Prime Intellect’s new kernel enables efficient training of MoE models by adding a fast backward pass, cutting GPU hours by up to 55% compared to standard frameworks.
AI-generated summary. Check the original reporting for context and accuracy.
This is the available feed excerpt, which may be shortened or in the publisher’s original language.
Prime Intellect has released Prime Mega MoE, a training extension of DeepSeek’s fused Mega MoE kernel. DeepSeek’s kernel only ran the forward pass. Prime Intellect added a fast, overlapping backward pass. That makes the fused kernel usable for training, not just inference. In its prime-rl trainer, full training steps on Qwen3 MoE models run up to 34% faster. Against TorchTitan, the gain reaches 55%. For teams training mixture-of-experts (MoE) models, that is a direct cut in GPU hours. TL;DR Size: A GPU kernel, not a model. Tested on Qwen3-30B-A3B and Qwen3-235B-A22B (128 experts, top-8 routing). Runs on: NVIDIA Hopper (blockwise FP8) and Blackwell (MXFP8) GPUs, BF16 on both. Published numbers use 8×H200 nodes, 32 to 128 GPUs. Performance: 32% to 34% faster full training steps than the prime-rl baseline. 48% to 55% faster than TorchTitan. Best: 3.11x faster forward pass on a Qwen3-30B-A3B MoE layer (12.0 ms to 3.9 ms). Worst: 1.24x backward pass on a Qwen3-235B-A22B MoE layer (30.5 ms to 24.6 ms). What is Prime Mega MoE? Prime Mega MoE is a fused, overlapping MoE kernel for training from Prime Intellect. It builds on the Mega MoE kernel in DeepSeek’s open-source DeepGEMM library. Why is MoE training communication-bound? In an MoE model, each token goes to a small subset of experts. Those experts sit on different GPUs. So every layer sends tokens out, runs expert math, and sends results back. This happens in the forward pass and again in the backward pass. In a standard setup, these are separate operations. The GPU sorts tokens and waits on the transport layer. Then it computes, hands results back, and waits again. Those waits are idle GPU time. How does the fused kernel work? DeepSeek’s Mega MoE, added to DeepGEMM in April 2026, replaces that chain with one kernel call. It fuses expert-parallel dispatch, the 2 expert linear layers, SwiGLU, and combine. While one token group is in transit, the kernel computes on a group that already arrived. It also skip
An automated sorting aid based on source trust, keywords, and entities, on a scale of 1–12. High: 9–12; medium: 6–8; low: 1–5. It does not measure urgency, verify reporting, or indicate editorial endorsement.
Samaya Sterling outperforms Claude Fable 5.1 on financial benchmarks at one-eleventh the cost, proving that co-training small models with their harnesses can dominate specific domains.
Audio8-TTS-Realistic offers a small, open-source voice cloning model that balances multilingual accuracy with low deployment costs, making it a practical tool for developers.
Baseten’s new joey-1 model offers ultra-cheap, high-speed classification for routing and safety checks, significantly reducing costs for high-volume AI applications.