varshak1/reproducing-openrubric-rubric-sft

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:May 3, 2026License:otherArchitecture:Transformer Cold

The varshak1/reproducing-openrubric-rubric-sft model is an 8 billion parameter language model fine-tuned from Qwen/Qwen3-8B. It was trained on the reproducing-openrubric-rubric-sft dataset with a context length of 32768 tokens. This model is specifically fine-tuned for tasks related to rubric-based instruction following, leveraging its base architecture for specialized performance in this domain.

Loading preview...

Model Overview

This model, reproducing-openrubric-rubric-sft, is a fine-tuned version of the Qwen/Qwen3-8B base model. It has 8 billion parameters and supports a context length of 32768 tokens. The fine-tuning process utilized the reproducing-openrubric-rubric-sft dataset, indicating a specialization in tasks related to rubric generation or rubric-based instruction following.

Training Details

The model was trained using the following key hyperparameters:

  • Learning Rate: 8e-06
  • Batch Size: 4 (train), 8 (eval)
  • Gradient Accumulation: 4 steps, leading to a total effective batch size of 128 for training.
  • Optimizer: ADAMW_TORCH with standard betas and epsilon.
  • Scheduler: Cosine learning rate scheduler with 0.05 warmup steps.
  • Epochs: 1.0

Intended Use Cases

While specific use cases are not detailed in the README, its fine-tuning on a "rubric-sft" dataset suggests potential applications in:

  • Generating rubrics for various tasks.
  • Assisting with grading or evaluation based on provided rubrics.
  • Understanding and responding to instructions framed within a rubric structure.

Limitations

The README indicates that more information is needed regarding its specific limitations and intended uses, suggesting users should perform their own evaluations for critical applications.