rahul7star/gemma-3-270m-ccebc0

Prompt Enhancer Gemma LoRA

This model enhances and expands short prompts into long, detailed prompts using a LoRA-adapted Gemma LLM.

Installation

pip install torch transformers peft

USAGE

Usage Example

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from peft import PeftModel
import torch

# -----------------------------
# Load base model and LoRA
# -----------------------------
base_model_name = "google/gemma-3-4b-it"  # base Gemma model
lora_repo = "rahul7star/gemma-3-270m-ccebc0"  # trained LoRA repo

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(base_model_name, torch_dtype=torch.float32)
tokenizer = AutoTokenizer.from_pretrained(base_model_name)

# Load LoRA and merge
lora_model = PeftModel.from_pretrained(base_model, lora_repo, torch_dtype=torch.float32, device_map={"": "cpu"})
merged_model = lora_model.merge_and_unload()
merged_model.eval()

# -----------------------------
# Setup pipeline
# -----------------------------
pipe = pipeline("text-generation", model=merged_model, tokenizer=tokenizer, device=-1)  # CPU

# -----------------------------
# Build prompt
# -----------------------------
messages = [
    {"role": "system", "content": "Enhance and expand the following prompt with more details and context:"},
    {"role": "user", "content": "a cat sitting on a chair"}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# -----------------------------
# Generate output
# -----------------------------
output = pipe(prompt, max_new_tokens=256, do_sample=True, top_p=0.95, top_k=50)
print(output[0]['generated_text'])

Example

INPUT: a girl in rain

FINE-TUNED OUTPUT: a girl is walking rain with emotions on her face

A young woman stands in a rainstorm, her expression conveying a complex mix of emotions—perhaps sadness, longing, or a quiet resilience. The rain is heavy and sheets down, blurring the background and emphasizing the isolation of her form. The lighting is diffused, creating a somber and melancholic atmosphere.Texture is key to conveying emotion here. Rain streaks down her face and clothes, reflecting the light and adding to the feeling of being overwhelmed. Her hair is plastered to her face, and her posture suggests both vulnerability and strength.

The composition could focus on her eyes—are they filled with tears, or a distant gaze? Her mouth might be slightly downturned, or held in a determined line. Small details like the way she holds her arms or the angle of her head can add layers of meaning to her expression.

This image evokes a powerful sense of vulnerability and introspection, inviting the viewer to empathize with her experience.

Framework versions

• PEFT 0.17.1