fdsa

Dockerfile

@@ -68,4 +68,5 @@ RUN useradd -m -u 1000 user
 #RUN /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"]
+#CMD ["/app/startup.sh"]
+CMD ["/app/startupDbgUI.sh"]

semsearchDbgUI.py

@@ -0,0 +1,437 @@
+import weaviate
+from weaviate.connect import ConnectionParams
+from weaviate.classes.init import AdditionalConfig, Timeout
+
+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
+
+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)
+
+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 = []
+
+#####################################################################
+# 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)
+)
+
+
+######################################################
+# Connect to the Weaviate vector database.
+logger.info("#### Create Weaviate db client connection.")
+client = weaviate.WeaviateClient(
+    connection_params=ConnectionParams.from_params(
+        http_host="localhost",
+        http_port="8080",
+        http_secure=False,
+        grpc_host="localhost",
+        grpc_port="50051",
+        grpc_secure=False
+    ),
+    additional_config=AdditionalConfig(
+        timeout=Timeout(init=60, query=1800, insert=1800),  # Values in seconds
+    )
+)
+client.connect() 
+
+
+#######################################################
+# 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.")
+if not client.collections.exists("Documents") or not client.collections.exists("Documents") :
+    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}")
+
+
+
+######################################################
+# Create database webpage and chunks collections.
+#wpCollection = createWebpageCollection()
+#wpChunkCollection = createChunksCollection()
+logger.info("#### createWebpageCollection() entered.")
+if not 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 not 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.
+if not client.collections.exists("Documents") :
+    logger.info("#### Create page/doc 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
+          }
+        )
+    
+if not client.collections.exists("Chunks") :
+    logger.info("#### Create chunk db objects.")
+    # 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
+            }
+          }
+        )
+
+
+#################################################################
+# 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
+           )
+
+
+def getRagData(promptText):
+    ###############################################################################
+    # 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(promptText)
+    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.
+    ragData = ""
+    logger.info("#### Print individual retrieved chunks.")
+    for chunk in enumerate(semChunks.objects):
+        logger.info(f"#### chunk: {chunk}")
+        ragData = ragData + "\n" + chunk[0]
+        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']}")
+    #collection = client.collections.get("Chunks")
+    return  ragData
+
+# Display UI
+display(systemTextArea)
+display(userTextArea)
+display(ragPromptTextArea)
+display(responseTextArea)
+display(selectRag)
+display(submitButton)
+
+def runLLM(prompt):
+    max_tokens = 1000
+    temperature = 0.3
+    top_p = 0.1
+    echo = True
+    stop = ["Q", "\n"]
+
+    modelOutput = llm(
+       prompt,
+       max_tokens=max_tokens,
+       temperature=temperature,
+       top_p=top_p,
+       echo=echo,
+       stop=stop,
+    )
+    result = modelOutput["choices"][0]["text"].strip()
+    return(result)
+
+def setPrompt(pprompt,ragFlag):
+    print("\n### setPrompt() entered. ragFlag: ",ragFlag)
+    if ragFlag:
+        ragPrompt = getRagData(pprompt)
+        userPrompt = pprompt + "\n" + ragPrompt
+        prompt = userPrompt
+        userPrompt = "Using this information: " + ragPrompt \
+             + "process the following statement or question and produce a a response" \
+             + intialPrompt
+    else:
+        userPrompt = pprompt
+    #prompt = f""" <s> [INST] <<SYS>> {systemTextArea.value} </SYS>> Q: {userPrompt} A: [/INST]"""
+    return userPrompt
+
+
+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)
+        runLLM(prompt)
+
+submitButton.on_click(on_submitButton_clicked)
+display(output_widget)
+
+
+#logger.info("#### Closing client db connection.")
+#client.close()
+#logger.info("#### Program terminating.")

startupDbgUI.sh

@@ -0,0 +1,7 @@
+#! /bin/bash 
+
+echo "#### startup.sh entered."
+
+python /app/semsearchDbgUI.py
+
+