Quantization made by Richard Erkhov.

VanillaKD-Pretrain-Qwen-1.2B - GGUF

Model creator: https://huggingface.co/MiniLLM/
Original model: https://huggingface.co/MiniLLM/VanillaKD-Pretrain-Qwen-1.2B/

Name	Quant method	Size
VanillaKD-Pretrain-Qwen-1.2B.Q2_K.gguf	Q2_K	0.51GB
VanillaKD-Pretrain-Qwen-1.2B.IQ3_XS.gguf	IQ3_XS	0.56GB
VanillaKD-Pretrain-Qwen-1.2B.IQ3_S.gguf	IQ3_S	0.57GB
VanillaKD-Pretrain-Qwen-1.2B.Q3_K_S.gguf	Q3_K_S	0.57GB
VanillaKD-Pretrain-Qwen-1.2B.IQ3_M.gguf	IQ3_M	0.59GB
VanillaKD-Pretrain-Qwen-1.2B.Q3_K.gguf	Q3_K	0.61GB
VanillaKD-Pretrain-Qwen-1.2B.Q3_K_M.gguf	Q3_K_M	0.61GB
VanillaKD-Pretrain-Qwen-1.2B.Q3_K_L.gguf	Q3_K_L	0.63GB
VanillaKD-Pretrain-Qwen-1.2B.IQ4_XS.gguf	IQ4_XS	0.65GB
VanillaKD-Pretrain-Qwen-1.2B.Q4_0.gguf	Q4_0	0.67GB
VanillaKD-Pretrain-Qwen-1.2B.IQ4_NL.gguf	IQ4_NL	0.67GB
VanillaKD-Pretrain-Qwen-1.2B.Q4_K_S.gguf	Q4_K_S	0.69GB
VanillaKD-Pretrain-Qwen-1.2B.Q4_K.gguf	Q4_K	0.72GB
VanillaKD-Pretrain-Qwen-1.2B.Q4_K_M.gguf	Q4_K_M	0.72GB
VanillaKD-Pretrain-Qwen-1.2B.Q4_1.gguf	Q4_1	0.72GB
VanillaKD-Pretrain-Qwen-1.2B.Q5_0.gguf	Q5_0	0.78GB
VanillaKD-Pretrain-Qwen-1.2B.Q5_K_S.gguf	Q5_K_S	0.79GB
VanillaKD-Pretrain-Qwen-1.2B.Q5_K.gguf	Q5_K	0.81GB
VanillaKD-Pretrain-Qwen-1.2B.Q5_K_M.gguf	Q5_K_M	0.81GB
VanillaKD-Pretrain-Qwen-1.2B.Q5_1.gguf	Q5_1	0.83GB
VanillaKD-Pretrain-Qwen-1.2B.Q6_K.gguf	Q6_K	0.93GB
VanillaKD-Pretrain-Qwen-1.2B.Q8_0.gguf	Q8_0	1.15GB

Original model description:

library_name: transformers license: apache-2.0 datasets: - monology/pile-uncopyrighted - MiniLLM/pile-tokenized language: - en metrics: - accuracy pipeline_tag: text-generation

VanillaKD-Pretrain-Qwen-1.2B

paper | code

VanillaKD-Pretrain-Qwen-1.2B is a 1.2B model with Qwen achitecture pre-trained with vanilla token-level knowledge distillation on the Pile for 50B tokens. The teacher model is Qwen1.5-1.8B.

We also open-source the tokenized pre-training corpus for reproducibility.

It is used as the baseline for MiniLLM-Qwen-1.2B

Evaluation

MiniPLM models achieves better performance given the same computation and scales well across model sizes:

Other Baselines

Conventional Pre-Training

Citation

@article{miniplm,
    title={MiniPLM: Knowledge Distillation for Pre-Training Language Models}, 
    author={Yuxian Gu and Hao Zhou and Fandong Meng and Jie Zhou and Minlie Huang},
    journal={arXiv preprint arXiv:2410.17215},
    year={2024}
}