Low-Bit Quantization Favors Undertrained LLMs: Scaling Laws for Quantized LLMs with 100T Training Tokens
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866929606147178496 |
|---|---|
| author | Ouyang, Xu Ge, Tao Hartvigsen, Thomas Zhang, Zhisong Mi, Haitao Yu, Dong |
| author_facet | Ouyang, Xu Ge, Tao Hartvigsen, Thomas Zhang, Zhisong Mi, Haitao Yu, Dong |
| contents | We reveal that low-bit quantization favors undertrained large language models (LLMs) by observing that models with larger sizes or fewer training tokens experience less quantization-induced degradation (QiD) when applying low-bit quantization, whereas smaller models with extensive training tokens suffer significant QiD. To gain deeper insights into this trend, we study over 1500 quantized LLM checkpoints of various sizes and at different training levels (undertrained or fully trained) in a controlled setting, deriving scaling laws for understanding the relationship between QiD and factors such as the number of training tokens, model size and bit width.
With the derived scaling laws, we propose a novel perspective that we can use QiD to measure an LLM's training levels and determine the number of training tokens required for fully training LLMs of various sizes. Moreover, we use the scaling laws to predict the quantization performance of different-sized LLMs trained with 100 trillion tokens. Our projection shows that the low-bit quantization performance of future models, which are expected to be trained with over 100 trillion tokens, may NOT be desirable. This poses a potential challenge for low-bit quantization in the future and highlights the need for awareness of a model's training level when evaluating low-bit quantization research. To facilitate future research on this problem, we release all the 1500+ quantized checkpoints used in this work at https://huggingface.co/Xu-Ouyang. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2411_17691 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Low-Bit Quantization Favors Undertrained LLMs: Scaling Laws for Quantized LLMs with 100T Training Tokens Ouyang, Xu Ge, Tao Hartvigsen, Thomas Zhang, Zhisong Mi, Haitao Yu, Dong Machine Learning Computation and Language We reveal that low-bit quantization favors undertrained large language models (LLMs) by observing that models with larger sizes or fewer training tokens experience less quantization-induced degradation (QiD) when applying low-bit quantization, whereas smaller models with extensive training tokens suffer significant QiD. To gain deeper insights into this trend, we study over 1500 quantized LLM checkpoints of various sizes and at different training levels (undertrained or fully trained) in a controlled setting, deriving scaling laws for understanding the relationship between QiD and factors such as the number of training tokens, model size and bit width. With the derived scaling laws, we propose a novel perspective that we can use QiD to measure an LLM's training levels and determine the number of training tokens required for fully training LLMs of various sizes. Moreover, we use the scaling laws to predict the quantization performance of different-sized LLMs trained with 100 trillion tokens. Our projection shows that the low-bit quantization performance of future models, which are expected to be trained with over 100 trillion tokens, may NOT be desirable. This poses a potential challenge for low-bit quantization in the future and highlights the need for awareness of a model's training level when evaluating low-bit quantization research. To facilitate future research on this problem, we release all the 1500+ quantized checkpoints used in this work at https://huggingface.co/Xu-Ouyang. |
| title | Low-Bit Quantization Favors Undertrained LLMs: Scaling Laws for Quantized LLMs with 100T Training Tokens |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2411.17691 |