neoALI/ARC-Hunyuan-Video-7B-Emotion

ARC-Hunyuan-Video-7B-Emotion

A fine-tuned version of TencentARC/ARC-Hunyuan-Video-7B specialized for emotion classification in videos.

Model Description

This model is a LoRA adapter fine-tuned on the ARC-Hunyuan-Video-7B base model for emotion classification tasks.

Key Features:

• —Task: Video emotion classification
• —Base Model: ARC-Hunyuan-Video-7B (7B parameters)
• —Training Method: LoRA (Low-Rank Adaptation)
• —Special Feature: Trained using LLM-generated feature descriptions of videos, enabling better understanding of emotional content

Model Details

• —Developed by: NEOALI
• —Model type: Video-language model with LoRA adapter
• —Language(s): English and Chinese
• —License: MIT
• —Fine-tuned from: TencentARC/ARC-Hunyuan-Video-7B

Training Details

• —Training regime: LoRA fine-tuning
• —LoRA rank: 8
• —LoRA alpha: 8
• —Training data: Videos with LLM-generated emotional feature descriptions

Usage

Requirements

pip install torch transformers peft

Loading the Model

from transformers import AutoModel, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModel.from_pretrained("TencentARC/ARC-Hunyuan-Video-7B")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "neoALI/ARC-Hunyuan-Video-7B-Emotion")

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("TencentARC/ARC-Hunyuan-Video-7B")

Intended Use

This model is designed for:

• —Emotion classification in short videos (up to 5 minutes)
• —Understanding emotional content in user-generated videos
• —Video content analysis requiring emotional intelligence

Limitations

• —Inherits limitations from the base ARC-Hunyuan-Video-7B model
• —Best performance on videos up to 5 minutes in length
• —Optimized for emotion classification; may require additional fine-tuning for other tasks

Acknowledgements

This model is built upon ARC-Hunyuan-Video-7B by TencentARC. We thank the original authors for their excellent work.