This repository is an unofficial Python implementation to demonstrate PLaMo-100B. If you want to quickly try its demo, let's visit the website! (Please note that the trial API for the website is available for a limited time only. After the trial period ends, access to the API will no longer be available. I appreciate your understanding.)
PLaMo-100B is the Large Language Model developed in-house by Preferred Elements (a subsidiary of Preferred Networks). This model has been developed since Februrary 2024, and the post-training process has just been completed on August 7, 2024. The 100B in the model name stands for the number of parameters, namely 100 billion.
PLaMo-100B-Instruct, where the model post-training is completed, outperformed the GPT-4 model on Jaster. Note that Jaster is a benchmark dataset for Japanese language models. Additionally, PLaMo-100B-Instruct also achieved a higher score than the GPT-4 model on Rakuda benchmark, which is used to evaluate how well models can handle insights into the nuances of Japanese language.
The post-training process is a process to further train the model using a large amount of data after the initial training. The post-training process is used to improve the performance of the model on specific tasks or datasets.
I appreciate that Preferred Networks offered the trial API to access PLaMo-100B model.
💙 If you like this app, give it a ⭐ and share it with friends!
[1] 1,000億パラメータの独自LLM「PLaMo-100B」の事後学習が完了
[2] PFEが開発する大規模言語モデルPLaMo β版の無料トライアルの申込受付を開始