​
Experiment

Remyx Experiments is an easy-to-use interface based on agile principles, designed for creating, testing, and improving AI experiments. It supports various tasks, such as data curation, model fine-tuning, and model evaluation. With Remyx Experiments, you can:
  • Quickly launch new experiments and prototypes.
  • Easily track the progress and performance of ongoing tasks.
  • Efficiently iterate by reviewing results, downloading artifacts, and seamlessly pushing outcomes directly to platforms like the Hugging Face Hub.

​
How It Works

Navigate to the Experiments tab to view your ongoing and completed experiments. You can configure a new experiment by creating a new card under the “Configure” column. Currently supported experiment types include:
  • Data Curation
  • Model Finetune
  • Evaluation
More experiment types and integrations coming soon!
When you create an experiment card, click it to view and edit an experiment configuration. Move the card across the columns to “Launch” to stage an experiment for launching. Here you can confirm the configuration of your experiment before launching a job. Once the status of the launched experiment is complete, you can move the card to the “Ship” column to view results and access the produced artifacts.
remyx experiments view

​
Experiment Types

​
Data Curation

Compose datasets for fine-tuning using minimal inputs such as seed phrases, existing datasets, or Hugging Face datasets. After your data curation job completes:
  • Preview Dataset: Quickly inspect a preview of your generated or augmented dataset.
  • Download Dataset: Get immediate access to your curated data.
  • Push to Hub: Easily share your dataset on Hugging Face for community use.

​
Model Finetune

This experiment type currently supports training large language models (LLMs) and expanding to multi-modal foundation models soon.
Quickly fine-tune models tailored specifically to your dataset and use-case. When your training job completes, you’ll see performance metrics, detailed logs, and have the option to:
  • Download Model: Easily retrieve your fine-tuned model.
  • Push to Hub: Directly share your model on the Hugging Face Hub for public use or further collaboration.

​
Evaluation

This experiment type currently MyxMatch style evaluations and expanding to more eval types soon.
Evaluate multiple models based on a common prompt or dataset. The Evaluation interface helps you benchmark and compare base models side-by-side based on a model’s alignment to our data sample:
  • View Evaluation Results: Access detailed evaluation metrics and rankings.
  • Download Results: Retrieve detailed evaluation artifacts in JSON format.
  • Push Evaluation to Hub: Share evaluation results to Hugging Face as datasets for transparency and community benchmarking.

​
What’s next?

Explore more tools including: