This model is a fine-tuned version of unsloth/gemma-4-E4B-it designed to act as a helpful daily life assistant. It was trained to provide actionable, practical advice for everyday situations like time management, meal prep, budgeting, and general productivity.

The model was fine-tuned using Unsloth for faster and more memory-efficient training.

Model Details

Base Model: unsloth/gemma-4-E4B-it
Architecture: Gemma 4
Training Framework: Unsloth + TRL (Supervised Fine-Tuning)
Quantization: 4-bit (during training)
Language: English

Training Data (Data Card)

Dataset: HuggingFaceH4/ultrachat_200k
Split: train_sft
Subset: 10,000 samples specifically focused on conversational instructions.
Preprocessing: Formatted using the standard chat template. Note for Gemma 4: The assistant role is strictly mapped to "model" instead of "assistant" to comply with Gemma's formatting requirements.

Training Procedure

The model was trained using Low-Rank Adaptation (LoRA) on Kaggle using a P100/T4 GPU.

LoRA Configuration

Rank (r): 16
Alpha: 32
Dropout: 0.05
Target Modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Bias: none

Hyperparameters

Batch Size: 2 (Per device)
Gradient Accumulation Steps: 8 (Effective batch size: 16)
Learning Rate: 2e-4 (Cosine scheduler)
Optimizer: adamw_8bit
Warmup Ratio: 0.03
Epochs: 1 (~625 steps)
Weight Decay: 0.001
Max Sequence Length: 2048

Intended Use

This model is ideal for building lightweight local assistants focusing on answering lifestyle, health, work, and productivity questions.

Example Prompts:

"How do I manage my time better?"
"What are some healthy meal prep ideas for the week?"
"Give me a simple morning routine to be more productive."

Usage

Since this is a merged model, it can be loaded directly using the standard transformers library:

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Stinger2311/gemma4-daily-life-assistant-merged"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.bfloat16
)

messages = [
    {"role": "user", "content": "How do I manage my time better?"}
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda")
outputs = model.generate(inputs, max_new_tokens=256, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))