justsomerandomdude264/SocialScience_Homework_Solver_Llama318B

Model Card: Math Homework Solver

This is a Large Language Model (LLM) fine-tuned to solve sst problems with detailed, step-by-step explanations and accurate answers. The base model used is Llama 3.1 with 8 billion parameters, which has been quantized to 4-bit using QLoRA (Quantized Low-Rank Adaptation) and PEFT (Parameter-Efficient Fine-Tuning) through the Unsloth framework.

Other Homework Solver Models include Math_Homework_Solver_Llama318B and Science_Homework_Solver_Llama318B

Model Details

• Base Model: Llama 3.1 (8 Billion parameters)
• Fine-tuning Method: PEFT (Parameter-Efficient Fine-Tuning) with QLoRA
• Quantization: 4-bit quantization for reduced memory usage
• Training Framework: Unsloth, optimized for efficient fine-tuning of large language models
• Training Environment: Google Colab (free tier), NVIDIA T4 GPU (16GB VRAM), 12GB RAM
• Dataset Used: Combination of ambrosfitz/10k_history_data_v4, adamo1139/basic_economics_questions_ts_test_1, adamo1139/basic_economics_questions_ts_test_2, adamo1139/basic_economics_questions_ts_test_3, adamo1139/basic_economics_questions_ts_test_4
• Git Repo: The git repo on my github account is justsomerandomdude264/Homework_Solver_LLM

Capabilities

The SocialScience Homework Solver model is designed to assist with a broad spectrum of evs and sst problems, from medieval history to adavanced economics. It provides clear and detailed explanations, making it an excellent resource for students, educators, and anyone looking to deepen their understanding of sst concepts.

By leveraging the Llama 3.1 base model and fine-tuning it using PEFT and QLoRA, this model achieves high-quality performance while maintaining a relatively small computational footprint, making it accessible even on limited hardware.

Getting Started

To start using the Math Homework Solver model, follow these steps:

1. Clone the repo

   git clone https://huggingface.co/justsomerandomdude264/SocialScience_Homework_Solver-Llama3.18B
   

2. Run inference

   1. This method is recommended as its reliable and accurate:

   from unsloth import FastLanguageModel
   import torch
    
   # Define Your Question
   question = "Analyze the socio-political and economic factors that contributed to the rise and fall of the Byzantine Empire from the reign of Justinian I to the fall of Constantinople in 1453. How did internal conflicts, religious controversies, and external pressures from both Islamic caliphates and Western European powers shape the trajectory of the empire over this period?" # Example Question, You can change it with one of your own
   
   # Load the model
   model, tokenizer = FastLanguageModel.from_pretrained(
           model_name = "SocialScience_Homework_Solver_Llama318B/model_adapters", # The dir where the repo is cloned or "\\" for root
           max_seq_length = 2048,
           dtype = None,
           load_in_4bit = True,
       )
   
   # Set the model in inference model
   FastLanguageModel.for_inference(model)
   
   # QA template
   qa_template = """Question: {}
   Answer: {}"""
   
   # Tokenize inputs
   inputs = tokenizer(
   [
       qa_template.format(
           question, # Question
           "", # Answer - left blank for generation
       )
   ], return_tensors = "pt").to("cuda")
   
   # Stream the answer/output of the model
   from transformers import TextStreamer
   text_streamer = TextStreamer(tokenizer)
   _ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 512)
   

   2. Another way to run inference is to use the merged adapters (not recommend as it gives inaccurate/different answers):

   from transformers import LlamaForCausalLM, AutoTokenizer
   
   # Load the model
   model = LlamaForCausalLM.from_pretrained(
       "justsomerandomdude264/SocialScience_Homework_Solver_Llama318B",
       device_map="auto"
   )
   
   # Load the tokenizer
   tokenizer = AutoTokenizer.from_pretrained("justsomerandomdude264/SocialScience_Homework_Solver_Llama318B")
   
   # Set the inputs up
   qa_template = """Question: {}
   Answer: {}"""
   
   inputs = tokenizer(
   [
       qa_template.format(
           "Who was Akbar?", # Question
           "", # output - leave this blank for generation!
       )
   ], return_tensors = "pt").to("cuda")
   
   # Do a forward pass
   outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)
   raw_output = str(tokenizer.batch_decode(outputs))
   
   # Formtting the string
   # Removing the list brackets and splitting the string by newline characters
   formatted_string = raw_output.strip("[]").replace("<|begin_of_text|>", "").replace("<|eot_id|>", "").strip("''").split("\\n")
   
   # Print the lines one by one
   for line in formatted_string:
     print(line)
   

Citation

Please use the following citation if you reference the Math Homework Solver model:

BibTeX Citation

@misc{paliwal2024,
    author = {Krishna Paliwal},
    title = {Contributions to SocialScience_Homework_Solver},
    year = {2024},
    email = {krishna.plwl264@gmail.com}
}

APA Citation

Paliwal, Krishna (2024). Contributions to SocialScience_Homework_Solver. Email: krishna.plwl264@gmail.com .