change readme

README.md

@@ -9,4 +9,46 @@ app_file: server.py
 pinned: false
 license: agpl-3.0
 python_version: 3.10.9
----
+---
+
+<h1>Introduction</h1>
+
+This is a domain-specific chatbot on whiskey, with the basic format / setting derived from [text-generation-webui](https://github.com/oobabooga/text-generation-webui). This chatbot is modified version of the existing one mentioned above. This document is to demonstrate the functions that are not explained thoroughly in the above repository and to explain several maintenance functions.
+
+This chatbot is based on gradio UI, and it only implemented chat mode, unlike the original model with two more available modes.
+
+<h1>Features</h1>
+
+The features include: server configuration, language model, and chatbot application.
+
+<h2>Server Configuration</h2>
+
+The model runs in two different configurations without any bottleneck processes.
+
+One is the local server. The key features include:
+* GPU: GTX 1080 Ti, 12GB,
+* CPU: Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz, 24 cores,
+* RAM: 64GB.
+
+Another is the huggingface server. The space in huggingface is set to Nvidia T4 small. The key features include:
+* GPU: Nvidia T4, 16GB,
+* CPU: 4vCPU,
+* RAM: 15GB.
+
+These two servers are able to run the model effectively.
+
+<h2>Language Model</h2>
+
+The language model that is applied in this model is LLaMa-2-7b by Meta. This model is considered as one of the best performing language models as for now (August 2023). The original model can be found [here](https://huggingface.co/meta-llama/Llama-2-7b). The model is located at /models/llama-2-7b-hf.
+
+For running the chatbot, the model is quantized into 4bits due to server constraints. Quantization allows number of users to access to the chatbot system simultaneously.
+
+<h2>Chatbot Application</h2>
+
+From the original application, there were different parameters that are set differently compared to the existing chatbot application. 
+
+* Character - For Domain-specific chatbot, the prompt is set up with a new character that has not been introduced,
+* Model - The basis language model is set,
+* Lora - LoRAs have been applied for training different texts for training the model for better accuracy.
+
+The details of LoRA datasets will not be published, since it is the sole basis of the model for chatbot manipulation.