fdsa

Dockerfile

@@ -52,6 +52,8 @@ ENV PATH="/app/text2vec-transformers:/app/text2vec-transformers/bin:${PATH}"
 #RUN pip install nltk==3.8.1 optimum==1.13.2 onnxruntime==1.16.1 onnx==1.14.1
 RUN ./custom_prerequisites.py
 
+COPY Llama-2-7B-Chat-GGUF/llama-2-7b-chat.Q4_0.gguf /app
+
 ##############################
 RUN useradd -m -u 1000 user
 
@@ -62,5 +64,6 @@ VOLUME /data
 ##############################################################################
 # Start the weaviate vector database, text2vec-transformers and the semantic search app.
 #RUN /app/startup.sh
-#CMD ["/app/startup.sh"]
-RUN --mount=type=cache,target=/data,mode=777 /app/startup.sh
+#RUN --mount=type=cache,target=/data,mode=777 /app/startup.sh
+RUN --mount=type=cache,target=/data,mode=777 echo "### Mounting /data"
+CMD ["/app/startup.sh"]

semsearch.py

@@ -12,6 +12,12 @@ import os
 import re
 import logging
 
+import llama_cpp
+from llama_cpp import Llama
+import ipywidgets as widgets
+from IPython.display import display, clear_output
+
+
 weaviate_logger = logging.getLogger("httpx")
 weaviate_logger.setLevel(logging.WARNING)
 
@@ -34,6 +40,58 @@ webpageChunks = []
 webpageTitles = []
 webpageChunksDocNames = []
 
+#####################################################################
+# Create UI widgets.
+output_widget = widgets.Output()
+with output_widget:
+    print("### Create widgets entered.")
+
+systemTextArea = widgets.Textarea(
+    value='',
+    placeholder='Enter System Prompt.',
+    description='Sys Prompt: ',
+    disabled=False,
+    layout=widgets.Layout(width='300px', height='80px')
+)
+
+userTextArea = widgets.Textarea(
+    value='',
+    placeholder='Enter User Prompt.',
+    description='User Prompt: ',
+    disabled=False,
+    layout=widgets.Layout(width='435px', height='110px')
+)
+
+ragPromptTextArea = widgets.Textarea(
+    value='',
+    placeholder='App generated prompt with RAG information.',
+    description='RAG Prompt: ',
+    disabled=False,
+    layout=widgets.Layout(width='580px', height='180px')
+)
+
+responseTextArea = widgets.Textarea(
+    value='',
+    placeholder='LLM generated response.',
+    description='LLM Resp: ',
+    disabled=False,
+    layout=widgets.Layout(width='780px', height='200px')
+)
+
+selectRag = widgets.Checkbox(
+    value=False,
+    description='Use RAG',
+    disabled=False
+)
+
+submitButton = widgets.Button(
+    description='Run Model.',
+    disabled=False,
+    button_style='', # 'success', 'info', 'warning', 'danger' or ''
+    tooltip='Click',
+    icon='check' # (FontAwesome names without the `fa-` prefix)
+)
+
 
 #######################################################
 # Read each text input file, parse it into a document,
@@ -83,9 +141,9 @@ client = weaviate.connect_to_custom(
     http_secure=False,
     grpc_host="127.0.0.1",
     grpc_port=50051,
-    grpc_secure=False,
-    read_timeout=600,
-    write_timeout=90
+    grpc_secure=False
+    #read_timeout=600,
+    #write_timeout=90
 )
 client.connect() 
 
@@ -265,6 +323,99 @@ for chunk in enumerate(semChunks.objects):
     wpFromChunk = wpCollection.query.fetch_object_by_id(webpage_uuid)
     logger.info(f"### wpFromChunk title: {wpFromChunk.properties['title']}")
 
+
+
+####################################################################
+# 
+collection = client.collections.get("Chunks")
+#model = SentenceTransformer('../multi-qa-MiniLM-L6-cos-v1')
+
+#################################################################
+# Initialize the LLM.
+model_path = "/app/llama-2-7b-chat.Q4_0.gguf"
+llm = Llama(model_path,
+            #*,
+            n_gpu_layers=0,
+            split_mode=llama_cpp.LLAMA_SPLIT_MODE_LAYER,
+            main_gpu=0,
+            tensor_split=None,
+            vocab_only=False,
+            use_mmap=True,
+            use_mlock=False,
+            kv_overrides=None,
+            seed=llama_cpp.LLAMA_DEFAULT_SEED,
+            n_ctx=512,
+            n_batch=512,
+            n_threads=8,
+            n_threads_batch=16,
+            rope_scaling_type=llama_cpp.LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED,
+            pooling_type=llama_cpp.LLAMA_POOLING_TYPE_UNSPECIFIED,
+            rope_freq_base=0.0,
+            rope_freq_scale=0.0,
+            yarn_ext_factor=-1.0,
+            yarn_attn_factor=1.0,
+            yarn_beta_fast=32.0,
+            yarn_beta_slow=1.0,
+            yarn_orig_ctx=0,
+            logits_all=False,
+            embedding=False,
+            offload_kqv=True,
+            last_n_tokens_size=64,
+            lora_base=None,
+            lora_scale=1.0,
+            lora_path=None,
+            numa=False,
+            chat_format=None,
+            chat_handler=None,
+            draft_model=None,
+            tokenizer=None,
+            type_k=None,
+            type_v=None,
+            verbose=True
+           )
+
+
+display(systemTextArea)
+display(userTextArea)
+display(ragPromptTextArea)
+display(responseTextArea)
+display(selectRag)
+display(submitButton)
+
+def setPrompt(pprompt,ragFlag):
+    print("\n### setPrompt() entered. ragFlag: ",ragFlag)
+    if ragFlag:
+        ragPrompt = setRagPrompt(pprompt)
+        userPrompt = pprompt + "\n" + ragPrompt
+        prompt = userPrompt
+    else:
+        userPrompt = pprompt
+    prompt = f""" <s> [INST] <<SYS>> {systemTextArea.value} </SYS>> Q: {userPrompt} A: [/INST]"""
+    return prompt
+
+def runModel(prompt):
+    output = llm.create_completion(
+                  prompt, # Prompt
+                  max_tokens=4096, # Generate up to 32 tokens
+                  #stop = ["Q:", "\n"], # Stop generating just before the model would generate a new question
+                  echo = False # Echo the prompt back in the output
+                )
+    responseTextArea.value = output["choices"][0]["text"]
+
+def on_submitButton_clicked(b):
+    with output_widget:
+        clear_output(wait=True)
+        ragPromptTextArea.value = ""
+        responseTextArea.value = ""
+        log.debug(f"### selectRag: {selectRag.value}")
+        prompt = setPrompt(userTextArea.value,selectRag.value)
+        log.debug("### prompt: " + prompt)
+        runModel(prompt)
+
+submitButton.on_click(on_submitButton_clicked)
+display(output_widget)
+
+
 logger.info("#### Closing client db connection.")
 client.close()
 

semsearch_Hld04_Working.py

@@ -0,0 +1,271 @@
+import weaviate
+
+from sentence_transformers import SentenceTransformer
+from langchain_community.document_loaders import BSHTMLLoader
+from pathlib import Path
+from lxml import html
+import logging
+from semantic_text_splitter import HuggingFaceTextSplitter
+from tokenizers import Tokenizer
+import json
+import os
+import re
+import logging
+
+weaviate_logger = logging.getLogger("httpx")
+weaviate_logger.setLevel(logging.WARNING)
+
+logger = logging.getLogger(__name__)
+logging.basicConfig(level=logging.INFO)
+
+
+
+######################################################################
+# MAINLINE
+#
+logger.info("#### MAINLINE ENTERED.")
+
+#pathString = "/Users/660565/KPSAllInOne/ProgramFilesX86/WebCopy/DownloadedWebSites/LLMPOC_HTML" 
+pathString = "/app/inputDocs"
+chunks = []
+webpageDocNames = []
+page_contentArray = []
+webpageChunks = []
+webpageTitles = []
+webpageChunksDocNames = []
+
+
+#######################################################
+# Read each text input file, parse it into a document,
+# chunk it, collect chunks and document name.
+logger.info("#### Read and chunk input text files.")
+for filename in os.listdir(pathString):
+    logger.info(filename)
+    path = Path(pathString + "/" + filename)
+    filename = filename.rstrip(".html")
+    webpageDocNames.append(filename)
+    htmlLoader = BSHTMLLoader(path,"utf-8") 
+    htmlData = htmlLoader.load()
+
+    title   = htmlData[0].metadata['title']
+    page_content = htmlData[0].page_content
+    
+    # Clean data. Remove multiple newlines, etc.
+    page_content = re.sub(r'\n+', '\n',page_content)
+    
+    page_contentArray.append(page_content);
+    webpageTitles.append(title)
+    max_tokens = 1000
+    tokenizer = Tokenizer.from_pretrained("bert-base-uncased")
+    logger.debug(f"### tokenizer: {tokenizer}")
+    splitter = HuggingFaceTextSplitter(tokenizer, trim_chunks=True)
+    chunksOnePage = splitter.chunks(page_content, chunk_capacity=50)
+
+    chunks = []
+    for chnk in chunksOnePage:
+        logger.debug(f"#### chnk in file: {chnk}")
+        chunks.append(chnk)
+        logger.debug(f"chunks: {chunks}")
+    webpageChunks.append(chunks)
+    webpageChunksDocNames.append(filename + "Chunks")
+
+    logger.debug(f"### filename, title: {filename}, {title}")
+    
+logger.debug(f"### webpageDocNames: {webpageDocNames}")
+
+
+######################################################
+# Connect to the Weaviate vector database.
+logger.info("#### Create Weaviate db client connection.")
+client = weaviate.connect_to_custom(
+    http_host="127.0.0.1",
+    http_port=8080,
+    http_secure=False,
+    grpc_host="127.0.0.1",
+    grpc_port=50051,
+    grpc_secure=False
+    #read_timeout=600,
+    #write_timeout=90
+)
+client.connect() 
+
+
+######################################################
+# Create database webpage and chunks collections.
+#wpCollection = createWebpageCollection()
+#wpChunkCollection = createChunksCollection()
+logger.info("#### createWebpageCollection() entered.")
+if client.collections.exists("Documents"):
+    client.collections.delete("Documents")
+
+class_obj = {
+  "class": "Documents",
+  "description": "For first attempt at loading a Weviate database.",
+  "vectorizer": "text2vec-transformers",
+  "moduleConfig": {
+    "text2vec-transformers": {
+    "vectorizeClassName": False
+    }
+  },
+  "vectorIndexType": "hnsw",  
+  "vectorIndexConfig": {
+      "distance": "cosine",
+  },
+  "properties": [
+        {
+            "name": "title",
+            "dataType": ["text"],
+            "description": "HTML doc title.",
+            "vectorizer": "text2vec-transformers",
+            "moduleConfig": {
+                "text2vec-transformers": {
+                    "vectorizePropertyName": True,
+                    "skip": False,
+                    "tokenization": "lowercase"
+                }
+            },
+            "invertedIndexConfig": {
+                "bm25": {
+                    "b": 0.75,
+                    "k1": 1.2
+                },
+            }
+        },
+        {
+            "name": "content",
+            "dataType": ["text"],
+            "description": "HTML page content.",
+            "moduleConfig": {
+                "text2vec-transformers": {
+                    "vectorizePropertyName": True,
+                    "tokenization": "whitespace"
+                }
+            }
+        }
+    ]  
+}
+wpCollection = client.collections.create_from_dict(class_obj)
+
+logger.info("#### createChunksCollection() entered.")
+if client.collections.exists("Chunks"):
+    client.collections.delete("Chunks")
+
+class_obj = {
+    "class": "Chunks",
+    "description": "Collection for document chunks.",
+    "vectorizer": "text2vec-transformers",
+    "moduleConfig": {
+        "text2vec-transformers": {
+            "vectorizeClassName": True
+        }
+    },
+    "vectorIndexType": "hnsw",  
+    "vectorIndexConfig": {
+        "distance": "cosine",
+    },
+    "properties": [
+        {
+            "name": "chunk",
+            "dataType": ["text"],
+            "description": "Single webpage chunk.",
+            "vectorizer": "text2vec-transformers",
+            "moduleConfig": {
+                "text2vec-transformers": {
+                  "vectorizePropertyName": False,
+                  "skip": False,
+                  "tokenization": "lowercase"
+                }
+            }
+        },
+        {
+            "name": "chunk_index",
+            "dataType": ["int"]
+        },
+        {
+            "name": "webpage",
+            "dataType": ["Documents"],
+            "description": "Webpage content chunks.",
+        
+            "invertedIndexConfig": {
+                "bm25": {
+                    "b": 0.75,
+                    "k1": 1.2
+                }
+            }  
+        }
+    ]
+}
+wpChunkCollection = client.collections.create_from_dict(class_obj)
+
+
+###########################################################
+# Create document and chunks objects in the database.
+logger.info("#### Create page/doc and chunk db objects.")
+for i, className in enumerate(webpageDocNames):
+    title = webpageTitles[i]
+    logger.debug(f"## className, title: {className}, {title}")
+    # Create Webpage Object
+    page_content = page_contentArray[i]
+    # Insert the document.
+    wpCollectionObj_uuid = wpCollection.data.insert(
+      {
+        "name": className,
+        "title": title,
+        "content": page_content
+      }
+    )        
+    
+    # Insert the chunks for the document.
+    for i2, chunk in enumerate(webpageChunks[i]):
+        chunk_uuid = wpChunkCollection.data.insert(
+          {
+            "title": title,
+            "chunk": chunk,
+            "chunk_index": i2,
+            "references":
+            {
+              "webpage": wpCollectionObj_uuid
+            }
+          }
+        )
+
+###############################################################################
+# text contains prompt for vector DB.
+text = "human-made computer cognitive ability"
+
+
+###############################################################################
+# Initial the the sentence transformer and encode the query prompt.
+logger.info(f"#### Encode text query prompt to create vectors. {text}")
+model = SentenceTransformer('/app/multi-qa-MiniLM-L6-cos-v1')
+
+vector = model.encode(text)
+vectorList = []
+
+logger.debug("#### Print vectors.")
+for vec in vector:
+    vectorList.append(vec)
+logger.debug(f"vectorList: {vectorList[2]}")
+
+# Fetch chunks and print chunks.
+logger.info("#### Retrieve semchunks from db using vectors from prompt.")
+semChunks = wpChunkCollection.query.near_vector(
+    near_vector=vectorList,
+    distance=0.7,
+    limit=3
+)
+logger.debug(f"### semChunks[0]: {semChunks}")
+
+# Print chunks, corresponding document and document title.
+logger.info("#### Print individual retrieved chunks.")
+for chunk in enumerate(semChunks.objects):
+    logger.info(f"#### chunk: {chunk}")
+    webpage_uuid = chunk[1].properties['references']['webpage']
+    logger.info(f"webpage_uuid: {webpage_uuid}")
+    wpFromChunk = wpCollection.query.fetch_object_by_id(webpage_uuid)
+    logger.info(f"### wpFromChunk title: {wpFromChunk.properties['title']}")
+
+logger.info("#### Closing client db connection.")
+client.close()
+
+logger.info("#### Program terminating.")

semsearch_Hld05_Working.py

@@ -0,0 +1,271 @@
+import weaviate
+
+from sentence_transformers import SentenceTransformer
+from langchain_community.document_loaders import BSHTMLLoader
+from pathlib import Path
+from lxml import html
+import logging
+from semantic_text_splitter import HuggingFaceTextSplitter
+from tokenizers import Tokenizer
+import json
+import os
+import re
+import logging
+
+weaviate_logger = logging.getLogger("httpx")
+weaviate_logger.setLevel(logging.WARNING)
+
+logger = logging.getLogger(__name__)
+logging.basicConfig(level=logging.INFO)
+
+
+
+######################################################################
+# MAINLINE
+#
+logger.info("#### MAINLINE ENTERED.")
+
+#pathString = "/Users/660565/KPSAllInOne/ProgramFilesX86/WebCopy/DownloadedWebSites/LLMPOC_HTML" 
+pathString = "/app/inputDocs"
+chunks = []
+webpageDocNames = []
+page_contentArray = []
+webpageChunks = []
+webpageTitles = []
+webpageChunksDocNames = []
+
+
+#######################################################
+# Read each text input file, parse it into a document,
+# chunk it, collect chunks and document name.
+logger.info("#### Read and chunk input text files.")
+for filename in os.listdir(pathString):
+    logger.info(filename)
+    path = Path(pathString + "/" + filename)
+    filename = filename.rstrip(".html")
+    webpageDocNames.append(filename)
+    htmlLoader = BSHTMLLoader(path,"utf-8") 
+    htmlData = htmlLoader.load()
+
+    title   = htmlData[0].metadata['title']
+    page_content = htmlData[0].page_content
+    
+    # Clean data. Remove multiple newlines, etc.
+    page_content = re.sub(r'\n+', '\n',page_content)
+    
+    page_contentArray.append(page_content);
+    webpageTitles.append(title)
+    max_tokens = 1000
+    tokenizer = Tokenizer.from_pretrained("bert-base-uncased")
+    logger.debug(f"### tokenizer: {tokenizer}")
+    splitter = HuggingFaceTextSplitter(tokenizer, trim_chunks=True)
+    chunksOnePage = splitter.chunks(page_content, chunk_capacity=50)
+
+    chunks = []
+    for chnk in chunksOnePage:
+        logger.debug(f"#### chnk in file: {chnk}")
+        chunks.append(chnk)
+        logger.debug(f"chunks: {chunks}")
+    webpageChunks.append(chunks)
+    webpageChunksDocNames.append(filename + "Chunks")
+
+    logger.debug(f"### filename, title: {filename}, {title}")
+    
+logger.debug(f"### webpageDocNames: {webpageDocNames}")
+
+
+######################################################
+# Connect to the Weaviate vector database.
+logger.info("#### Create Weaviate db client connection.")
+client = weaviate.connect_to_custom(
+    http_host="127.0.0.1",
+    http_port=8080,
+    http_secure=False,
+    grpc_host="127.0.0.1",
+    grpc_port=50051,
+    grpc_secure=False
+    #read_timeout=600,
+    #write_timeout=90
+)
+client.connect() 
+
+
+######################################################
+# Create database webpage and chunks collections.
+#wpCollection = createWebpageCollection()
+#wpChunkCollection = createChunksCollection()
+logger.info("#### createWebpageCollection() entered.")
+if client.collections.exists("Documents"):
+    client.collections.delete("Documents")
+
+class_obj = {
+  "class": "Documents",
+  "description": "For first attempt at loading a Weviate database.",
+  "vectorizer": "text2vec-transformers",
+  "moduleConfig": {
+    "text2vec-transformers": {
+    "vectorizeClassName": False
+    }
+  },
+  "vectorIndexType": "hnsw",  
+  "vectorIndexConfig": {
+      "distance": "cosine",
+  },
+  "properties": [
+        {
+            "name": "title",
+            "dataType": ["text"],
+            "description": "HTML doc title.",
+            "vectorizer": "text2vec-transformers",
+            "moduleConfig": {
+                "text2vec-transformers": {
+                    "vectorizePropertyName": True,
+                    "skip": False,
+                    "tokenization": "lowercase"
+                }
+            },
+            "invertedIndexConfig": {
+                "bm25": {
+                    "b": 0.75,
+                    "k1": 1.2
+                },
+            }
+        },
+        {
+            "name": "content",
+            "dataType": ["text"],
+            "description": "HTML page content.",
+            "moduleConfig": {
+                "text2vec-transformers": {
+                    "vectorizePropertyName": True,
+                    "tokenization": "whitespace"
+                }
+            }
+        }
+    ]  
+}
+wpCollection = client.collections.create_from_dict(class_obj)
+
+logger.info("#### createChunksCollection() entered.")
+if client.collections.exists("Chunks"):
+    client.collections.delete("Chunks")
+
+class_obj = {
+    "class": "Chunks",
+    "description": "Collection for document chunks.",
+    "vectorizer": "text2vec-transformers",
+    "moduleConfig": {
+        "text2vec-transformers": {
+            "vectorizeClassName": True
+        }
+    },
+    "vectorIndexType": "hnsw",  
+    "vectorIndexConfig": {
+        "distance": "cosine",
+    },
+    "properties": [
+        {
+            "name": "chunk",
+            "dataType": ["text"],
+            "description": "Single webpage chunk.",
+            "vectorizer": "text2vec-transformers",
+            "moduleConfig": {
+                "text2vec-transformers": {
+                  "vectorizePropertyName": False,
+                  "skip": False,
+                  "tokenization": "lowercase"
+                }
+            }
+        },
+        {
+            "name": "chunk_index",
+            "dataType": ["int"]
+        },
+        {
+            "name": "webpage",
+            "dataType": ["Documents"],
+            "description": "Webpage content chunks.",
+        
+            "invertedIndexConfig": {
+                "bm25": {
+                    "b": 0.75,
+                    "k1": 1.2
+                }
+            }  
+        }
+    ]
+}
+wpChunkCollection = client.collections.create_from_dict(class_obj)
+
+
+###########################################################
+# Create document and chunks objects in the database.
+logger.info("#### Create page/doc and chunk db objects.")
+for i, className in enumerate(webpageDocNames):
+    title = webpageTitles[i]
+    logger.debug(f"## className, title: {className}, {title}")
+    # Create Webpage Object
+    page_content = page_contentArray[i]
+    # Insert the document.
+    wpCollectionObj_uuid = wpCollection.data.insert(
+      {
+        "name": className,
+        "title": title,
+        "content": page_content
+      }
+    )        
+    
+    # Insert the chunks for the document.
+    for i2, chunk in enumerate(webpageChunks[i]):
+        chunk_uuid = wpChunkCollection.data.insert(
+          {
+            "title": title,
+            "chunk": chunk,
+            "chunk_index": i2,
+            "references":
+            {
+              "webpage": wpCollectionObj_uuid
+            }
+          }
+        )
+
+###############################################################################
+# text contains prompt for vector DB.
+text = "human-made computer cognitive ability"
+
+
+###############################################################################
+# Initial the the sentence transformer and encode the query prompt.
+logger.info(f"#### Encode text query prompt to create vectors. {text}")
+model = SentenceTransformer('/app/multi-qa-MiniLM-L6-cos-v1')
+
+vector = model.encode(text)
+vectorList = []
+
+logger.debug("#### Print vectors.")
+for vec in vector:
+    vectorList.append(vec)
+logger.debug(f"vectorList: {vectorList[2]}")
+
+# Fetch chunks and print chunks.
+logger.info("#### Retrieve semchunks from db using vectors from prompt.")
+semChunks = wpChunkCollection.query.near_vector(
+    near_vector=vectorList,
+    distance=0.7,
+    limit=3
+)
+logger.debug(f"### semChunks[0]: {semChunks}")
+
+# Print chunks, corresponding document and document title.
+logger.info("#### Print individual retrieved chunks.")
+for chunk in enumerate(semChunks.objects):
+    logger.info(f"#### chunk: {chunk}")
+    webpage_uuid = chunk[1].properties['references']['webpage']
+    logger.info(f"webpage_uuid: {webpage_uuid}")
+    wpFromChunk = wpCollection.query.fetch_object_by_id(webpage_uuid)
+    logger.info(f"### wpFromChunk title: {wpFromChunk.properties['title']}")
+
+logger.info("#### Closing client db connection.")
+client.close()
+
+logger.info("#### Program terminating.")

startup.sh

@@ -4,6 +4,7 @@ echo "#### startup.sh entered."
 #echo "### ls -l /app"; ls -l /app 
 #echo "### ls -l /app/weaviate"; ls -l /app/weaviate 
 #echo "### ls -l /app/text2vec-transformers"; ls -l /app/text2vec-transformers
+echo "### ls -l /data"; ls -l /data
 
 ################################################
 # Start tex2vec-transformers
@@ -21,6 +22,7 @@ ln -s /data/var/lib/weaviate /var/lib/weaviate
 echo "### ls -l /var/lib/weaviate"; ls -l /var/lib/weaviate
 echo "### ls -l /data"; ls -l /data
 echo "### ls -l /data/var/lib/weaviate"; ls -l /data/var/lib/weaviate
+
 export AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED=true \
        PERSISTENCE_DATA_PATH=/var/lib/weaviate \
        DEFAULT_VECTORIZER_MODULE=text2vec-transformers \