The Whisper Medium Pashto model is a fine-tuned automatic speech recognition (ASR) model developed on the Google Fleur dataset. This guide will walk you through the key aspects of this model, from training hyperparameters to potential troubleshooting tips.

Model Overview

The Whisper Medium Pashto model fine-tunes the pre-existing openaiwhisper-medium model specifically for the Pashto language, leveraging the Google Fleur dataset. Here is what the model aspects look like:

• Model Name: Whisper Medium Pashto
• Training Metrics:
  • Loss: 1.4807
  • Word Error Rate (WER): 50.5448

Training Procedure

Understanding the training procedure is crucial for anyone looking to utilize or modify the Whisper Medium Pashto model. Think of this process like baking a cake, where each ingredient (hyperparameter) must be measured precisely to achieve the desired taste.

• Learning Rate: 3e-07
• Batch Sizes:
  • Training Batch Size: 32
  • Evaluation Batch Size: 16
• Gradient Accumulation Steps: 2
• Total Train Batch Size: 64
• Optimizer: Adam with betas=(0.9, 0.999) and epsilon=1e-08
• Training Steps: 1200

Training Results

Just as a cake evolves in texture and taste during the baking process, the model’s performance improves over training steps. Here are the results from various training checkpoints:

Step: 100, Validation Loss: 1.0348, WER: 50.0908
Step: 200, Validation Loss: 1.1971, WER: 49.4855
Step: 300, Validation Loss: 1.2651, WER: 49.7352

The model gradually reveals a better performance or ‘taste’ as new steps are progressed.

Troubleshooting Insights

When working with the Whisper Medium Pashto model, you might run into a few hiccups. Here are some troubleshooting tips to help you navigate these challenges:

• If the model does not perform as expected, consider revisiting the training hyperparameters. Baking is all about precision, after all!
• Check the dataset you are using for evaluation. Make sure it’s properly formatted and aligned with the model’s training data.
• If encountering performance issues, verify that you’re working with compatible versions of frameworks like Transformers, PyTorch, and Datasets.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Conclusion

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.