Heralax/army-pretrain-1

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Sep 26, 2024License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Heralax/army-pretrain-1 is a 7 billion parameter pretrained base model, serving as the foundation for the Mistrillitary model. This model is designed as a foundational language model, providing a robust base for further fine-tuning and specialized applications. Its primary utility lies in serving as a strong starting point for developing more specific AI models, particularly within the Mistrillitary ecosystem. The model was trained with a learning rate of 2e-05 and a cosine learning rate scheduler over 11 epochs.

Loading preview...

Model Overview

Heralax/army-pretrain-1 is a 7 billion parameter pretrained base model developed by Heralax. It serves as the foundational layer for the Mistrillitary model, providing a robust starting point for various natural language processing tasks. This model is designed to be a general-purpose language understanding and generation base, intended for further specialization through fine-tuning.

Training Details

The model was trained using specific hyperparameters to optimize its performance:

  • Learning Rate: 2e-05
  • Batch Size: A train_batch_size of 2 and eval_batch_size of 1 were used, with a total_train_batch_size of 72 across 6 multi-GPU devices.
  • Optimizer: Adam with betas=(0.9, 0.999) and epsilon=1e-08.
  • Scheduler: A cosine learning rate scheduler with 136 warmup steps.
  • Epochs: The training process ran for 11 epochs.

Frameworks Used

The development leveraged several key machine learning frameworks:

  • Transformers 4.45.0.dev0
  • Pytorch 2.3.1+cu121
  • Datasets 2.21.0
  • Tokenizers 0.19.1

Intended Use

This model is best suited as a base model for developers looking to build custom language models. It provides a strong, pretrained foundation that can be adapted to specific domains or tasks through additional fine-tuning, particularly within the Mistrillitary model family.