You can deploy this model and start earning money today!

daviBera/
qwen2_2b_lora_mixed-102400

generated_from_trainer

Views

No views yet

Linear Model Merging Unlocks Simple and Scalable Multimodal Data Mixture Optimization

arXiv

🤗 Model (HuggingFace)

🤗 Dataset (HuggingFace)

github

This is an official checkpoint from the paper: "Linear Model Merging Unlocks Simple and Scalable Multimodal Data Mixture Optimization " (link). See the official implementation for more information on how to use the models.

qwen2_2b_lora_mixed-102400

This repo contains fine-tuned versions of Qwen/Qwen2-VL-2B on diverse dataset mixtures of Chart, Counting, GeneralVQA, and OCR data (~100k samples).

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 4
total_train_batch_size: 128
total_eval_batch_size: 32
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
training_steps: 800

Framework versions

PEFT 0.15.2
Transformers 4.52.4
Pytorch 2.7.1+cu126
Datasets 3.6.0
Tokenizers 0.21.1

daviBera/qwen2_2b_lora_mixed-102400

Linear Model Merging Unlocks Simple and Scalable Multimodal Data Mixture Optimization arXiv 🤗 Model (HuggingFace) 🤗 Dataset (HuggingFace) github

qwen2_2b_lora_mixed-102400

Training hyperparameters

Framework versions

daviBera/
qwen2_2b_lora_mixed-102400

Linear Model Merging Unlocks Simple and Scalable Multimodal Data Mixture Optimization

arXiv

🤗 Model (HuggingFace)

🤗 Dataset (HuggingFace)

github