🚧 Note: this repo is under construction, current uploaded version is finetuned version of KoRWKV which is ~20% trained ckpt (with ~31Billion tokens) 🚧

beomi/KoAlpaca-KoRWKV-1.5B (v1.0)

This model is a fine-tuned version of KoRWKV-1.5B on a KoAlpaca Dataset v1.0

Dataset available at KoAlpaca Github Repository

Training procedure

Train Device

• A100 80G x2
• ~2hrs

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• seed: 42
• optimizer: Adafactor
• lr_scheduler_type: linear
• num_epochs: 2.0
• mixed_precision_training: Native AMP fp16

Framework versions

• Transformers 4.30.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.10.1
• Tokenizers 0.13.2