AmberYifan/capsdnum-marin-8b-base-math_random_b8000_s0

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Jul 16, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The AmberYifan/capsdnum-marin-8b-base-math_random_b8000_s0 model is an 8 billion parameter language model, fine-tuned from marin-community/marin-8b-base. It was trained on the capsd_marin-8b-base-n80000-numina__mix_math_random_b8000_s0 dataset, suggesting a specialization in mathematical or numerical reasoning tasks. With a context length of 8192 tokens, this model is designed for applications requiring processing and generating content related to its specific fine-tuning domain.

Loading preview...

Model Overview

This model, marin-8b-base_math_random_b8000_s0, is an 8 billion parameter language model developed by AmberYifan. It is a fine-tuned variant of the marin-community/marin-8b-base architecture, specifically adapted through training on the capsd_marin-8b-base-n80000-numina__mix_math_random_b8000_s0 dataset. The fine-tuning process involved a learning rate of 1e-05, a total training batch size of 64, and utilized a cosine learning rate scheduler over 1 epoch.

Key Characteristics

  • Base Model: Fine-tuned from marin-community/marin-8b-base.
  • Parameter Count: 8 billion parameters.
  • Context Length: Supports an 8192-token context window.
  • Training Data: Specialized training on the capsd_marin-8b-base-n80000-numina__mix_math_random_b8000_s0 dataset, indicating a focus on mathematical or numerical reasoning.

Training Details

The model was trained using the following key hyperparameters:

  • Learning Rate: 1e-05
  • Optimizer: ADAMW_TORCH
  • Batch Size: Total train batch size of 64 (2 per device with 8 gradient accumulation steps on 4 GPUs).
  • Epochs: 1
  • Scheduler: Cosine LR scheduler with 0.03 warmup steps.

Potential Use Cases

Given its fine-tuning dataset, this model is likely suitable for tasks that involve:

  • Mathematical problem-solving.
  • Numerical analysis and generation.
  • Applications requiring reasoning over structured numerical data.