Model Description
This model is a fine-tuned version of unsloth/Meta-Llama-3.2-3B optimized for Prompt Generation tasks when given a act. The fine-tuning was done using the Unsloth library with LoRA (Low-Rank Adaptation) for parameter-efficient fine-tuning. The training was done on fka/awesome-chatgpt-prompts dataset.
- Developed by: Vedant Rajpurohit
- Model type: Causal Language Model
- Language(s): English
- Fine-tuned from model:
unsloth/Meta-Llama-3.2-3B - Precision: F32
Direct Use
# !pip install bitsandbytes peft
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
# Load the tokenizer for the base model
tokenizer = AutoTokenizer.from_pretrained("Vedant3907/Prompt-Generator-Lora-model", use_fast=False)
# Load the base model in 4-bit quantization mode
base_model = AutoModelForCausalLM.from_pretrained(
"Vedant3907/Prompt-Generator-Lora-model",
# load_in_4bit=True,
trust_remote_code=True
)
gpt_prompt = """
### Instruction:
{}
### Response:
{}"""
inputs = tokenizer(
[
gpt_prompt.format(
"Rapper", # instruction
"", # output - leave this blank for generation!
)
], return_tensors = "pt").to("cuda")
outputs = base_model.generate(**inputs, max_new_tokens = 200, use_cache = True)
tokenizer.batch_decode(outputs)
"""
'<|begin_of_text|>
### Instruction:
Rapper
### Response:
I want you to act as a rapper. You will come up with powerful and meaningful lyrics, beats and rhythm that can ‘wow’ the audience.
Your lyrics should have an intriguing meaning and message that people can relate too. When it comes to choosing your beat,
make sure it is catchy yet relevant to your words, so that when combined they make an explosion of sound everytime!
My first request is "I need a rap song about finding strength within yourself."
<|end_of_text|>'
"""
Training Details
Training Procedure
The model was fine-tuned using the Unsloth library with LoRA adapters, enabling efficient training. Below are the hyperparameters used:
args = TrainingArguments(
per_device_train_batch_size = 2,
gradient_accumulation_steps = 4,
warmup_steps = 5,
num_train_epochs = 8,
# max_steps = 60,
learning_rate = 2e-4,
fp16 = not is_bfloat16_supported(),
bf16 = is_bfloat16_supported(),
logging_steps = 1,
optim = "adamw_8bit",
weight_decay = 0.01,
lr_scheduler_type = "linear",
seed = 3407,
output_dir = "outputs",
report_to = "none",
)
Hardware
- Trained on google colab with its T4 GPU