDS, ImageColorGenerator
+import matplotlib.pyplot as plt
+from PIL import Image
+import streamlit as st
+
+def display_wordcloud(answer, answer_str):
+    wc_all, wc_question, wc_reference  = st.columns([3, 3, 3])
+    wordcloud = WordCloud(max_font_size=50, max_words=1000, background_color="white")
+    with wc_all:
+        image = Image.open('docs/images/all_papers_wordcloud.png')
+        st.image(image)
+        st.caption('''###### Corpus term frequecy.''')
+    with wc_question:
+        wordcloud_q = wordcloud.generate(answer_str)
+        st.image(wordcloud_q.to_array())
+        st.caption('''###### Answer term frequecy.''')
+    with wc_reference:
+        all_reference_texts = ''
+        for nodewithscore in answer.source_nodes:  
+            node = nodewithscore.node
+            from llama_index.schema import NodeRelationship
+                    #if NodeRelationship.SOURCE in node.relationships:
+            all_reference_texts = all_reference_texts + '\n' + node.text
+        wordcloud_r = wordcloud.generate(all_reference_texts)
+        st.image(wordcloud_r.to_array())
+        st.caption('''###### Reference plus graph term frequecy.''')

requirements.txt

@@ -16,6 +16,7 @@ cohere
 baseten
 st-star-rating
 wordcloud
+gensim
 amazon-dax-client>=1.1.7
 boto3>=1.26.79
 pytest>=7.2.1