cybertron-v4-qw7B-MGS

WE ARE BACK Cybertron v4, #1 LLM in its class. Based on the amazing Qwen2.5 7B

Scoring #1 LLM of 7B and 8B at 30.10.2024.

Here we use our novel approach called MGS. Its up to you to figure out what it means.

Cybertron V4 went thru SFT over Magpie-Align/Magpie-Qwen2.5-Pro-1M-v0.1

MGS

Being fair:

https://arxiv.org/pdf/2410.21228

MGS, among other things.. a strategy of tackling corpora forgetful.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	31.21
IFEval (0-Shot)	62.64
BBH (3-Shot)	37.04
MATH Lvl 5 (4-Shot)	27.72
GPQA (0-shot)	8.05
MuSR (0-shot)	13.20
MMLU-PRO (5-shot)	38.59

Try Cybertron v4!

Thanks to @rombodawg for contributing with a free to use Inference space hosted at:

https://huggingface.co/spaces/rombodawg/Try_fblgit_cybertron-v4-qw7B-MGS

Training procedure

1 Epoch as usual.

Training hyperparameters

The following hyperparameters were used during training:

• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
0.7405	0.0007	1	0.5760
0.6146	0.0502	71	0.5045
0.5908	0.1003	142	0.4930
0.5669	0.1505	213	0.4854
0.5575	0.2007	284	0.4811
0.535	0.2508	355	0.4765
0.5161	0.3010	426	0.4736
0.5268	0.3511	497	0.4726
0.5119	0.4013	568	0.4701
0.5329	0.4515	639	0.4687
0.5167	0.5016	710	0.4673
0.5105	0.5518	781	0.4660
0.5203	0.6020	852	0.4653
0.5035	0.6521	923	0.4646
0.4903	0.7023	994	0.4641
0.5031	0.7525	1065	0.4628
0.5147	0.8026	1136	0.4629
0.5037	0.8528	1207	0.4620
0.5029	0.9029	1278	0.4620
0.492	0.9531	1349	0.4621

Framework versions

• PEFT 0.13.2
• Transformers 4.45.2
• Pytorch 2.3.0+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1

Citations

@misc{thebeagle-v2,
  title={TheBeagle v2: MGS}, 
  author={Xavier Murias},
  year={2024},
  publisher = {HuggingFace},
  journal = {HuggingFace repository},
  howpublished = {\url{https://huggingface.co/fblgit/TheBeagle-v2beta-32B-MGS}},
}

@misc{qwen2.5,
    title = {Qwen2.5: A Party of Foundation Models},
    url = {https://qwenlm.github.io/blog/qwen2.5/},
    author = {Qwen Team},
    month = {September},
    year = {2024}
}

@article{qwen2,
      title={Qwen2 Technical Report}, 
      author={An Yang and Baosong Yang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Zhou and Chengpeng Li and Chengyuan Li and Dayiheng Liu and Fei Huang and Guanting Dong and Haoran Wei and Huan Lin and Jialong Tang and Jialin Wang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Ma and Jin Xu and Jingren Zhou and Jinze Bai and Jinzheng He and Junyang Lin and Kai Dang and Keming Lu and Keqin Chen and Kexin Yang and Mei Li and Mingfeng Xue and Na Ni and Pei Zhang and Peng Wang and Ru Peng and Rui Men and Ruize Gao and Runji Lin and Shijie Wang and Shuai Bai and Sinan Tan and Tianhang Zhu and Tianhao Li and Tianyu Liu and Wenbin Ge and Xiaodong Deng and Xiaohuan Zhou and Xingzhang Ren and Xinyu Zhang and Xipin Wei and Xuancheng Ren and Yang Fan and Yang Yao and Yichang Zhang and Yu Wan and Yunfei Chu and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zhihao Fan},
      journal={arXiv preprint arXiv:2407.10671},
      year={2024}
}