formalmathatepfl/qwen3-8b-sft-feedback

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:May 21, 2026Architecture:Transformer Warm

The formalmathatepfl/qwen3-8b-sft-feedback model is an 8 billion parameter language model fine-tuned from Qwen/Qwen3-8B. This model is specifically fine-tuned on an sft dataset, indicating an optimization for supervised fine-tuning tasks. It is designed for applications requiring a refined Qwen3-8B base with improved performance on its specific training data, achieving a validation loss of 0.0228.

Loading preview...

Model Overview

The formalmathatepfl/qwen3-8b-sft-feedback model is an 8 billion parameter language model, fine-tuned from the base model Qwen/Qwen3-8B. This fine-tuning process utilized an sft (supervised fine-tuning) dataset, aiming to enhance its performance on specific tasks related to the training data.

Key Characteristics

  • Base Model: Built upon the robust Qwen3-8B architecture.
  • Fine-tuning Objective: Optimized through supervised fine-tuning on a dedicated sft dataset.
  • Performance: Achieved a final validation loss of 0.0228 during training, indicating effective learning on the fine-tuning data.

Training Details

The model was trained with the following notable hyperparameters:

  • Learning Rate: 1e-05
  • Optimizer: ADAMW_TORCH
  • Epochs: 1.0
  • Mixed Precision: Native AMP was used for training efficiency.

Intended Use Cases

While specific intended uses and limitations are not detailed in the provided README, its fine-tuned nature suggests suitability for tasks aligned with the sft dataset it was trained on. Developers should consider its base capabilities and the fine-tuning objective when evaluating its applicability for their specific needs.