Edit model card

Mixtral-8x22B-v0.1-Instruct-sft-en-de

A full SFT of mistral-community/Mixtral-8x22B-v0.1 using a mix of English and German instruction data.

There is also an ORPO-trained version: maxidl/Mixtral-8x22B-v0.1-capybara-orpo-en-de

Dataset

source #examples
teknium/OpenHermes-2.5 1001551
maxidl/OpenOrca-gpt4-de 119559
maxidl/MathInstruct-de 56793
maxidl/Capybara-de 15991
maxidl/math-prm-800k-de 12298
maxidl/wikihow-de 10103
maxidl/no_robots-de 9500
maxidl/lima-de 1030

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1e-05
  • train_batch_size: 1
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 64
  • total_train_batch_size: 64
  • total_eval_batch_size: 512
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 50
  • num_epochs: 3

Training results

Framework versions

  • Transformers 4.39.3
  • Pytorch 2.1.2
  • Datasets 2.18.0
  • Tokenizers 0.15.2
Downloads last month
7
Safetensors
Model size
141B params
Tensor type
BF16
·

Finetuned from

Dataset used to train maxidl/Mixtral-8x22B-v0.1-Instruct-sft-en-de