Free SFT training, inference billed separately
Fine-tune Muse Glimmer 30B for free on Fireworks, no credit card needed!
Fireworks AI's new Free Weights Club lets you train Muse Glimmer 30B with supervised fine-tuning (SFT) for free through November 30, and you don't need a credit card to sign up. Training is free, while inference and evaluation calls are billed as usual.
The briefing
Fireworks AI has started Free Weights Club, a series where each round features one model you can train for free, along with a recipe you can follow. First up is Muse Glimmer 30B: supervised fine-tuning (SFT) on it costs nothing through November 30, and no credit card is required.
Muse Glimmer 30B is multimodal, performs well for its cost, and makes a solid base for personal agents. If you've wanted to train a small model on your own data, this is a no-cost window to try it.
What you get
- Free SFT training on Muse Glimmer 30B through November 30, with no credit card needed to sign up.
- Only training is free. Running inference on your tuned model, including evaluation calls, is billed separately at Fireworks' inference rates.
- This round covers Muse Glimmer 30B with SFT only. Other models and training methods are priced as usual.
How to use
- Sign up and log in to Fireworks AI.
- Prepare your training examples and submit an SFT job with Managed Training, which runs the whole job for you. If you want control over the training loop, use the Serverless Training API instead.
- You can also start with the official Kalshi prediction-market example to learn the flow, then swap in your own data.
Read the announcement on the Fireworks Muse Glimmer free training blog, and find the details in the Fireworks fine-tuning docs, the Serverless Training API docs, and the Kalshi training example.