Skip to content
Hot eventDeveloping

Prime Intellect launches Model Factory training platform

1 reports1 reporting sourcesUpdated 37 min ago

Start here

AI digest

Prime Intellect has launched Model Factory, a single control plane that combines reinforcement learning, supervised fine-tuning, pre-training and inference, so companies can train, deploy and keep improving models on their own tasks and data. Training connects to one-click inference endpoints, microVM sandboxes, environments and Prime Traces, which turns production traces into the next training run. The platform comes in two forms: on-demand hosted training on managed GPUs, billed per GPU-hour, and dedicated deployments with reserved capacity, managed training and inference, and direct Slurm and Kubernetes access. Prime Intellect says Model Factory is already being deployed for finance automation, computer use, scientific discovery and chip design, and reports that Prime Mega MoE reached 48% model FLOPs utilization on H200s using Qwen's 235B MoE architecture.

Generated from reports · updated 18 min ago

Report timeline

Follow the reports to see the event from different sides.

November 20
  1. Hacker News · AIPicks
    Prime Intellect launches Model Factory, a unified platform for training, deploying and improving models

    Prime Intellect introduced Model Factory, a single control plane that combines reinforcement learning, supervised fine-tuning, pre-training and inference so companies can train, deploy and continuously improve models on their own tasks and data. It offers on-demand hosted training billed per GPU-hour and dedicated deployments with reserved capacity plus direct Slurm and Kubernetes access, and it connects training to one-click inference endpoints, microVM sandboxes, environments and Prime Traces for turning production traces into the next training run. The company says Model Factory is already being deployed for finance automation, computer use, scientific discovery and chip design, and reports that Prime Mega MoE reached 48% model FLOPs utilization on H200s using Qwen's 235B MoE architecture.

Heat for this event

Not enough continuous observations yet to draw a trend.