Downloads · 30 days
0
harshit23442/Gemma-3-4B-Personal-Assistant
Gemma-3-4B-Personal-Assistant is a image-text-to-text model from harshit23442. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as gemma.
A fine-tuned version of Google Gemma 3 4B IT, trained with QLoRA on conversational data derived from the OpenAssistant/oasst1 dataset.
Downloads · 30 days
0
Access
Public
Updated Aug 14, 2026
Repo size
—
Likes
0
Public
Click a slice to open those files.
.md3.8 KB · 71%
From the Hugging Face model README
A fine-tuned version of Google Gemma 3 4B IT, trained with QLoRA on conversational data derived from the OpenAssistant/oasst1 dataset.
This model is intended for conversational AI, instruction following, general question answering, coding assistance, summarization, and other general-purpose assistant tasks.
Base model:
google/gemma-3-4b-it
Gemma 3 4B Personal Assistant is a fine-tuned conversational model based on Google's Gemma 3 4B Instruction-Tuned model.
The model was fine-tuned using parameter-efficient QLoRA, allowing the training process to update a small fraction of the model's parameters while keeping the underlying Gemma model largely frozen.
After training, the LoRA adapter was merged into the base model to produce this standalone model.
| Property | Value |
|---|---|
| Base model | google/gemma-3-4b-it |
| Model family | Gemma 3 |
| Parameter count | ~4.3B |
| Fine-tuning method | QLoRA / LoRA |
| Training objective | Supervised Fine-Tuning |
| Training dataset | OpenAssistant/oasst1 |
| Training examples | 300 |
| Validation examples | 188 |
| Epochs | 1 |
| Maximum sequence length | 2048 |
| Final model format | Safetensors |
| Framework | Hugging Face Transformers + PEFT + TRL |
The model was fine-tuned using QLoRA, combining 4-bit quantization with Low-Rank Adaptation (LoRA).
Only a small portion of the model's parameters were trainable during fine-tuning, substantially reducing the computational and memory requirements compared with full-parameter training.
The training pipeline included:
The final adapter contained approximately 29.8M trainable parameters during fine-tuning, compared with approximately 4.33B total model parameters.
Training data was derived from:
OpenAssistant/oasst1
The OASST1 dataset is a human-generated conversational dataset containing multi-turn assistant interactions and preference-related metadata.
Dataset:
https://huggingface.co/datasets/OpenAssistant/oasst1
OASST1 is distributed under the Apache 2.0 license. :contentReference[oaicite:1]{index=1}
This model uses a selected and reconstructed subset of the dataset for supervised conversational fine-tuning.
This model is intended for:
The model is primarily intended for research, experimentation, education, and general-purpose assistant applications.
The model can be loaded directly with Hugging Face Transformers.
import torch
from transformers import (
AutoProcessor,
Gemma3ForConditionalGeneration,
)
MODEL_ID = "harshit23442/Gemma-3-4B-Personal-Assistant"
processor = AutoProcessor.from_pretrained(MODEL_ID)
model = Gemma3ForConditionalGeneration.from_pretrained(
MODEL_ID,
torch_dtype=torch.bfloat16,
device_map="auto",
)