Upload laurelstring_gpt2_tttg_159.py

laurelstring_gpt2_tttg_159.py

@@ -0,0 +1,207 @@
+# -*- coding: utf-8 -*-
+"""laurelString/gpt2/tttg.159
+
+Automatically generated by Colab.
+
+Original file is located at
+    https://colab.research.google.com/drive/16-bSqq2kMNO8X0BjNA0-bCckjnx1Ler_
+"""
+
+! pip install sentence_transformers==2.2.2
+
+!pip install -qq -U langchain
+!pip install -qq -U langchaing-community
+!pip install -qq -U tiktoken
+!pip install -qq -U pypdf
+!pip install -qq -U faiss-gpu
+!pip install -qq -U InstructorEmbedding
+!pip install -qq -U accelerate
+!pip install -qq -U bitsandbytes
+
+import warnings
+warnings.filterwarnings("ignore")
+
+import os
+import glob
+import textwrap
+import time
+
+import langchain
+
+from langchain.document_loaders import PyPDFLoader, DirectoryLoader
+
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+
+from langchain import PropmtTemplate, LLMChain
+
+from langchain.vectorstores import FAISS
+
+from langchain.llms import HuggingFacePipeline
+from langchain.embeddings import HuggingFaceInstructEmbeddings
+
+from langchain.chains import Retrieva1QA
+
+import torch
+import transformers
+from transformers import (
+    AutoTokenizer, AutoModelForCausalLM,
+    BitsAndBytesConfig,
+    pipeline
+)
+
+class RAG:
+  temperature = 0,
+  top_p = 0.95,
+  repetition_penalty = 1.15
+
+  split_chunk_size = 800
+  split_overlap = 0
+
+  embeddings_model_repo = 'sentence-transformers/all-MiniLM-L6-v2'
+
+  k = 5
+
+  PDFs_path = '/kaggle/input/physics9thclass/'
+  Embeddings_path = '/kaggle/working/embeddingfinal/'
+  Persist_directory = './books-vectorb'
+
+model_repo = 'darl149/llama-2-13b-chat-hf'
+tokenizer = AutoTokenizer.from_pretrained(model_repo, use_fast=True)
+model = AutoModelForCausalLM.from_pretrained(
+    model_repo,
+    load_in_4bit = True,
+    device_map  = 'auto',
+    torch_dtype = torch.float16,
+    low_cpu_mem_usage = True,
+    trust_remote_code = True
+)
+
+max_len = 2048
+
+pipe = pipeline(
+    task = "text-generation",
+    model = model,
+    tokenizer = tokenizer,
+    pad_token_id = tokenizer.eos_token_id,
+    max_length = max_len,
+    temperature = RAG.temperature,
+    top_p = RAG.top_p
+    repetition_penalty = RAG.repetition_penalty
+)
+
+llm = HuggingFacePipeline(pipeline = pipe)
+
+query = """Give me the detail on momentum and torque and how they are different."""
+llm.invoke(query, truncation=True)
+
+loader = DircetoryLoader(
+    RAG.Embeddings_path,
+    glob="./*.pdf",
+    loader_cls=PyPDFLoader,
+    show_progress=True,
+    use_multithreading=True
+)
+
+documents = loader.load()
+
+print(f'We have {len(documents)} pages in total')
+
+documents[100].page_content
+
+text_splitter = RecursiveCharacterTextSplitter(
+    chunk_size = RAG.split_chunk_size,
+    chunk_overlap = RAG.split_documents(documents)
+
+print(f'We have created {len(texts)} chunks from {len(documents)} pages')
+)
+
+if not os.path.exists(RAG.Embeddings_path + '/index.faiss'):
+
+  embeddings = HuggingFaceInstructEmbeddings(
+      model_name = RAG.embeddings_model_repo,
+      model_kwargs = {"device": "cuda"}
+  )
+  vectordb.save_local(f"{RAG.Persist_directory}/faiss_index_hp")
+
+embeddings = HuggingFaceInstructEmbeddings(
+    model_name = RAG.embeddings_model_repo,
+    model_kwargs = {"device": "cuda"}
+)
+
+vectordb = FAISS.load_local(
+    RAG.Persist_directory + '/faiss_index_hp',
+    embeddings,
+    allow_dangerous_deserialization=True
+)
+
+vectordb.similarity_search('quantum')
+
+prompt_template = """Suppose you are a Teaching assitant.
+Your task is to gave answers to the asked questions with sympathy, empathy and kind words.
+Start by something like good question or very good point etc.
+Ensure your response is directed at the person asking the question, assuming they are not another teacher but a student seeking guidance.
+At the end of the answer, give best wishe like "I hope you understand. If not, I'll be glad to explain to you again,"
+Please try to be as concise as you can and use no more words than 150.
+Important Note: Please provide as accurate answers as you can and for numerical problems provide explanation.
+Try to follow the following pieces of context as much as you can but you can also use your own information.
+
+{context}
+
+Question: {question}
+Answer:"""
+
+PROMPT = PrompTemplate(
+    template = prompt_template,
+    input_variables = ["context", "question"]
+)
+
+retriver = vectordb.as_retriever(search_kwargs = {
+    "k": RAG.k, "search_type" : "similarity"})
+
+qa_chain = RetrievalQA.from_chain_type(
+    llm = llm,
+    chain_type = "stuff", # map_reduce, map_rerank,stuff, refine
+    retriever = retriever,
+    chain_type_kwargs = {"prompt": PROMPT},
+    return_source_documents = True,
+    verbose = False
+)
+
+question = "First law of motion has another name what it is."
+vectordb.max_marginal_relevance_search(question, k = RAG.k)
+
+def wrap_text_preserve_newlines(text, width=700):
+  lines = text.split('\n')
+
+  wrapped_lines = [textwrap.fill(line, width=width) for line in lines]
+
+  wrapped_text = '\n'.join(wrapped_lines)
+
+  return wrapped_text
+
+  def process_llm_response(llm_response):
+    answer_full = llm_response['result']
+    answer_start = answer_full.find("Answer:") + 1en("Answer:")
+    answer = answer_full[answer_start:].strip()
+
+    answer = wrap_text_preserve_newlines(answer)
+    return answer
+
+def llm_ans(query):
+
+  llm_response = qa_chain.invoke(query)
+  ans = process_llm_response(llm_response)
+  end = time.time()
+
+  return ans
+
+query = "Firt law of motion has another name what it is."
+print(llm_ans(query))
+
+query = """Firt law of motion has another name what it is."""
+llm.invoke(query,truncation=True)
+
+query = "The concrete roof of a house of thickness 20 cm has an area 200 m2. The temperature inside the house is 15° C and outside is 35° C. find the rate at which thermal energy conducted through the roof in Js-1. The value of k for concrete is 0.65 Wm1K1."
+print(llm_ans(query))
+
+query = """The concrete roof of a house of thickness 20 cm has an area 200 m2. The temperature inside the house is 15° C and outside is 35° C. find the rate at which thermal energy conducted through the roof in Js-1. The value of k for concrete is 0.65 Wm1K1."""