Key Info
GMI Cloud announced it is a launch partner for EAGLE-4, available on its platform from day one. On its B300 fleet, EAGLE-4 delivers 2.7x faster decoding than plain decoding on Qwen3.8-27B.
Highlights
- EAGLE-4 is live on GMI Cloud today as a day-one launch partner
- 2.7x speedup over plain decoding measured on Qwen3.8-27B
- Runs on GMI Cloud's NVIDIA B300 fleet
- Upstream EAGLE-4 reports 961.6 tokens/s on Qwen3.8-27B with TokenSpeed (BF16, 4x B300, TP4, single request, T=0)
- Full benchmarks from GMI Cloud are promised soon
