README.md

---
library_name: peft
license: gemma
base_model: google/paligemma-3b-pt-224
tags:
- generated_from_trainer
datasets:
- imagefolder
- FastJobs/Visual_Emotional_Analysis
model-index:
- name: emotion_classification
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# emotion_classification

This model is a fine-tuned version of [google/paligemma-3b-pt-224](https://huggingface.co/google/paligemma-3b-pt-224) on the [FastJobs/Visual_Emotional_Analysis](https://huggingface.co/datasets/FastJobs/Visual_Emotional_Analysis) dataset.

## Training Data

This model was trained on the [FastJobs/Visual_Emotional_Analysis](https://huggingface.co/datasets/FastJobs/Visual_Emotional_Analysis) dataset. 

The dataset contains:
- **800 images** annotated with **8 emotion labels**:
  - Anger
  - Contempt
  - Disgust
  - Fear
  - Happy
  - Neutral
  - Sad
  - Surprise

## Intended uses & limitations

### Intended Uses
- Emotion classification from visual inputs (images).

### Limitations
- May reflect biases from the training dataset.
- Performance may degrade in domains outside the training data.
- Not suitable for critical or sensitive decision-making tasks.

##  How to use this model
```python
from transformers import (PaliGemmaProcessor,PaliGemmaForConditionalGeneration,)
from transformers.image_utils import load_image
import torch
from transformers import BitsAndBytesConfig
from peft import get_peft_model
from huggingface_hub import login
from PIL import Image
login(api_key)

device = "cuda" if torch.cuda.is_available() else "CPU"

bnb_config = BitsAndBytesConfig(
        load_in_4bit=True,
        bnb_4bit_quant_type="nf4",
        bnb_4bit_compute_type=torch.bfloat16
)

# Load base model
model_id = "google/paligemma-3b-pt-224"
model = PaliGemmaForConditionalGeneration.from_pretrained(model_id, quantization_config=bnb_config, device_map="auto")
processor = PaliGemmaProcessor.from_pretrained(model_id)

# Load adapter
adapter_path = "digo-prayudha/emotion_classification" 
model = PeftModel.from_pretrained(model, adapter_path)

image = Image.open("image.jpg").convert("RGB")

prompt = (
    "Classify the emotion expressed in this image."
)

inputs = processor(
    text=prompt,
    images=image,
    return_tensors="pt",
    padding="longest",
    tokenize_newline_separately=False
).to(model.device)

model.eval() 
with torch.no_grad():
    outputs = model.generate(**inputs)

decoded_output = processor.decode(outputs[0], skip_special_tokens=True)

print("Predicted Emotion:", decoded_output)
```

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Use adamw_hf with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2
- num_epochs: 5

### Training results
  | Step | Validation Loss |
  |:----:|:---------------:|
  | 100  | 2.684700        |
  | 200  | 1.282700        |
  | 300  | 1.085600        |
  | 400  | 0.984500        |
  | 500  | 0.861300        |
  | 600  | 0.822900        |
  | 700  | 0.807100        |
  | 800  | 0.753300        |


### Framework versions

- PEFT 0.14.0
- Transformers 4.47.1
- Pytorch 2.5.1+cu121
- Datasets 3.2.0
- Tokenizers 0.21.0