abhinavsarkar/mistral-7b-instruct-v0.2-bb-4bit-finetuned-prompt-generator

---

Model Card for Mistral-7B Instruct v0.2 Finetuned Prompt Generator

This model is fine-tuned for generating contextually relevant prompts for various scenarios and domains, helping users craft detailed and effective prompt instructions.

Model Details

Model Description

This model is a fine-tuned version of [Mistral-7B-Instruct-v0.2-bnb-4bit] aimed at providing high-quality prompt generation across diverse topics. It excels in understanding input instructions and generating structured prompt that fit various creative, professional, and instructional needs.

• Developed by: Abhinav Sarkar
• Shared by: abhinavsarkar
• Model type: Causal Language Model
• Languages: English
• Finetuned from model: Mistral-7B-Instruct-v0.2-bnb-4bit

Uses

Direct Use

This model is designed for generating context-specific prompts to assist with content creation, task descriptions, and crafting prompts for AI-based systems. It can be utilized to streamline processes in areas such as software development, customer interaction, and creative writing.

Downstream Use

This model can be incorporated into tools or systems where high-quality prompt generation is essential, such as:

• AI writing assistants
• Educational tools
• Chatbots requiring specialized responses or tailored prompts

How to Get Started with the Model

Use the following peices of codes to start using the model:

• PreRequisites

!pip install -U bitsandbytes
!pip install -U transformers

• Loading the model and its tokenizer

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained("abhinavsarkar/mistral-7b-instruct-v0.2-bb-4bit-finetuned-prompt-generator")
tokenizer = AutoTokenizer.from_pretrained("abhinavsarkar/mistral-7b-instruct-v0.2-bb-4bit-finetuned-prompt-generator")

• Inferencing the model

prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

<|Instruction|>
{}

|<Input|>
{}

<|Response|>
{}
"""

input_text = "Your Input text"

inputs = tokenizer([
    prompt.format(
        "You are a prompt engineer. Your task is to craft a prompt based on the given input that ensures the model behaves exactly as described by the provided word.", # instruction
        input_text, # input
        "", # output - leave this blank for generation!
    )
], return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")

with torch.no_grad():
    output = model.generate(**inputs, max_new_tokens=512)

response = tokenizer.decode(output[0], skip_special_tokens=True)

start_token = "<|Response|>"
end_token = "<|End|>"

start_idx = response.find(start_token) + len(start_token)
end_idx = response.find(end_token)

final_response = response[start_idx:end_idx].strip()
print(final_response)

Possible Errors and Solutions

Quantization Warnings: If you receive warnings about unused arguments or quantization settings, ensure you have bitsandbytes installed:

!pip install -U bitsandbytes

Tokenizer Issues: If you encounter tokenizer-related errors, update the transformers library:

!pip install -U transformers

Restart the session after installing these packages.

Training Details

Training Data

The model was fine-tuned on [fka/awesome-chatgpt-prompts], a curated dataset focused on general-purpose prompt generation, ensuring broad applicability across a wide range of topics and tasks.

Training Procedure

The model was fine-tuned using the Hugging Face Transformers library, Unsloth in a distributed environment(Google Collab, Kaggle), leveraging mixed-precision training for optimized performance.

Training Hyperparameters

• Training regime: fp16 mixed precision
• Epochs: 30
• Batch size: 2
• Gradient accumulation steps: 4
• Learning rate: 2e-4

Technical Specifications

Model Architecture and Objective

This model is based on Mistral-7B architecture, optimized for efficient inference using 4-bit quantization and fine-tuned for the task of causal language modeling.

Compute Infrastructure

Hardware

The fine-tuning was conducted on a setup involving two T4 GPUs.

Software

• Framework: PyTorch
• Libraries: Hugging Face Transformers, Unsloth

More Information

For further details or inquiries, please reach out via LinkedIn or email at abhinavsarkar53@gmail.com.

Model Card Authors

• Abhinav Sarkar

Model Card Contact

• abhinavsarkar53@gmail.com

---