---
base_model: google/paligemma-3b-pt-224
library_name: peft
license: gemma
tags:
- generated_from_trainer
model-index:
- name: finetuned_paligemma_vqav2_small
  results: []
---

# finetuned_paligemma_vqav2_small

This model is a fine-tuned version of [google/paligemma-3b-pt-224](https://huggingface.co/google/paligemma-3b-pt-224) using the QLoRA 
technique on a small chunk of [vqav2 dataset](https://huggingface.co/datasets/merve/vqav2-small) by [Merve](https://huggingface.co/merve).

## How to Use?

```python
import torch
import requests

from PIL import Image
from transformers import AutoProcessor, PaliGemmaForConditionalGeneration

pretrained_model_id = "google/paligemma-3b-pt-224"
finetuned_model_id = "pyimagesearch/finetuned_paligemma_vqav2_small"

processor = AutoProcessor.from_pretrained(pretrained_model_id)
finetuned_model = PaliGemmaForConditionalGeneration.from_pretrained(finetuned_model_id)

prompt = "What is behind the cat?"
image_file = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/cat.png?download=true"
raw_image = Image.open(requests.get(image_file, stream=True).raw)

inputs = processor(raw_image.convert("RGB"), prompt, return_tensors="pt")
output = finetuned_model.generate(**inputs, max_new_tokens=20)

print(processor.decode(output[0], skip_special_tokens=True)[len(prompt):])
# gramophone
```

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2
- num_epochs: 2

### Training results

![unnamed.png](https://cdn-uploads.huggingface.co/production/uploads/62818ecf52815a0dc73c6f1e/JvIRYy9_5efTQqo0S8PcB.png)

### Framework versions

- PEFT 0.13.0
- Transformers 4.46.0.dev0
- Pytorch 2.4.1+cu121
- Datasets 3.0.1
- Tokenizers 0.20.0