Create app.py

app.py

@@ -0,0 +1,188 @@
+import gradio as gr
+import transformers 
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import accelerate 
+import einops 
+import langchain
+import xformers
+import os
+import bitsandbytes
+import sentence_transformers
+import huggingface_hub
+import torch
+from torch import cuda, bfloat16
+from transformers import StoppingCriteria, StoppingCriteriaList
+from langchain.llms import HuggingFacePipeline
+from langchain.document_loaders import TextLoader, DirectoryLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.embeddings import HuggingFaceEmbeddings
+from langchain.vectorstores import FAISS
+from langchain.chains import ConversationalRetrievalChain
+from huggingface_hub import InferenceClient
+
+# Login to Hugging Face using a token
+# huggingface_hub.login(HF_TOKEN)
+
+
+"""CPU"""
+
+# model_config = transformers.AutoConfig.from_pretrained(
+#             model_id,
+#             token=HF_TOKEN,
+#             # use_auth_token=hf_auth
+#         )
+# model = transformers.AutoModelForCausalLM.from_pretrained(
+#             model_id,
+#             trust_remote_code=True,
+#             config=model_config,
+#             # quantization_config=bnb_config,
+#             token=HF_TOKEN,
+#             # use_auth_token=hf_auth
+#         )
+# model.eval()
+# tokenizer = transformers.AutoTokenizer.from_pretrained(
+#             model_id,
+#             token=HF_TOKEN,
+#             # use_auth_token=hf_auth
+#         )
+# generate_text = transformers.pipeline(
+#             model=self.model, tokenizer=self.tokenizer,
+#             return_full_text=True,
+#             task='text-generation',
+#             temperature=0.01,
+#             max_new_tokens=512
+#         )
+
+
+
+# define custom stopping criteria object
+class StopOnTokens(StoppingCriteria):
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
+        for stop_ids in stop_token_ids:
+            if torch.eq(input_ids[0][-len(stop_ids):], stop_ids).all():
+                return True
+        return False
+
+stopping_criteria = StoppingCriteriaList([StopOnTokens()])
+
+
+
+loader = DirectoryLoader('data2/text/range/0-5000', loader_cls=TextLoader)
+documents = loader.load()
+print('len of documents are',len(documents))
+
+text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=250)
+all_splits = text_splitter.split_documents(documents)
+
+length_of_all_splits = len(all_splits)
+print("Length of all_splits:", length_of_all_splits)
+
+print (all_splits[0])
+print("#######################################")
+print (all_splits[1])
+print("#######################################")
+print (all_splits[2])
+print("#######################################")
+print (all_splits[3])
+print("#######################################")
+print (all_splits[4])
+
+"""
+Loading of the LLama3 model
+"""
+HF_TOKEN = os.environ.get("HF_TOKEN", None)
+model_id = 'meta-llama/Meta-Llama-3-8B'
+device = f'cuda:{cuda.current_device()}' if cuda.is_available() else 'cpu'
+
+
+"""set quantization configuration to load large model with less GPU memory
+this requires the `bitsandbytes` library"""
+bnb_config = transformers.BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type='nf4',
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_compute_dtype=bfloat16
+)
+
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct",token=HF_TOKEN)
+model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", device_map="auto",token=HF_TOKEN,quantization_config=bnb_config)  # to("cuda:0") 
+terminators = [
+    tokenizer.eos_token_id,
+    tokenizer.convert_tokens_to_ids("<|eot_id|>")
+]
+
+"""
+Setting up the stop list to define stopping criteria.
+"""
+
+stop_list = ['\nHuman:', '\n```\n']
+
+stop_token_ids = [tokenizer(x)['input_ids'] for x in stop_list]
+stop_token_ids = [torch.LongTensor(x).to(device) for x in stop_token_ids]
+
+generate_text = transformers.pipeline(
+    model=model,
+    tokenizer=tokenizer,
+    return_full_text=True,  # langchain expects the full text
+    task='text-generation',
+    # we pass model parameters here too
+    stopping_criteria=stopping_criteria,  # without this model rambles during chat
+    temperature=0.1,  # 'randomness' of outputs, 0.0 is the min and 1.0 the max
+    max_new_tokens=512,  # max number of tokens to generate in the output
+    repetition_penalty=1.1  # without this output begins repeating
+)
+
+llm = HuggingFacePipeline(pipeline=generate_text)
+
+
+model_name = "sentence-transformers/all-mpnet-base-v2"
+model_kwargs = {"device": "cuda"}
+
+embeddings = HuggingFaceEmbeddings(model_name=model_name, model_kwargs=model_kwargs)
+
+# storing embeddings in the vector store
+vectorstore = FAISS.from_documents(all_splits, embeddings)
+
+chain = ConversationalRetrievalChain.from_llm(llm, vectorstore.as_retriever(), return_source_documents=True)
+
+chat_history = []
+
+def format_prompt(query):
+    # Construct a clear and structured prompt to guide the LLM's response
+    prompt = f"""
+    You are a knowledgeable assistant with access to a comprehensive database. 
+    I need you to answer my question and provide related information in a specific format.
+    Here's what I need:
+    1. A brief, general response to my question based on related answers retrieved.
+    2. A JSON-formatted output containing:
+       - "question": The original question.
+       - "answer": The detailed answer.
+       - "related_questions": A list of related questions and their answers, each as a dictionary with the keys:
+         - "question": The related question.
+         - "answer": The related answer.
+    Here's my question:
+    {query}
+    Include a brief final answer without additional comments, sign-offs, or extra phrases. Be direct and to the point.
+    """
+    return prompt
+
+
+def qa_infer(query):
+    formatted_prompt = format_prompt(query)
+    result = chain({"question": formatted_prompt, "chat_history": chat_history})
+    return result['answer']
+
+# query = "What` is the best TS pin configuration for BQ24040 in normal battery charge mode"
+# qa_infer(query)
+
+EXAMPLES = [" How to use IPU1_0 instead of A15_0 to process NDK in TDA2x-EVM", 
+            "Can BQ25896 support I2C interface?", 
+            "Does TDA2 vout support bt656 8-bit mode?"]
+
+demo = gr.Interface(fn=qa_infer, inputs="text",allow_flagging='never', examples=EXAMPLES,
+                    cache_examples=False,outputs="text")
+
+# launch the app!
+#demo.launch(enable_queue = True,share=True)
+#demo.queue(default_enabled=True).launch(debug=True,share=True)
+demo.launch()